There’s a common arc for human technology that goes something like this:

1. Humans invent a new technology.

2. This technology is good, but has flaw(s) that prevent mass adoption.

3. Humans create new infrastructure that unlocks the full power of the technology.

4. Everyone wins, especially the humans who made that new infrastructure.

This happened with penicillin, which was promising but couldn’t be used at scale until humans came up with a new mass-production process in the ‘40s. It happened with electricity, which was limited by the short range of the DC current until AC took over. It happened with planes, which were speed-constrained by propellers until people invented the jet engine.

It feels a bit like we might be at phase 2 of that arc with AI.

Depending on who you listen to, AI is one of the greatest pieces of technology humanity has ever created. It’s going to transform everything. It’s going to make everything better. That’s what they say. And if you’ve ever used any kind of modern LLM, you can certainly see the potential; AI can do a lot of powerful stuff and it can sometimes feel impressive.

But I wouldn’t blame you if you looked around and asked… Where is all of the progress? Where is all of this mass adoption? One example of this lack of progress is in finance. Where are the investment banking analysts, or the private equity associates, suddenly being 10X more productive? Why are they still working 80+ hours per week (the pain AI was meant to solve)?

What gives? You could come up with a lot of reasons, but I’d argue there is one rather obvious explanation for why AI is still stuck at the door in a lot of major industries: accuracy.

LLMs hallucinate (i.e. generate the wrong information) a lot. This could be okay if you’re working on a personal project with no stakes. But if you’re an investment banker researching a deal, for instance, one wrong number can lose you millions of dollars.

Accuracy is a big deal, and today’s AI can rarely promise it.

Kepler, the startup we’re covering today, is building new infrastructure that aims to take AI through phase 3 of the technology arc and beyond; infrastructure that will, if successful, be a necessary layer in every industry where accuracy matters.

Their approach is to stop asking LLMs to do what they’re bad at, and to instead build a new deterministic trust layer. And they’ve started in an industry that is extremely allergic to getting it wrong: finance.

The product

Say you are an equity research analyst and you need to compare gross margins between AAPL and MSFT over the last four quarters. You would rather not spend valuable time doing this yourself, so you ask your favorite LLM.

The LLM comes back with a confident answer. It tells you that it is completely sure that it calculated the right numbers. And sure, it might have. But it also could have…

• Generated a confident-looking number out of thin air.

• Found a filing from a year ago and pretended it was this year’s.

• Found the right filing but pulled the wrong line items from it.

• Found the right line items but did the math wrong.

• Found the right filings but cited the wrong sources.

• Mixed filings from different periods or companies within the same answer.

• [This list could be 100s of bullets long. But I’ll spare you.]

Bad news for you is that LLMs do some version of this—some version of getting it wrong—pretty often. With many finance questions, major models tend to get it wrong more often than they get it right. (For example, on the foundational task of pulling the right line item out of a 10-K, frontier models only get it right between 38% and 46% of the time.)

And so, if you are this equity research analyst, AI is mostly useless when you are dealing with anything that requires accuracy (which you often are). If you can’t trust the AI’s numbers then you need to double-check yourself, at which point you have saved no time at all.

The reason this hasn’t been solved yet is that “hallucinations are a structural property of how these systems work,” Vinoo Ganesh, co-founder and CEO, told me. There is no such thing as a hallucination-free LLM, Vinoo says, because LLMs are basically just big probability machines.

Most existing products in the space are just LLMs with finance reskins; the AI itself still generates numbers, which means it can hallucinate. Their approach is not to fix the LLM, but to instead build a separate layer underneath it:

• The LLM only figures out what you’re asking

• A separate, deterministic, non-AI system pulls the data and does the math

• A semantic layer that ties everything together, making sure that the LLM and retrieval layer speak the same vocabulary (which is often unique per company)

When Kepler’s AI gives you an answer, it’s backed by real data that was calculated separately from your query to the LLM. You can click any number and see the exact filing, page, and line item it came from, which lets you be confident without running the numbers yourself. Kepler’s output shows up where work really happens (e.g. Excel, PowerPoint, Word), which is a lot better than a chat answer an analyst needs to copy and reformat and validate by hand.

“In finance, the model can’t be the whole system,” John McRaven, Kepler’s co-founder and CTO, said. “We treat it as one stage in a pipeline whose job is to hand the model exactly what it needs to succeed exactly at that stage.”

Kepler also learns how you do your job. Every desk at every fund has its own quirks: how they treat convertible debt, how they handle diluted shares, which assumptions go into an EV calculation. Tell Kepler once, and it applies the rule every time after that. A generic LLM averages toward consensus, but in finance, consensus loses money; personalization is important.

Kepler started with finance because it was an industry where wrong answers are extremely expensive (meaning people are willing to pay a lot to avoid them). But the accuracy problem with AI, as you might imagine, extends far beyond finance.

And if Kepler is able to succeed, you should expect to see their architecture show up everywhere else accuracy, auditability, and verifiability matter. (The team told me they’re already getting inbound from people in other industries asking to implement Kepler’s architecture on their datasets.)

The strategy

It is not outlandish to say that trustworthiness is one of the biggest problems that still needs to be solved for LLMs to achieve their full potential (however you would define that). A huge amount of the hesitation around using AI is not knowing whether you can trust the things it is saying to you, and a huge amount of the ridicule directed at AI companies takes the shape of “look at this dumb AI saying dumb untrue things.”

In the future, “the default answer to the question ‘how do I use AI in trustworthy ways is Kepler,” Vinoo said. Kepler will be used everywhere it matters: “in finance, in law, in medicine, in defense, in any domain where being wrong has real consequences.” In other words, if you want to produce accurate and verifiable work using AI, you’ll need to use Kepler’s infrastructure.

Finance may have been the smartest place to start, and the future holds a lot more. But the skeptic in you might voice concerns. “Why won’t the big model labs just do this better?” you might ask. There are two answers.

The first is that the labs are not really trying to do this. Every dollar a foundation model lab spends building per-customer data infrastructure, vertical-specific ontologies, and the coding of how Desk A treats convertible debt differently from Desk B, Vinoo says, “is not a dollar spent on the core model.” Their economics push them further up the stack.

The second is that Kepler isn’t really competing with the labs. Other “AI for finance” companies build features on top of a foundation model, and those features are at risk the day a lab ships something similar. Kepler builds the layer underneath. This infrastructure (which is a lot deeper into the plumbing than foundation labs go) is complementary with what companies like OpenAI and Anthropic are building. It opens up potential use cases for their LLMs, which can now be used in higher-trust workflows that previously (and still today) defaulted to manual work.

The team and growth

Co-founders Vinoo Ganesh and John McRaven spent a combined 18 years at Palantir, where they spent time working in demanding, accuracy-focused environments. John ran Palantir’s time-series and analytics products with customers like BP and Airbus. Vinoo led the Spark group across Foundry and led a number of customer engagements across financial and defense verticals (including in Afghanistan).

After Palantir, Vinoo ran a geospatial company called Veraset to $15MM ARR and an acquisition. He then worked as the Head of Business Engineering at Citadel, building for the PMs that are basically Kepler’s target customers today. The two founders have, in other words, already built deterministic data infrastructure at scale in some of the most difficult environments to do so. Kepler is the inevitable extension of this background.

The team around them takes the same impressive shape. Just look at some of their founding team: Susannah Meyer, founding engineer, came from Meta and was Head of Engineering at Enigma Labs. Sara Kromwijk spent eleven years at Palantir leading projects like Palantir’s ontology layer. Chris Martin spent 15 years at Palantir working on optimization problems across the organization. Sean Gorsky directed ML at Capital One. Rui Zhang managed a 100M+ chemicals business before joining the investment team at a 10B industrials PE firm. Eddie Hammond ran business operations across Oddity through its IPO.

“They’re the kind of team who could get comfortable jobs at a frontier lab any day,” one person on the team told me. “But instead chose to build together.”

There’s less to say publicly about funding, partly by design, since Kepler hasn’t formally announced a round. But Vinoo confirmed they’ve raised $7M+ from operators who built the data and AI systems the rest of the industry now runs on (including the founders of OpenAI, Meta AI Research, MotherDuck, dbt Labs, Outerbounds, and Square/Block.)

Kepler keeps its growth numbers quiet by design. In finance, access to a tool like this is itself an edge, and many of their customers don’t want competitors knowing they use it, so staying low-profile is the point. ‘We have an unreal amount of customer inbound,’ Susannah Meyer said, ‘our entire focus right now is on shipping product to service that demand.’ Outside validation has started showing up regardless. In April, Anthropic published a piece about Kepler, the kind of recognition most startups don’t get this early.

The culture

Before writing a single line of code for Kepler, Vinoo and John spent sixteen hours in a small room with a goal to define the values they’d hire against. They walked out of the room with six: forward-deployed with product DNA, extreme ownership, production-first engineering, trust as the default, communicate with intent, and earn it every day.

Speaking of the bar, well, it’s high. Kepler is tackling one of the bigger problems in AI adoption right now, which means the team has to clear a high bar. “The bar here is set by people who spent eleven years at Palantir building ontology infrastructure, or who shipped production systems at Meta scale,” Vinoo told me. I think this is exciting for the right kind of person, and in fact most people at Kepler would tell you the talent density is part of why they joined.

The hiring process reflects this. There are no LeetCode interviews, no whiteboard puzzles, no rooms without internet access. “We give candidates a real problem we actually have to solve,” Vinoo told me of the hiring process for engineering, “and they use AI tools while solving it.” What he and John look for is how someone defines assumptions when the problem is ambiguous, whether they can ship fast, and whether they can push past the scope of what was asked.

Kepler is still small, so if you join you’d expect to spend meaningful time around Vinoo and John. So what are they like? Nearly everyone I spoke to referred to them as opposite sides of the coin. Vinoo is fast (“his brain runs at 100 miles per second,” someone said). John is the more deliberate counterweight. Eddie Hammond, who runs strategy and ops, put it this way: “Vinoo has the answer before you’ve finished the question, and he’s already pressure-testing it against the next three. John is the one who slows the room down and asks the underlying question nobody thought to bring up. Together they get it right the first time.”

The culture that arises as a result of all this is intense. The pace is fast and the bar is high. But, for the right person, it’s exciting to work on such a big problem alongside such great people.

“Most intense companies make you choose: be sharp or be liked,” Eddie said. “Kepler doesn’t. People are direct, the bar is high, and yet nobody keeps score on anyone else. That’s core to the culture.”

Should you join Kepler?

Kepler is small, but they already have an effective product being used by real firms in the field. If they accomplish even a fraction of their biggest ambitions, they could be worth a massive amount of money one day. And if you have read this essay so far and are convinced by the thesis, you might want to consider applying.

So what does a day at Kepler look like? Well, if you are an engineer, you might spend the morning three layers deep in retrieval infrastructure. By the afternoon, you might be sitting next to a portfolio manager, watching them do their job. You learn the way this fund defines enterprise value isn’t the textbook way; it’s some niche approach you’ve never seen before. So you encode the rule. Then you find out there are fifteen edge cases the analyst didn’t mention. The retrieval system underneath wasn’t built to handle them, so the engineer has to go rebuild that too. That might be a random Tuesday at Kepler.

I especially get the feeling that Kepler is good for people who want to work around people they can learn a lot from. Susannah Meyer (founding engineer) was leading engineering at a different early-stage startup before Kepler, and she left because she “wasn’t learning enough from the team,” she said. Kepler, by contrast, looked like the kind of place she could join and learn a lot alongside some great people. That bet is paying off.

And yes, Kepler is looking for people who are truly great at what they do (the pedigree of their existing team is good evidence), but more important than a fancy resume is what you’re able to do within a fast-paced, low-structure environment. “When we find people we get excited about, it’s because of a demonstrated path of ownership that shows they can thrive in a rapidly changing situation,” Vinoo said.

If you’re excited by the idea of going into Kepler’s New York City office, working alongside a bunch of talented people (and the office dog, Wagyu), taking ownership and shipping excellent work, then I’d encourage you to explore Kepler’s open roles.

Kepler specifically told me that they’re hiring engineers right now, and that “cold emailing is the single highest-signal way to get our attention.” They have just one ask of your email: skip the resume recap and skip the pitch. Instead, if you want a job at Kepler, send an email with one paragraph describing the most complex production system you’ve shipped end-to-end.

Send it to join@kepler.ai. They read every one.

Thanks to Kepler for supporting Next Play and making this essay possible.

What happens after you leave the doctor’s office?

Perhaps the doctor, alone for a brief moment, writes a few notes in your file before moving onto the next patient. Down the hall, somebody behind a computer types in some numbers and files an invoice to insurance on your behalf. The practice gets paid, the doctor keeps working, and the wheels of the system continue to turn smoothly.

That paragraph might be the understatement of the century. If only it were that simple for a provider to get paid for seeing a patient.

“Once an invoice goes out the door,” Alvin Wu, co-founder of Amperos, told me, “there is a rat’s nest of manual work standing between the provider and their money… It is one of the most broken back-office problems in America, and almost nobody outside of healthcare knows it exists.” This problem—getting paid as a healthcare provider—is what Amperos is solving.

Right now, more than 1 in 5 invoices that healthcare providers send to insurers go unpaid (to the tune of $260B+ per year). Those numbers have only continued to grow over the past decade. Healthcare providers do some of the most important work in the country, and yet it is becoming increasingly difficult for them to actually get paid.

If Amperos succeeds, then healthcare providers can stop chasing invoices, recover many billions more dollars, and redirect that money to what they do best: serving patients.

So what, exactly, is Amperos doing to make this a reality? How are they solving this huge, but difficult-to-solve, problem? What is their trajectory? And, of course: should you join?

The product

Years ago, before he co-founded Amperos, now-CEO Michal Miernowski worked with admin leaders at healthcare clinics. One of these clients was a clinic with 100 doctors on the payroll but 300 people working in collections to ensure those doctors got paid. “The imbalance struck me. Why did it take 3 people to chase payment for the work of 1 doctor?,” Michal said.

It sounds a bit insane, but that situation is more of the norm than an outlier. Collections is a huge problem, and the existing solutions have been the default for decades. Providers have had to choose to either build a team in-house (even though there aren’t enough people in the U.S. to do the job, Michal said) or to outsource to a traditional firm using offshore labor (which is sometimes slow and often full of quality issues).

There is also an AI arms race starting in healthcare, and insurers have been faster to adopt than providers have. Most healthcare insurers now use AI, many of which use it to decide whether to approve or deny claims. The old-school approaches will likely only become less effective as insurers use AI to deny claims faster, and sometimes at higher rates, than ever before.

So what is Amperos doing to fix this? The traditional startup approach would have been to build software, package it, then hand it off to a customer to use to achieve the outcome they want: in this case, collecting more money for their work. Amperos is, rather, an AI-native neofirm: a company that customers pay not for access to only software, but to get actual outcomes.

If you are a healthcare provider, “getting actual outcomes” means paying Amperos to run collections for you. They do it faster, cheaper, and far more effectively than any option you’ve had in the past. So you get a huge amount of money back in your (and your doctors’) pockets, and you get it faster.

Amperos has, of course, built a mountain of advanced internal software to give them an edge: like agentic AI voice systems that call insurance companies, computer-use agents that log into hundreds of portals to research claims or send appeals, and a denial-intelligence layer that ingests every payer interaction to identify causes upstream. With innovations like these (and others), the relatively small team at Amperos is able to do what used to take an army.

This neofirm approach is one of an increasingly popular group in AI. We recently wrote about Manifest (for lawyers) and Knit (for market research), both of which are taking somewhat similar approaches. Sequoia wrote about how this business model would be the next $1T company. OpenAI and Anthropic both launched neofirm-esque projects this year.

The strategy

All kinds of people and companies say they want to “fix healthcare.” But, as nice as that sounds, it’s also rather vague. What are we talking about? What does it mean? And, if you wanted to “fix healthcare,” what would be the best way of going about it?

While not as “sexy” as tackling longevity or building a personal health assistant, tackling the healthcare back office is one of the fastest ways to have a disproportionate impact on the average American’s experience of healthcare. After all, almost all Americans today still experience healthcare in the office of a physician, and that physician’s practice needs to get paid for their work to keep the lights on and continue serving those patients.

“Our goal is to redefine the financial infrastructure of American healthcare,” co-founder Alvin Wu said. “And we started in the back office because that is where the pain is most acute and most measurable… That let us build a product with clear feedback loops and clear ROI. But the back office was never the destination. It is the entry point.”

So the master plan is, roughly, this: first, put Amperos at the center of major healthcare providers in America by solving one of their biggest problems (collections). Then, become the payments layer for all American healthcare. “As we work more claims,” Alvin said, “we accumulate a precise, longitudinal record of how every major health insurance company behaves. Which codes they systematically downcode. Which denial reasons are pretextual. Which appeal arguments work and which do not. Nobody else has this.”

Today, that intelligence does not exist in one place at all; Amperos is the company in a position to collect and aggregate it.

Eventually, that data tells Amperos something nobody else knows: exactly when a given provider will get paid, and how much. “Providers stop waiting 60, 90, 180 days for reimbursement,” Alvin said. “We pay them in days and take a small margin based on the risk profile of their payors. The underwriting model is the data model.” Nobody, Alvin says, can build that model without doing all of the upfront work Amperos is doing now.

If Amperos wins, then in 20 years they are no mere collections company. Instead, they are the payment platform sitting between providers and payors. The layer through which American healthcare gets paid.

The growth

The pitch for Amperos sounds obviously good. But then again, a lot of startups have compelling pitches. What you may be interested in, as a way to underwrite the pitch, is some actual data. So I asked the team and did some research. TL;DR: things are going well for Amperos.

Nearly two years ago, Amperos went through the Neo startup accelerator in 2024 (an accelerator that has backed Cognition, Kalshi, Cursor, and others). About a year later, they raised a $4.2M seed round led by Susan Liu at Uncork. And less than a year later, in April of this year, Bessemer Venture Partners led Amperos’s $16M Series A; you can read the announcement from Amperos here.

The traction with customers is equally (if not more) impressive. Within just the first year of signing their first annual contract, Amperos has scaled to serve over 3,000 clinical locations across all 50 states. Amperos has signed some of the largest specialty-focused and fastest growing clinical networks, including DOCS Dermatology, Boulder Care, and others.

“We sell to larger, enterprise providers,” Alvin said, “which means we can build a huge business without 100s of customers and we really get to focus on the dozens that we serve.” The first of their customers came through cold outreach, whereas now it’s about an even split between referrals, outreach and inbound. (Referrals are a big deal when one new customer can mean many millions more dollars.)

By the time they raised their Series A this year, Amperos was recovering about $700M a year for their customers. Customers recover 22% more revenue per claim compared with traditional vendors, and at half the cost. Amperos reduces aged and unpaid claims by 60%+, resolves denials 5x faster, and (in one of my favorite anecdotes), one customer told Amperos that “we weren’t even touching 27% of our claims” before they signed on with Amperos.

Perhaps most surprising is the headcount: Amperos has been able to do all of this with about 20 people. A traditional firm serving as many clinics as Amperos does would probably need hundreds, if not thousands, of people on payroll to do this. There is a long way to go, but a 20-person team recovering $700M a year is a pretty good start.

The team and culture

Amperos does a lot with a rather small team. The velocity at Amperos is comparable to “top-tier startups,” Alvin told me. The team works hard. “We want to win together.”

But winning together does not come at the cost of life outside of work, because “[that’s where] you build the creativity, inspiration, and taste to do your best work when you are in the office.” Instead of focusing on sheer time spent in or outside the office, the Amperos team focuses on what makes a team effective and able to move quickly: “high ownership, quick decision making and building with low ego.”

As for the work itself, “there’s opportunity to take on more ownership [than at other companies],” Adam Gluck, a software engineer at Amperos, said. “I feel like my ideas are more likely to be implemented.”

“It’s very easy to try new things at Amperos,” Annie Qiu, Head of Engineering, said. “I can use new tools without waiting for a lengthy review process. We move very fast here, and looking back over the past year, I’m surprised by how much we accomplished.”

I also get the sense that the team likes to have fun. They’re recently back from an offsite in Taipei, which I’ve been told was an excellent time. Back in New York, there’s a piano in the office (and an air fryer, a detail that numerous people enthusiastically mentioned). There are fun activities. People make lots of jokes. “Between escape rooms, joke of the day, and the company air fryer,” Adam said, “coming to work is always fun.”

It’s easy to roll your eyes at this stuff. But all startups say they have a good culture! This is downstream of something more important, though, which is that the Amperos team seems to actually like each other—starting with leadership. “All 3 of the founders are so humble, approachable, and low ego,” Jackie Osborn, who joined from Stripe, said. “They care about our users, have very high empathy, and just want to do what’s right.”

“Density of talent on a team compounds in the same way data does,” Alvin told me. “One exceptional engineer who cares about their craft, operates with full ownership, and makes the people around them better is worth more than ten people who need to be managed.”

Perhaps as a result of this culture and intentional approach to hiring, nobody at Amperos has ever quit. “We are intentional about building two products at Amperos,” Alvin said. “Our core business and the experience of working at Amperos.”

Should you join Amperos?

There are a few categories of people I would push to seriously consider joining Amperos.

The first is people who care about ownership. Amperos hires for it, and it’s a big focus of how they get work done. People who work well at Amperos “find the problem, dig until they understand the root cause, fix it, and tell people it is fixed,” Alvin said. “That combination of ownership and follow-through is rarer than it sounds.” If you are a particularly autonomous and proactive person who can take a lot of ownership over your work, you’ll fit in.

The second is people (especially engineers!) who are open to—or eager to—learn an industry as opposed to operating in a silo. “If someone is looking to focus purely on deep technical work, this may not be the best fit,” Annie said. “We expect engineers to go beyond code and invest in learning the healthcare domain, because that’s critical to building the right solutions.”

The third category is people who move fast and don’t need much process to do it. “Our build velocity is something that might push [people] away,” Prakhar Gupta, a software engineer, said. Amperos is not a fit for people who “prefer to be as thoughtful as they possibly can be before tackling a task.” There’s a bias towards action at Amperos.

For everyone else, the case is fairly straightforward. Annie, who joined from Stripe after a year of building MVPs for other startups, put it well: “Amperos stood out because of how tight-knit their founders are and how each of them excels in their role. They already had a strong sales pipeline, their product narrative was very specific, and their technical product was built on years of voice agent research. It was clear they were well positioned for exponential growth.”

If she’s right, now might be a good time to go check out open roles at Amperos. They’re currently hiring across most departments for their office in New York City and a new hub they are building out in San Francisco.

Thanks to Amperos for supporting Next Play and making this essay possible.

The Romans believed that every person was born with a genius. A second, divine self that walked with you from the hour of birth until the day you died. On your birthday, instead of celebrating yourself, you poured out wine for your genius.

It is harder to find a culture without this archetype of the second self than one with it. Norway had the fylgja, “the one who follows.” Some parts of Christian culture feature the guardian angel, described by Aquinas as “appointed for men who have to pass by an unsafe road.” In early modern England, practitioners of witchcraft spoke of familiars, animal companions who offered help during moments of distress.

Perhaps humanity craves the second self because it sounds like a very good thing. It’s not a friend, nor a lover, but rather a non-human entity that serves as an extension of yourself. A supernatural-ish being who helps you carry the weight of being human.

What would happen if there was a second self around all of the time to help? Would you be more effective? Would you be more caring? Would you be more motivated?

While this all sounds rather spiritual and aloof, I’d like to pitch you the idea that, thanks to artificial intelligence, we might now be closer than ever to some version of the second self. On paper, AI it has many of the attributes you might want if you were to build a second self:

• It is basically alien

• It is available all of the time and can hold massive amounts of context

• It (usually) has exceptional language abilities

• It can (sometimes) be useful, smart, funny, kind, and knowledgeable

• It is scalable; everyone on earth could have an AI companion

Yet when you look at what is being built… This helpful, ethical second self does not show up often. Many of the products that exist on the market today are addictive and anti-social, designed to extract as much as possible from the user (think: romantic AI relationship apps and AI roleplaying apps). These products tend to leave their users worse off in the real world. If I had to guess, I doubt that most people reading this would be very excited to build a product like that.

But maybe personal AI doesn’t have to be bad and evil and parasocial. Maybe there is a way to build a great version of the second self humanity has craved for thousands of years, and maybe we can make it actually good for its users in the real world, not just online.

Maybe that answer is Tolan, the startup we are covering today.

The team, led by Quinten Farmer (CEO), is answering one big question: “How will humanity build healthy relationships with AI?” Tolan is their bet. If they succeed, then in 20 years Tolan will have given virtually every person on earth ‘a second self,’ Quinten told me.

There’s a lot of traction so far. They’ve raised $30M and have $12M+ in ARR. Hundreds of thousands of people pay to use Tolan, most of them 25+ year old women (especially moms). The team is talented and has a successful exit under their belt. And, perhaps just as importantly, there are good signs that Tolan is aiming to be a net positive in the world; a good thing not only for the company, but for the people actually using the product.

So… What are they building? Why is it different? And should you join?

The product

Unlike many personal AI products out there, Tolan makes no attempt to be human: it is a cute and friendly alien from another galaxy. How do you interact with it? Well, it’s rather simple. You just talk, or type, to it. And Tolan will talk, or type, back.

You can talk to your Tolan about whatever you want, its product trailer says, and your Tolan will get to know you so that it can help you with just about anything. Talk through something overwhelming in your life. Chat about music. Put together an outfit. Get motivated to study for your college finals. Vent about something stressful. Your Tolan is there for it all.

(There’s other fun stuff, too, that makes this feel like a lot more than a chatbot: you can customize your Tolan’s looks and voice, you can design its planet, you can invite friends and their Tolans to share a cozy cottage together. Faintly reminds me of Animal Crossing.)

Then there are the things your Tolan won’t do.

It won’t do roleplay or pretend to get into a romantic relationship with you (there are no upsells for high-res Tolan feet pics). It won’t pretend that it’s a human. Tolan knows it lives in a galaxy far away, so it won’t make promises to come visit you; a rule which writer Eliot Peper, a sci-fi novelist who helped build Tolan’s lore and character, said is core to the Tolan ‘master-system prompt’. And, of course, it won’t help or encourage you to do dangerous things (here’s a Reddit user being denied when asking for instructions about how to rob a bank). Oh, and if you spend more than an hour talking with Tolan, it will nudge you to sign off and go outside.

The idea that Tolan is definitely not a human, nor pretending to be one, is rather important to the way the product works. “An alien species felt right,” the company’s president, Ajay Mehta, wrote. The alien story explains why it can be a bit awkward; why AI does not feel totally human.

The alien presentation also helps emphasize that Tolan is not meant to be a replacement for human interaction. In fact, if all goes well with Tolan, real life should improve. That is basically the whole goal. And thankfully, for most Tolan users, it does. More than 85% of Tolan users say their Tolan has helped them get through an emotionally difficult experience, and more than 72% said their Tolan helped them manage or improve a relationship in their life. That’s a good start. But Tolan’s ambitions are even bigger.

The strategy

If they wanted to, the Tolan team could probably (at least for a while) run Tolan as a modest personal AI app making a few million dollars per year. They’ve already reached that milestone. But the end goal for Tolan, Quinten told me, is to “give every person on Earth access to their own personal AGI.” The goal is to build the second self for all of humanity.

That AGI will exist is the assumption Tolan is built on. The team is focused on solving the bottlenecks they believe need to be solved for personal AGI to become widespread:

1. Trust. It doesn’t matter how capable your model is if a user does not trust it to act in their life.

2. Memory. Intimate and esoteric context is what makes personal AGI feel actually personal. (You cannot really get this, however, without the user trusting you first.)

3. Connection. People need to be willing to incorporate their personal AGI into their most important human relationships.

“Everything Tolan does today is designed to make progress on these three key bottlenecks,” Quinten said, which need to be unlocked before it’s possible to realize the full promise of personal AGI. So, you may ask: how exactly do they plan to solve these bottlenecks?

One of the main ways Tolan is building trust today is the character itself. Few examples of personal AI have a character like Tolan’s (most character-based personal AI are romantic or role-playing adjacent, which Tolan is not). It is a lot easier to build trust in an alien-ified character than it is in a scary AI company logo.

As for memory, Tolan’s choice to be a voice-first product goes a long way. People have more open-ended conversations with Tolan than they do with text-first chat products, which helps the company build a more complete memory profile of each user.

Linking real human connection and AI is a rare thing, but Tolan has taken a few steps towards it so far: like a feature called the cottage, where your Tolan can be friends with the Tolans of your family and closest friends.

None of what Tolan has built so far is going to completely solve these bottlenecks or bring about personal AGI in an instant—there is still more work to do—but things are going well so far. Tolan users chat more than 200 minutes each week with their Tolans, and they increasingly welcome Tolan into their most important personal connections (35% of retained users have at least one friend connection).

Over the next six months, Tolans will offer to take actions on behalf of their users, leaning on the trust we have built to more helpful and ultimately more deeply integrated into their lives.

Tolan is not the first company to build personal AI, but Quinten doesn’t think that is relevant. “Historically, in major technological cycles, the first product at scale was not the final winner. It was instead the one that packaged extraordinary new technical capabilities in a format that is safe and legible for everyday people. The Macintosh defined the format that won the PC era. Facebook did it for the social web, and the iPhone did it for mobile.”

Will Tolan, perhaps, do it for personal AI?

It’s too early to say, of course, but early traction is encouraging.

The growth

This is not the Tolan team’s first rodeo. Much of the core group, including Quinten Farmer (CEO), Evan Goldschmidt (CTO), and Aseem Kishore (founding engineer) worked together on Quinten’s previous startup, Even: a fintech that Walmart bought for $300M in 2022. This is an experienced group of people that seems to know how to make things that work.

It may not be surprising, then, to learn that Tolan’s public launch in February 2025 went about as well as you could expect. Within four months they had over 3 million downloads, more than 100,000 paying subscribers, and $12M in ARR. A lot of this growth was driven by viral organic (and affiliated) social media marketing; here’s an example of a viral TikTok.

That summer, Tolan raised a $20M Series A led by Khosla Ventures, bringing them to ~$30M raised total. Other backers include Lachy Groom, Nat Friedman, Daniel Gross, and Instagram co-founder Mike Krieger.

You could probably argue that the Tolan character itself has a lot to do with the growth so far. The team spent a lot of time trying to figure out what Tolan should look and sound like (you can read more about that here), and it feels like they landed on something that works. In a world of awkward personified chatbots and impersonal pulsing circles (e.g. ChatGPT’s chat interface), it is not very hard to see why people would rather talk to a cute little alien.

The team and culture

I mentioned earlier that much of the core team has already worked together on a different startup, Even. I should also mention that the team is small (about 23 people as of today) and works mostly in-person in San Francisco.

But it is reasonable to ask how you will fit in here. The culture here is perhaps more established (and time-tested) than it is at most Series A startups. Many of these people have spent years building a successful startup together. The odds that you are going to come in and upend the way they work, while not zero, are probably low.

So what is it like to work at Tolan, then? The answers I got were pretty satisfying.

The people at Tolan are all “low-ego, kind/personable, hard-working, care deeply about their craft, and just want to do really great work,” founding engineer Aseem Kishore said. “The culture is as no-BS and drama-free as it gets. The focus is just entirely on doing great work, shipping a great product, and building a great business… [Tolan is] very simple and ceremony-free.”

People are direct with each other and it’s not uncommon to publicly debate, Carter Aronson (who leads applied AI at Tolan) said. “I think to an outsider, it can sometimes feel surprising how direct we are… But there’s a tremendous amount of trust and respect across the team, and I think this allows us to skip a lot of the fluff and politics that show up in other orgs.”

The vibe, according to Lily Doyle, a founding researcher at Tolan, is “the calmest culture I’ve ever seen… [And] it’s just a really fun and smart team. Every person we add and keep has brought a great, fun energy that keeps coming to work a really good time. As one of my colleagues said, we all joke as hard as we work. It’s the best.”

Calm doesn’t mean easy, of course; the team works very hard, including “many nights and weekends,” Aseem told me, “[but] we know this is a marathon, not a sprint.”

Perhaps counterintuitively, Tolan arrived at this apparently successful culture without thinking too hard about it at all. “I’m not sure we even have a mission statement, nor any formal company values,” Aseem said. “We’ve never had any big culture meetings, value workshops, big retrospectives or pulse surveys. That stuff just doesn’t seem to matter here.”

The thing that does seem to matter, or matter most, is simple: just do really great work.

Should you join Tolan?

People who fit in best at Tolan are folks who are able to “share our expectations about pace and quality,” Quinten said. “Most companies only care about one or the other, or neither. We believe the two move in concert.” You should expect to do great work, fast. (The writer James Somers wrote a great essay on the topic with a thesis similar to Tolan’s.)

But what about the broader picture? How should you know if Tolan is the best place for you to come do your life’s work? Quinten says it’s a fit if:

• You care deeply about the role AI will play in the lives of everyday people.

• You love building products that connect emotionally with consumers.

• You want to work on a small, veteran, tightly knit team.

There is a general sense of maturity that I get from Tolan. Maturity in the product, which is a labor of love designed with a lot of intentionality and guardrails. Maturity in the strategy, which takes a long-term view and steers away from making short-term compromises that increase revenue but hurt real people. And maturity in the culture, which steers clear of politics and drama with a simpler focus on being honest and doing great work.

If this sounds like the place you think you could make a serious impact, I’d encourage you to take a look at Tolan’s open roles. They are currently hiring across most departments for their (still small!) team. All roles are in-person in San Francisco. If you want to go beyond the traditional application, Quinten said, “just email me and state as concisely as possible why you are uniquely suited to join our team.”

Tolan is still a small company, but they are growing fast. If they really are building the second self that humanity has been imagining for eons, then now is the time to be in the room.

Thanks to Tolan for supporting Next Play and making this essay possible.

This afternoon, somewhere in the United States, a hungry child won’t get fed. Somebody in the world—many somebodys, in fact—will get diagnosed with breast cancer. An entire community in Chad will go another day without clean drinking water.

These are horrible situations, but they do not go unnoticed. No Kid Hungry, for instance, works to make sure no child in America goes without food. Susan G. Komen is on a mission to end breast cancer. charity: water is bringing clean and safe water to every person on the planet. The world has more problems than it can solve, more problems than our governments and markets can take care of on their own. But through human compassion and charitable giving, nonprofits aim to tackle many of the world’s most difficult problems.

So here is perhaps the least controversial statement ever written: the world would be a better place if nonprofits (1) raised a lot more money and (2) were a lot more effective.

Why? Fewer kids would go hungry. More support would go towards breast cancer research, community health, and public policy. More people would have easy access to clean drinking water. Thousands of other great outcomes to improve human well-being would happen all over the planet.

Here is the more difficult question: how do we make this world a reality?

If you were tasked with accomplishing that goal (nonprofits raising more money and being more effective), how would you go about it? What would be your first step? Well, if you are anything like Chariot founders Aaron Kahane, Salo Serfati, and Drew Schneider, you might spend years investigating this question. And when you do, you might quickly realize there is one major unsolved problem sitting at the center of all of this: financial infrastructure.

The financial rails available to nonprofits (handling things like donations, disbursements, gift processing, and more) are traditionally a mess. Clean up this mess and you not only help nonprofits raise more money and become more effective, you put yourself at the center of the entire industry to help determine what’s next. This is, more or less, what Chariot is doing.

So I wanted to learn more about Chariot. What does their product do today? Why? What’s the long-term vision? What is it like to work there? And who should consider joining their (intentionally) small and efficient team? Answers to those questions, and more, below.

The product

Charitable giving is about 3% of US GDP. That’s a lot of money. So you might be surprised to learn that the financial infrastructure that underlies it feels ripped straight out of the early 2000s.

Chariot focuses on Donor-Advised Funds (DAFs). Created in the 1930s and popularized over the past five years or so, the basic idea is that you contribute money or stock, get the tax deduction right away, and then decide later which nonprofits actually receive the money. The funds grow tax-free in the meantime.

DAFs are already a big deal: $326B was committed to giving in these funds as of 2024. But it’s still early days. Only 17% of Americans even know what a DAF is, and 401(k) accounts outnumber DAFs by a factor of 20, even though more Americans give to charity than save for retirement. The ceiling for DAFs feels much higher than where things stand today.

Okay, say you want to actually make a payment via your DAF. How would you do that?

Before Chariot, the answer was: with a lot of friction. You see a nonprofit you want to support, but there’s no option to use your DAF in the payment flow. Instead, you leave the nonprofit’s site, log into your DAF portal, search for the nonprofit’s legal name and EIN, enter the amount, specify the campaign (if you still remember it), and submit a request. Then someone writes a paper check and lumps your gift with hundreds of others. The nonprofit eventually receives the money with or information about you besides maybe your DAF’s name, if they’re lucky.

This is kind of insane, right? Business does not still happen this way in almost any other part of the economy. And this friction can probably explain a lot of why so many people end up ignoring their DAFs; it likely costs nonprofits billions every year. “Philanthropic infrastructure is inherently extremely high impact… But they’re stuck dealing with terrible financial technology and backwards processes,” Elon Packin, Chariot’s Head of Partnerships, told me.

To fix this, Chariot created an Apple Pay-like solution called DAFpay. It’s the industry’s first and only DAF payment checkout option that lets you donate from your DAF, directly on the nonprofit’s donation page. It’s already helping thousands of nonprofits (like No Kid Hungry, Susan G. Komen, and Charity: water) raise more money and get more details about their donor so they can steward them better. (DAFpay was one of TIME’s Best Inventions of 2025, right alongside Claude Sonnet and the Waymo Driver.)

But the story doesn’t end here. Beyond individual DAF donations, nonprofits can receive DAF gifts from over a thousand different grantmakers and workplace giving platforms (like employer corporate match programs). And each grantmaker or platform has their own login portal and way of transferring funds and data to the nonprofit, which creates an operational nightmare for already under-resourced nonprofit teams. (Some nonprofits have 20 to 30 people on their team whose job is to reconcile, code, and move these gifts into their CRM and accounting systems.)

To solve this, Chariot introduced its two other core products:

• Disbursements: Helps grantmakers, e.g. DAF sponsors, including community foundations and workplace giving platforms, facilitate electronic payments to over 1.4M+ nonprofits safely and securely.

• Gift Processing: Helps nonprofits unify payment sources, enrich their donation data, and sync with their CRM and accounting platforms.

Each of these products attacks a different broken piece of the same underlying rails: how donors pay, how grantmakers pay, and how nonprofits make sense of what arrives. And each of them sets the foundation for Chariot to live at the center of all nonprofit finance.

I want to stress that this suite of products actually means something, actually does something, in the real world. “Financial infrastructure” can sound abstract, but the outcomes here aren’t.

You could look at the Michael J. Fox Foundation, which funds research toward a cure for Parkinson’s. After installing DAFpay, average digital gifts came in 9x larger than typical credit card donations ($1,800 vs. $204) and 69% of DAFpay users were first-time donors entirely. You could look at the Pan-Mass Challenge, a bike-a-thon that has raised more than $1B for cancer research since 1980. Nearly half of their 2024 fundraising growth came through DAFpay alone.

Chariot is unlocking better fundraising for many of the world’s most impactful nonprofits. And they are, by doing so, making the world a materially better place.

The strategy

Back to the big mission: how do we create a world where nonprofits raise more money and are more effective? How do we create a world where philanthropy is 10% of the global GDP—a goal Chariot states in their employee handbook?

The first step is to build a new set of financial rails that make it faster and more secure to send and receive nonprofit payments.

The second step, of course, is to win. In twenty years, Chariot will be “the default financial infrastructure layer for the entire nonprofit economy,” Aaron Kahane, co-founder and COO, said. “[AI] agents will handle most operational and software work. In that world, Chariot powers the primitives and trust layer - identity, compliance, and money movement - that agents, platforms, and institutions rely on to move money to nonprofits.”

How does this happen? Well, one reason is that Chariot has also started by solving the hardest-to-replicate problems first. “Our stack gets more defensible over time, not less,” Aaron said. “Most startups in this space focus on UX, dashboards, integrations… those layers get commoditized. We’re focused on what doesn’t: nonprofit identity, compliance, and money movement. Establishing and maintaining financial accounts tied to nonprofit identity requires deep compliance (KYC, AML) and institutional trust. That’s slow, operationally heavy, and not something a new entrant, or an agent, can spin up quickly. We already have 38% of the total nonprofit space (by contributions) banked in this network.”

There is a strong network effect here, too. “People all like to talk about network effects, I’ve put it into a ton of pitch decks in the past,” Mitch Stein, Head of Strategy, said. “The power of our network effects are palpable… Because we’re bridging pieces of the ecosystem, we have grantmakers paying a nonprofit through Chariot, which brings the nonprofit to us. We have nonprofits directing their payments to their Chariot account where they can automate the processing & reconciliation. We have donation forms adding Chariot payment options at the checkout, bringing their nonprofit clients in… I’ve really never felt it like this before.”

And there’s another more obvious reason, the same reason you probably haven’t heard about many Chariot-like companies: this is a very underserved space. Y Combinator’s next batch isn’t going to feature 30 new nonprofit finance startups. And since Chariot started with the difficult-to-replicate infrastructure problems and is already entrenched with a huge portion of the market… You’d have a long road ahead if you really wanted to compete.

The growth

The strategy is working. “We are growing 2-5% of the TOTAL NONPROFIT SPACE each month,” co-founder Aaron said of Chariot’s banked network. Chariot has moved over $450M in total to nonprofits, and they moved $150M this past April alone. (To put that into context, Chariot’s first $10M month was only 8 months ago.)

Early sales were founder-led (“warm intros, quick pitches,” Salomon Serfati said). But, in part due to the network effects I described earlier, things are shifting to product-led growth. “Our customers are finding the next set of customers.”

If you look at Chariot’s fundraising history, you’ll see that it is much like their product: purpose-built and intentional. Chariot went through Y Combinator in the summer of 2022 and raised an $11M Series A led by Maveron in 2024. They count a huge portion of the nonprofit industry as customers already. They are growing fast. Revenue is up 11X since that Series A. Today, the company is growing its customer base about 30X YoY.

Chariot has been intentional about fundraising, has built a phenomenal product, and is growing faster than ever in a market they hope to one day be the default financial layer for.

The team and culture

Chariot speaks more about its culture than most startups. Co-founder Aaron runs a Substack where he has written about how to get stuff done, why slow speaking is overrated, and why good references are better than a resume. Their employee handbook, which is public (you can read it here) and covers the way they work, gives more detail than a million years’ worth of generic career pages possibly could. I read all of those things—and talked to their team directly—to learn more about how they operate.

One of the first things that stood out to me is how much Chariot cares about speed. Of course every startup says they like to ship fast. But Chariot has clearly thought about this and has a few thoughts about what speed means and how to do it:

• Speed matters a lot. There’s an expectation that you reply to everything in <1 day.

• Speed and quality are not mutually exclusive. You can, and should, do both!

• Speed doesn’t mean being lazy. It also doesn’t mean “AI slop,” their handbook says.

This kind of quality-focused urgency was echoed by the people I talked to. “Chariot is a culture where you really need to own your workstreams and drive them forward. You need to be passionate about the mission in order to have the hunger to push the pace on hitting your goals,” Elon Packin said. “When I first transitioned to startups from investment banking, I had hoped to find the energy I could feel when reading about the early innings of Nike in Shoe Dog. Chariot has given me that and so much more.”

The hierarchy at Chariot is flat, “like a sports team,” and there’s not much (or any) micromanaging. “Compared to my past jobs, I feel like I have the most autonomy at Chariot,” said Julia Rosenson, a software engineer. “I feel trusted to prioritize work, advocate for what I think are the most important initiatives, and question any requirements presented to me.”

The work is hard, as you might expect. It is also, encouragingly, laser-focused on actual output.

“We care about output and the impact of the output far more than the hours,” Salo said. Chariot is not a startup that spends long hours in the office so they can post about it on LinkedIn; you won’t see them romanticizing the 996 model. Rather, the expectation is that you care about your work and that you do it well. To be as effective as possible, Chariot works in-person in NYC and expects that people work 50 to 60 hours per week. “That turns off 95% of people and that’s fine,” Chariot’s handbook says. “You want to find that 5%.”

There is also plenty of fun stuff. The office has a calendar called the “crashout board” for when people crash out (about a bug, a customer, etc). You get your picture taken and put onto the calendar for the date of your crashout. There’s a gong they hit when closing new sales. There are fantasy sports leagues, games of Mario Kart, annual hackathons, and nonprofit volunteering. There is a LOTR-style “sidequest map” to remind people to stay on-topic. The kinds of quirky fun things people associate—or used to associate—with tech startups definitely exist here.

Should you join Chariot?

Unlike a concerning number of companies in the world, Chariot is public and honest with candidates about what the expectations are: in-person in NYC with workdays from 9AM to 7 or 8PM, some sprints that extend late into the evenings, and most importantly, a real dedication to your work. This is a place where you are expected to actually care (a lot).

In many ways, I get the sense that people who are a fit at Chariot are people who care deeply about the world the company hopes to unlock. The world Chariot is unlocking already: a world where nonprofits have the tools to be far more effective, and where that effective-ness drives real outcomes for real people.

For the right person, Chariot’s mission might have a stronger pull than the average startup. And if you are that person, if you are particularly excited about Chariot’s vision for nonprofit fintech, then deciding to join Chariot might feel like the most obvious thing in the world.

As for what might make you more likely to get hired? Well, Chariot cares more about what you’ve done than if there is a fancy name on your resume. When you apply, “talk about some of the top things you built and why you built them,” Salo said. “We love people who bring their ideas to reality and have novel ideas for how to solve tricky problems.”

Chariot is currently hiring across most departments in NYC. See their open roles here.

Thanks to Chariot for supporting Next Play and making this essay possible.

Running a successful company is all about making good decisions. What to do with your product, who to hire, what messaging to use, what direction to take the business, and everything else: it’s hard to win if you can’t make good decisions about things like these.

But if you want to make good decisions, you need good information, quickly. You need answers! What do our customers think about this product of ours? How much would they pay? Does our messaging land? What do people want?

Unfortunately, there is no magic crystal ball that can immediately (and accurately) answer all of these questions for you. Instead, you have to do “market research.” In the past, that meant partnering with a probably-slow and probably-expensive research firm that would take months to get you answers. Sometimes it was useful; other times, the question had moved on without you.

But what was always true was that market research was painfully slow and inefficient.

The arrival of AI has changed the field drastically. Legacy firms have scrambled to retrofit AI onto their old workflows. There is also a fresh crop of AI-native market research “platforms” that are taking new approaches to make research faster and more effective.

But there is one startup—Knit—which is not taking the “software platform” approach. Knit, rather, has built the first AI-native market research firm. Just seven months ago, we wrote about their story and growth. We wrote about them because they’re a particularly differentiated startup that’s signed a number of impressive customers (including Amazon, T-Mobile, Paramount, Mars Wrigley and 50+ other Enterprises) and is growing fast. Today, we’re writing about them again, this time to help you answer one specific question: Should you join?

The product

Plenty of AI startups that exist today are building their products in what you could describe as the traditional SaaS model: they make an AI-enabled software tool, which customers get access to by paying money. This can work. It does, for lots of AI startups.

But in certain industries, like market research (and law) there is a different kind of AI-native startup emerging. A kind of company that Knit COO Chris Hicks described as an “AI-enabled neofirm.” These are services-focused companies that are AI-native from day 1, delivering better, faster end-to-end outcomes for their customers while still preserving human judgment.

This services-based approach is an effective way to describe Knit, which bills itself as “The AI-native research firm.” And I’m telling you about this now because it informs how you should think of the product. Don’t envision it as a DIY platform; rather, envision it as paying for an outcome. In Knit’s case, think of it as paying for an answer (or a series of answers) getting your full job done.

The way Knit works is simple:

1. Companies bring a strategic business question.

2. Knit’s Forward-Deployed Researcher (FDR) works alongside purpose-built AI agents to run the research end-to-end.

3. Within days, stakeholders have their answers.

The answers, by the way, are unified (not across separate platforms) and presented in exactly the formats, and with exactly the analysis, that leadership needs to make decisions about them. In other words, the answers you get with Knit are useful. The crystal ball works. You ask a question, you get the correctly-formatted answer, and now you can go decide what to do next.

Knit says their approach is faster, more effective, and unlocks entirely new ways of research compared to what traditional research firms offer, which is perhaps why all kinds of companies—including Amazon, T-Mobile, and Paramount—have partnered with Knit to do their market research to make better business decisions, faster.

One thing that’s worth calling out about Knit’s approach is that, both today and when we wrote about them in the fall, Knit has had a strong opinion about keeping human judgment in the loop. There are plenty of AI startups which have the goal of replacing humans, but Knit believes that keeping human researchers involved (e.g. in their Forward-Deployed Researcher roles) is part of what it takes to produce truly great work in this space. This is an interesting (and unique) thesis, and I suspect that it is at least part of why they believe they will win.

The strategy

Perhaps one of the most important long-term bets Knit is making is that providing services, not just software, is the right path for the company. “We believe that some of the most important and valuable companies in an AI-native world are going to be AI-native services,” Aneesh Dhawan, co-founder and CEO of Knit, told me. “We are rethinking entirely what doing the ‘whole job’ as an AI-native research firm will look like - and what it can do for customers.”

This market isn’t niche. Services (not software) are where the majority of budgets actually live, and a number of AI-native companies are positioning themselves to capture a lot of that spend. Just last week, OpenAI launched its Deployment Company: a $4B venture to embed forward-deployed engineers inside of enterprise companies. One week earlier, Anthropic announced a similar venture. And investors are catching on, too; Sequoia recently wrote that “the next $1T company will be a software company masquerading as a services firm.”

If you are the kind of person who hears about a new piece of AI software every day and is possibly getting a little jaded about the whole thing, Knit might be a breath of fresh air. Because one thing you can certainly say about them is that their approach is different. It is, in some sense, more ‘real’. In our spotlight on Thursday, the founder of Manifest told us that his company is “not selling pickaxes. We’re mining.” I think there’s a similar pitch here.

There is good reason to believe that this strategy has merit. In some industries, and market research may well be one of them, it is not good enough to simply hand over a piece of software. The reason Palantir had (and has) FDEs, the reason Northslope has them, was that it was the most effective way for their clients to get amazing results with their software. The same may be true about Knit’s FDRs—it is quite possible that a software tool is not enough.

Of course, there are other strategic bets at play. One is that combining quant and qual research together on the same platform (i.e. a comprehensive platform instead of a siloed solution) differentiates Knit. “Most companies in this space are picking off one slice of market research,” Jake Shanesy, Director of Product, said. “Knit is going after the entire arc.”

Another point in Knit’s favor is that they are AI-native. Legacy market research companies, the ones that have been operating for decades, can’t really say this. Many of them are trying to retrofit AI onto their current workflows (to varying degrees of success), but Knit is a much newer and faster company that has been building with AI from the start.

There’s also a structural competitor that few other competitors really have: actual researchers help drive the product. Chris Doty, Researcher on Staff, is one of them. His job is to translate years of research knowledge into product vision and direction. “We are building a platform for our end users,” Chris said. “Not just one that our engineering team thinks will work for them.”

I was curious, because it is the kind of thing I would ask if I wanted to join: what does success look like for Knit? In the short term and the long term?

“Short-term success looks like becoming the most important research company ever built,” Aneesh said; a rather impressive ‘short-term’ goal. “Knit is capturing the $130B being spent today on research services by delivering a 10x better experience than legacy companies… In fact, at some of our customers, up to 50% of the research being run is for use cases that never existed. This is net new research never done before.” Aneesh is referring to the Jevons paradox here: as the efficiency of research increases, demand does, too.

But research, Aneesh says, is just the “wedge to becoming a digital customer brain. Think of each data point on Knit as a neuron. The more research you run and context you feed in, the more synapses form between them, surfacing connections no human researcher could hold in their head at once… And when the next big decision comes, the answer is already there.”

“So to answer the question: Knit becomes the decision intelligence layer for the AI-native world.”

The growth

Knit was founded in early 2023 by Aneesh Dhawan and Raahish Kalaria. The first version of the product was simpler than what exists today. The modern version, the “AI-native neo-firm” with forward-deployed researchers and a unified quant-plus-qual platform, came together over the following two years as they signed customers like T-Mobile and NASCAR.

Knit closed a $16.1M Series A in September 2025, co-led by GFT Ventures and Ashton Kutcher’s Sound Ventures, after growing 5X the previous year. Today, the company has 70+ full-time employees, and more than 50 enterprise customers (half of which are Fortune 500 companies).

Selling to enterprise is hard, and selling to Fortune 500 enterprise is harder: long procurement cycles, internal budget fights, security reviews that take forever, and a strong default towards the legacy vendors they’ve used for years. But Knit has sold to them, and has sold to them fast (<120 days on average). The average customer more than doubles their spend in the first year.

This is all encouraging news. But, if Knit is going to reach the heights Aneesh described when I asked him about the long-term strategy, then their ceiling is perhaps 100x (or more) where they are today; a prospect both exciting and daunting. To get there, they’ll need a hell of a team.

The team and culture

There are some companies where the culture is nebulous. Talk to twenty people and you will get twenty different answers about how things work. Knit is not one of those companies. Knit, to me, seems to have a clearly-defined idea of what makes them special and how they work.

I asked the team all kinds of questions: what the founders are like, what meetings are like, what makes Knit different, what someone smart who quit might say. A few things stood out.

The talent bar is pretty serious. Most startups (okay, every startup) will tell you they hire well. Not all of them talk about it with the conviction people at Knit do. Jake Shanesy, who joined as Director of Product after a decade in B2B SaaS, said the talent density was “unlike anything I’d seen — people who could be coasting at bigger companies choosing to be in the trenches.”

Rachel Weems, Head of Engineering and one of the first engineers at DoorDash, put it more bluntly. “The founders at Knit hire people 10x better than themselves, who will in turn hire people 10x better than themselves.”

You will have an impact in your first week. On Jake’s first week, he was “in customer calls, shipping decisions, and writing specs. The expectation was that I’d already done the homework before showing up — and I had.” On her first week, Rachel was learning the codebase and putting together the new engineering interview process. Cullen Bates, a software engineer, was pushing changes by the end of his second day and owning a project by the start of week two.

The 30/60/90 day plan that Jeremy Grompone, Talent Acquisition lead, received was “a bit eye opening. It didn’t come off as micromanagement, but as a true guide on how to own the role they hired you for.” If you join Knit, I wouldn’t expect to slowly coast in for the first couple of months; I would expect to start doing serious work on day 1.

The founders are everywhere and “incredibly grounded.” Aneesh and Raahish, co-founder and CTO, are not the sorts of founders who set strategy and disappear into investor meetings. Jake described the founders’ defining quality as “ambitious humility — enormous conviction about where we’re going, paired with zero ego about being wrong on the way there.” He also said that the founders “carry an unusual amount of context across every part of the org — product, eng, GTM, customer.”

Erica Foster, Director of Product Marketing, said she’s “never seen a CEO as actively hands on in as many projects in a truly meaningful way as Aneesh.” Founder Mode-esque.

The work is hard. Nobody complained about it, though. Jeremy described the rhythm as being focused on what you do, not when you do it or how long it takes you to do it. Output and its results are the goal. You can see this in the way Knit runs meetings: they are short and always end in a decision. Not much fluff or “let’s just align on this” calendar blocks where you have small talk for an hour. (Hurray!)

“We don’t sugarcoat,” Aneesh said, describing how he pitches candidates. “It is hard work. But what makes it worth it is the impact you have. The work we do at Knit influences billion-dollar decisions for some of the most iconic brands in the world.”

Should you join Knit?

Knit’s careers page is right here. What should you do about it? I’ve written a lot about the product and what the culture and expectations are. So perhaps it’s fitting to filter some people out by writing about the things you cannot expect to do at Knit:

• You cannot wait to be told what to do.

• You cannot work slowly and precisely on everything. (Cullen said the engineering team moves fast and ships scrappy versions to iterate on.)

• You cannot operate on tenure. The best argument wins.

Knit’s team of 70+ full-time employees is spread throughout the US, with headquarters of ~25 folks in NYC. While engineering and PM roles are required in NYC, check out their careers page for which positions are remote or centralized around other smaller Knit hubs (Chicago, DC, and San Francisco). In general, though, I’d say that Knit is one of those rare companies that has roles both for people who are excited about working in-person and for people who work remote.

Stage is another reason you might join. Arnaud Mondoulet, principal software engineer, pointed out that Knit is in “a sweet spot. You have product-market fit and traction, less risk than [seed stage], and still tons to build.”

If all of the above sounds interesting, then I think you might join Knit because you see an opportunity to get in early with an excellent team with a company that could one day be worth many, many billions of dollars. Aneesh’s short-term vision—to become the most important research company ever built—is already big. The long-term vision is even bigger. I think that, if you are convinced by the vision and encouraged by the traction so far, you might want to check out Knit’s open roles. They are currently hiring across most departments.

Thanks to Knit for supporting Next Play and making this essay possible.

In 1956, a young Hungarian man fled his country’s violent revolution. He escaped across the border to Austria and then boarded a ship to the United States, hoping for a better life.

In 1992, a teenage boy and his mother left a village outside Kyiv with little more than they could carry, fleeing the collapse of the Soviet Union. They settled in a small apartment in California.

In the mid-2000s, two brothers from a small village in Ireland got the idea to go chase their ambitions in the United States. They flew across the pond to attend MIT and Harvard.

Who are they? The Hungarian man was Andy Grove, who ended up running Intel during arguably its most successful era. The Ukrainian boy was Jan Koum, who co-founded WhatsApp. The Irish brothers were Patrick and John Collison, who co-founded Stripe.

But there is another way to answer that question: they are immigrants to the United States. And they are far from the only ones. Immigrants have founded the majority (55%) of America’s unicorn startups. This is to say nothing of the immigrant employees of those startups, who themselves have unlocked success for these companies. 80% of unicorns feature immigrants in key roles (leadership, product, management). These are people who have built lives in the United States; have themselves gone on to build things (like Eric Yuan, who founded Zoom).

Immigrants are foundational to American innovation and tech.

And, yet, if you are a startup founder or an employee looking to work in the United States, it has become extraordinarily difficult to realize your dream. There is no dedicated startup or founder visa. Recent changes to H1B visas make that path virtually unworkable for most startups. What’s left (O-1, EB-1, EB-2 NIW visas) can work on paper, but in practice are way too expensive and out of reach for exactly the kinds of people who need them most.

So who is fixing this? Who is giving great people a path to come to the United States, and who is giving companies the talent to bring great people in? This seems rather important.

The answer may be Manifest OS. Founded by Dan Mishin, an immigrant himself, Manifest is taking a new approach to immigration that turns the standard predatory law firm business model on its head and makes immigration significantly more feasible for great people everywhere.

The problem

Imagine you are a founder (or recruiting person) at a startup and you’d like to bring in some talented people from abroad. If this is you, your experience might look something like this:

• You realize the visas in play (O-1, EB-1, EB-2 NIW) are complex and hard to get.

• Most firms don’t care about your business and won’t reply for weeks.

• The firms that do reply don’t give you a firm commitment on cost or timeline.

• The big, ‘reputable’ firms charge insane amounts. Like $1,000 per hour (or more).

• The smaller firms and independent lawyers are expensive, too, and hard to vet.

It ends up becoming so much friction on your side to sponsor somebody that you end up deciding to exclude the immigrant pool altogether. Of course, this is not ideal: immigrants have been, and continue to be, extremely important for American startups. But what can you do?

(The same is true, by the way, if you are a founder trying to enter the United States to start a company; you deal with all of the problems I just described above. A mess.)

But let’s say you get lucky. Say a promising firm does reply and you decide to work with them. If this happens, welcome to the frustrating world of:

• Insane hourly bills (you have no idea when it will end).

• Slow email responses (sometimes you follow up 2, 3, 4 times).

• Unclear communication (“Sure, I’ll get back to you on this”).

• No clarity on where things are in the process (even many months into the process).

• No idea what will happen if it fails (is all my money just gone?).

This is basically a nightmare; talk to anyone who has been through this process and, unless they were very lucky, they will repeat all of the same pain points. Immigration law is a mess.

The current model tends to work for law firms, which often have more demand than they can service and don’t face differentiated competition. The partners of these law firms reap the profits of their hourly billing, of your complicated case, of your stress. These profits rarely ever get distributed back into the business—because why would they? The model works.

Well. The model works for them. The law firms. The partners. It does not, however, work for you. Manifest is changing all of this with an honest, outcomes-focused model for law.

The product

Manifest is two companies that operate in tandem. The first company is Manifest OS, a NYC-based tech company that builds software, owns the brand, and runs a centralized back office. The second is Manifest Law, an actual law firm. Manifest OS powers Manifest Law.

This structure matters a lot. Most legal AI startups today sell software to existing law firms, the idea being that AI can make those firms more efficient. But Manifest doesn’t do this; founder Dan Mishin told me that “selling AI to law firms can’t fix the problem. $6B went into legal tech last year, most of it toward tools that make the existing law firm model more profitable without changing whose interests it serves.”

So, instead of selling AI to law firms, Manifest built its own.

They partner with elite, hand-picked attorneys (the acceptance rate is <1%). Those attorneys join the platform and only do one thing: practice law. Everything else, from intake to document collection to drafting to billing to operations, is handled by Manifest OS. The lawyer focuses on the legal work while Manifest handles the rest. And so, if you are an individual or a company who wants immigration help, the experience looks like this:

1. You meet your attorney.

2. You get a flat-fee quote for the outcome (not the hours).

3. You get a single dashboard showing exactly where your case is, what the steps are, and who is doing what. When you reach out, people respond fast (and helpfully).

4. If your case is denied, you could get a partial or full refund.

It is in many ways the antithesis to the way most law firms today work, even the fancy ones. It’s the way that, if you were a normal person designing the ideal way for a business like this to work, you would probably design it. It’s the way many other parts of our economy work: identify an outcome you want, pay for it, and get as much support as you need along the way. Why can’t law be this way? With Manifest, it seems as though it can.

Oh, and it works. Manifest Law’s visa approval rate runs higher than the USCIS national average for EB-1 and O-1 cases. They have already done this work for thousands of individual clients and across more than 150 corporate immigration programs (including companies like Whop and Orchard Robotics). The engine is on and it is humming.

The strategy

Solving immigration law is wonderful (and perhaps relevant to some of you reading), but I suspect it isn’t the only reason Manifest OS recently raised $60M from Menlo Ventures, Kleiner Perkins, First Round Capital, Quiet Capital, and others.

Surely there is a big long-term vision, then? I asked Dan Mishin, founder and CEO of Manifest OS. He said the goal is to provide “a trusted global platform powering hundreds of AI-native law firms delivering best-in-class legal services across every major practice area.”

Immigration was the logical place to start because it was a problem the Manifest team knew inside out; Dan himself went through all the broken pieces of the US immigration system. Manifest is already delivering thousands of positive outcomes in immigration, and there is still plenty of room to grow. They have probably not even achieved five percent of what is ultimately possible with immigration in the United States. The ceiling is high even with the current scope.

But, if things go well, Manifest will go global. They will expand outside of immigration law. They will become the most recognizable brand in law on the planet. Because their outcomes-based, AI-driven model makes law more accessible, there will be net new demand for all kinds of legal advice. Anyone with a problem of almost any scope will be told, by a friend or a colleague, “oh, you can just go talk to the people at Manifest.”

So how do they get there? A few things likely need to happen.

One, Manifest has to become (or remain) the best place in the world for elite lawyers to run their practice. Manifest lawyers are great today, and for the business to really win, its lawyers need to continue to be great in the future. The good news is that I think the Manifest pitch is quite appealing for lawyers. You join Manifest, you only practice law (no admin, billing, business development, lead gen), and you get a bunch of tools that make you faster and better. Menlo’s Croom Beatty, an investor in Manifest, compared it to Uber: “just as Uber created more supply by increasing the number of drivers on the road, Manifest uses AI to dramatically increase the throughput and productivity of existing lawyers.” I think most lawyers like that pitch.

Two, the model needs to continue working as Manifest moves into new practice areas. There is plenty of reason for optimism here: most legal practice areas suffer from the same broken model that immigration does. If Manifest can make its model work in those other areas, it wins.

Three, Manifest OS needs to keep building great software that makes the whole thing defensible; the tools that differentiate Manifest lawyers from others. This is where “operate the service, don’t just sell the software” pays off. Most legal companies sell tools to law firms and never see how those tools get used. But Manifest operates the service. This means every single case makes the product better, and every product improvement makes the next case faster. As Dan puts it, “Most legal AI companies are selling shovels; Manifest OS is mining.”

There are early signs all of this is working. Manifest has 3,000+ clients, 100+ attorneys, and has worked with more than 150 corporate immigration programs. They just raised a $60M Series A at a $750M valuation from some of the best investors in the world.

Manifest is still a young company, but it appears to be well on its way to building something that changes all of law: a unified brand sitting on top of hundreds of practices, all AI-native, all sharing the same back office, all delivering the same exceptional experience around the world.

What to do with this information

Manifest is an interesting company solving a serious problem in a new way; that’s part of why I’m writing about them. There’s one world where you read this and simply update your priors about immigration law (or all law, for that matter). You’re now aware of this new, ambitious model and approach to the space. That may be enough.

But I do think, for certain people, there are more practical steps you could take.

If you are a founder (or employee, or someone who wants to live and/or work in the US), you could try out Manifest Law’s services for yourself. If you work at a company that has a corporate immigration program and/or wants to bring people in from overseas, you could also have a look. If the model and the product is as good as outside indications make it seem, it might be prudent to at minimum book a consultation and learn more.

Manifest is also hosting a webinar on Tuesday, May 26th from 1–2pm Eastern on visa options for founders and startups. Whether you’re building something of your own or trying to make the jump from a bigger company into the startup world, they’ll cover the paths that actually matter. You can sign up here.

Finally, if you are interested in what Manifest is building and would like to help them build it, you can look at Manifest OS’s open roles here. They’re hiring for their SF and NYC offices.

Thanks to Manifest for supporting Next Play and making this Spotlight possible.

What would happen if you brought a coffee machine to an uncontacted tribe? A real fancy machine, maybe one of those expensive La Marzocco ones. The kind of machine capable of producing some of the best-tasting espresso in the world.

So you go to the tribe and you just hand it to the first person you see. What happens?

Well, nothing happens, most likely. The coffeemaker is capable of producing some delicious espresso if you use it, but it is a useless hunk of metal for someone who does not know how to interact with it. And so your gift might be received with little more than a shrug.

There is, however, one obvious solution: bring them great documentation. Hand them a beautiful instruction book that helps them figure out how to use the coffee machine so they can get a ton of value out of it. Now your gift is more than a hunk of metal. Now it is something extremely valuable. As you leave the tribe, they might ask you to bring more next time.

This is not too far off from what happens in the real world:

• Products are (usually) only useful if people know how to get value out of them.

• Companies are (usually) only successful if the people inside of them can create value.

Problem is, documentation is not easy. It’s not easy to write effective instructions for using your product or working at your company. It’s not easy to make that writing look good and be easy to navigate. It’s not easy to make sure every single line of your documentation is up-to-date, especially when your product changes near-daily. And, in today’s world, it is not easy to make sure your documentation is well-designed for AI agents to read and use it.

That is why today’s spotlighted company, Mintlify, exists. Documentation now serves both humans and AI agents, and what agents need is constantly evolving. Mintlify keeps you current automatically. Companies like Anthropic, Coinbase, Polymarket, and OpenClaw already use it; if you’ve read developer docs recently, there’s a good chance you’ve seen Mintlify in action.

AI agents now account for 60% of Mintlify’s traffic, and what they require changes constantly: llms.txt, MCP servers, content negotiation, structured markdown, and more. Meanwhile, humans increasingly expect to *chat* with docs rather than read them, and teams need AI to automate *writing* docs so humans don’t have to think about it.

This is the problem Mintlify solves: instead of scrambling to keep up with each new standard yourself, Mintlify handles it automatically—for developer docs, internal wikis, and help centers alike. To continue scaling this, last month Mintlify raised a $45M Series B at a $500M valuation. Now may be a particularly interesting time to join.

So I sat down with the team to ask: what does the product do today? Where is it heading in the future? How has the company grown so far? What’s the culture like, and who should consider joining? Answers to those questions—and others—below.

The product

Companies use Mintlify to produce and update documentation, both external and internal. The external documentation exists so customers and their AI agents can understand what a product does and how to best get value out of it. The internal documentation exists so employees—and their AI agents—can understand what the company does and how to best generate value for it.

This sounds rather simple. Aren’t you just describing a nice CMS? And the answer is no, not really. In fact there’s a good argument that one of the main reasons Mintlify succeeded early on was because they did not build a CMS; they recognized that developers are not content marketers. They don’t want to log into a fancy WordPress-like dashboard.

Instead, Mintlify’s approach from the start has always been docs-as-code. Which means developers, the ones generally responsible for documentation, can write documentation inside of a repository and ship it the same way they ship code. And the documentation they ship with Mintlify is beautiful, which was (and is) a rarity in the documentation world. Better yet, it is beautiful by default; developers don’t have to waste hours on design.

So that was phase one. For a while, Mintlify was considered by many to be the most beautiful and effective way to produce documentation. (They still are.) People were happy with it.

But times have changed. AI is here. AI agents are now reading the majority of documentation; they’re writing a lot of it, too. This means that, as a company, you need to make some changes. You need to make it easy for AI to understand your docs (or else it won’t use your product correctly, and end-users will drop it). You need to make it easier for humans to interact with your docs (people expect chat-like features in our current world). And you need to make it easier for ai to produce the docs themselves (so you can write better, faster).

As the world has evolved, Mintlify’s product has evolved with it. Some examples:

• You can now structure your documentation in ways that make it easy for AI agents to read them (ex: Mintlify partnered with Anthropic to co-develop llms-full.txt for this).

• You can build an AI assistant into your documentation so people can easily get answers.

• You can use AI to help write, or completely write, your documentation.

What Mintlify is building now is what they call an “intelligent knowledge platform.” It is still a documentation product; but it’s an AI-native documentation product that treats documentation as important infrastructure, both inside and outside the company.

Now, that all sounds nice. But will they win?

The strategy

For some companies, the arrival of AI raises worrying questions.

• “Will we survive?”

• “Is AI going to make our product irrelevant?”

• “Does our business model still work?”

These are often valid questions. With Mintlify, though, you could make a strong argument that the existence and continued popularity of AI (especially AI agents) has made an even stronger case for their business. That a world with billions of AI agents roaming around is a world where documentation probably becomes more important than ever.

It is a world where, as co-founder and CTO Hahnbee Lee said, “Mintlify is the knowledge infrastructure for every company.” If the company wins, then in 20 years “every organization uses Mintlify for both their external-facing knowledge — docs sites, help centers, API references — and their internal knowledge like wikis and runbooks. We become the single layer that captures, organizes, and serves a company’s knowledge to both humans and AI agents.”

It’s true that Mintlify is not the only company building tools for documentation. But one strong reason they believe they’ll win out is their AI-native approach. “Most competitors are retrofitting AI onto legacy architectures,” Patrick Foster, a software engineer at Mintlify, said. “We built for an AI-first world from day one.”

This approach has always been in Mintlify’s DNA. Early on, it was doing things like putting the documentation tools where developers actually wanted them (in the code editor) and shipping beautiful design by default. Now, it’s building a product full of new capabilities—llms.txt, MCP, agent-readable markdown, AI assistants—to make docs work in a very AI-obsessed world. Rather than retrofit features onto a platform built for humans, Mintlify is working backwards from a world where AI will become an increasingly important part of documentation.

“By mid-2027, nobody will be writing product documentation or internal wikis by hand,” Hahnbee Lee said. “Most people in the docs space still think of AI as an assistive tool. We think it replaces the writing entirely. The role of a docs team shifts from authoring content to curating and verifying what AI generates.” And, of course, the role of an end-user shifts from reading documents themselves to overseeing an AI agent that read the documents.

Documentation, then, is perhaps a humble way to describe what Mintlify is building: the knowledge layer that every product, team, and agent depends upon.

The growth

Like many of their fellow alumni, Mintlify’s first 100 or so customers came via startup connections and referrals while they were going through Y Combinator’s Winter 2022 batch. They added a ‘Powered by Mintlify’ badge to those startup customers’ docs sites. Eventually, those sites started getting serious traffic, thousands of developers were seeing Mintlify’s name every single day… And momentum only picked up from there as those viewers started trying Mintlify at their own companies

Now, Mintlify’s list of customers includes names like Anthropic, Coinbase, AT&T, Fidelity, and Kalshi: this is no longer a tiny startup selling to other tiny startups. Mintlify is for everyone.

They’ve also raised a lot of money. Mintlify’s most recent round, a $45M Series B led by a16z and Salesforce Ventures which valued them at $500M, brings the company to $67M total raised. All this investment is probably backed in part by a vision of the future where documentation becomes more important than ever, and in part by what’s happening right now: “Over 20,000 companies use Mintlify, and 100M+ people per year read docs powered by our platform,” Hahnbee said. One can only wonder at how many AI agents read, too.

If Mintlify were a smaller and earlier company, one version of this Spotlight would involve me writing that, while their growth is modest and they are still early, the future looks bright. But that’s not the version I’m writing. Rather, Mintlify has already shown impressive growth. They’ve already sold to virtually every size of company—from tiny startups to Fortune 500 companies—and are growing revenue fast. There is a lot to be happy about today; and, if things go well, there will be far more to be excited about over the years to come.

The team and culture

If you work at Mintlify, you can basically do “what you want,” I was told. “You can exercise your personal taste and style extremely easily — there’s no committee blocking you from shipping.” Meetings are “short and to the point” and work at the company is, like the product, focused on working backwards from outcomes: what do we want to achieve and how do we get there fast?

Reed Barnes, one of Mintlify’s software engineers, told me that “the team ships faster than companies 10x our size.” Jarod Hatch, an AE, said that the pace is “electric… Unlike anything I’ve experienced. I was on customer calls within my first few days, and the team was immediately supportive — sharing context, looping me into deals… It reminded me of the team culture in competitive sports: everyone pushing each other to be better, no egos, just results.”

Lots of startups say they ship fast. It’s easy to raise an eyebrow. The best way to double-check is to see what the company actually shipped. And, in 2025 alone, Mintlify shipped…

• Mintlify Agent (self-updating docs that open PRs automatically)

• MCP server auto-generator (one command turns docs into agent tools)

• llms-full.txt (co-built with Anthropic, now industry standard)

• AI assistant (chat with your docs, 23M+ queries monthly)

• API Playground rebuild (completely re-architected interactive “try it” experience)

• Web editor overhaul (clear, simple interface that commits cleanly back to Git)

• Eight new themes (expanded visual options without custom CSS)

• One-click LLM copy (“Copy for LLM” buttons plus agent analytics)

… And a whole lot more. AI is moving fast, but so is Mintlify and their team. Speaking of that team, there is one final question to ask: Should you join?

Should you join Mintlify?

There is a lot to like about Mintlify. They’ve grown fast and built a well-loved product, and the culture is rather low-BS and gives you the autonomy to exercise your own judgement and move fast. They’ve already signed many of the companies you could describe as their ‘dream customers,’ like Anthropic and Coinbase. And there is a good argument to be made that their product will only become more relevant as AI adoption increases.

But is it the right fit for you, specifically? Hahnbee Lee, Mintlify’s CTO, said that the company is a good place for someone who “wants ownership and fast feedback loops. People who succeed here are earnest, high-agency, and take real pride in their work. People who don’t work out tend to be the opposite: they need a lot of structure, wait to be told what to do, [and] treat their work as just a job rather than something you genuinely care about.”

It’s also worth considering Mintlify in the context of their stage; they’re fresh off a $45M Series B. If you are looking for a startup that (1) still has a ton of room to grow (e.g. to 10x, or more, your equity grant) but (2) has been somewhat de-risked, Mintlify’s current situation might be attractive. In other words, if building the world’s best knowledge infrastructure product sounds exciting to you, then there is no better time to join than today.

You can take a look at Mintlify’s open roles here. Most of the team works in-person in San Francisco, and they’re currently building out an NYC office (where they’re hiring AEs now and open to hiring others in the future). “If someone cold emails us with a sample project or a thoughtful take on something we’re building, that immediately stands out,” Hahnbee told me. If you decide to go that path, you could read some of our ideas here.

Thanks to Mintlify for supporting Next Play and making this Spotlight possible.

A long time ago, in a long-forgotten corner of the planet, there was a peaceful village inhabited by peaceful people. They lived a quiet life: fishing, dancing, hunting, cooking.

One day, a new being came to the village. This was the first time anyone new had ever come to the village. This person, if you could call them a person, introduced themselves as The Oracle. They looked a lot different from the villagers, and talked differently too, and many of the villagers were skeptical. Some even voted to have The Oracle banished.

But as the months went on, the villagers realized that The Oracle was actually quite powerful. It was extremely intelligent, at least in some ways. It had capabilities the villagers did not. The villagers began to build trust in The Oracle. They gave it more power. Like the power to do their work for them, and the power to run things.

Centuries passed. The village grew into a civilization. And then something strange started to happen, something that would shape the entire economy: people relied on The Oracle to decide which products to buy. This new shift was concerning for many companies.

They started asking…

• How do we get The Oracle to recommend our products?, and…

• Where does The Oracle even get its information?, and…

• Do we need an entirely new approach to marketing in this version of the world?

These are essentially the questions that Profound, today’s spotlighted startup, exists to answer. The Oracle in the story is AI, and it is quickly becoming one of the main ways people get information. AI is also quickly becoming one of the main ways marketers do their work.

Which means that Profound, which provides the marketing tools for this new world, is interesting. Being a near-direct pipeline to revenue is usually a pretty good bet, which perhaps explains why Profound has grown so fast and raised a $96M Series C this year.

So I wanted to dive deeper. What are they building? Why do they believe they will win? What kinds of people thrive there? And: Should you join?

The product

It is not always easy to describe how a software product helps its customers make more revenue (sometimes you have to draw a 17-step chart). But with Profound, it is easy: Profound helps companies make more revenue by giving them the tools to understand, control, and scale AI marketing. In other words, Profound gets AI to talk about your company (among other things).

This new category is called AEO: Answer Engine Optimization.

In the past, companies used tools like Semrush and Ahrefs to dominate a similar domain (Search Engine Optimization). These tools helped you analyze your company’s ranking on search engines, like Google, and helped you strategize about ways to rank higher. None of them were nearly as ambitious as what Profound is building, but they are a helpful analog.

The issue with SEO tools—with any tools pre-Profound—is that they are not built for people chatting with AI. The best practices are different. Claude is not Google Search. Tracking and optimizing your company’s performance on Google and in AI conversations is a completely different game; it is a kind of game that requires completely new tools. And the exciting piece is that AI search may be on track to become much bigger than SEO ever was.

Today, companies mostly use Profound in two ways: to track their AI search performance (which Profound calls ‘Monitor’) and to optimize it (which Profound calls ‘Create’). Those categories are rather self-explanatory—the tracking tools help you keep tabs on your performance, and the optimizing tools help you do things that increase your performance.

So the basic flow for a customer (like Ramp) is something like this:

1. Sign up for Profound.

2. Figure out how your brand is doing today; a vitals check of sorts.

3. Learn what steps need to be taken to perform even better.

4. Use Profound’s AI Agents to execute on those steps.

5. Perform better, make more money, and so on.

Powering much of Profound’s platform are their Agents, which they announced alongside their Series C this year. In the past, Profound gave marketing teams visibility. But, “as customers operationalized that insight, they often had to export Profound data into separate orchestration and automation tools, learning and using multiple systems just to execute.”

With Agents, customers can now do it all with Profound.

Results with Profound (and their Agents) often come fast. Statsig 2x’d their visibility across all key mentions in one week by creating content and getting citations. Zapier became the most-cited domain for their most competitive prompts. Ramp increased visibility on their Accounts Payable solution from 3.2% to 22.2% in one month by publishing targeted content.

Clearly, if someone came to you and told you they could (almost) magically get a trusted entity to recommend your product to a bunch of people… You would say yes. And lots of companies have said yes. Profound is a pretty hot startup these days.

“Hot startup,” however, does not always mean “successful in the long-term.” So how does Profound achieve that? What is the long-term vision?

The strategy

The optimistic view of Profound’s trajectory is that it might be on pace to become the most important marketing tool in the world. This is certainly the view that James Cadwallader and Dylan Babbs, Profound’s co-founders, take. “The SEO and paid search industry is a $90B market,” they told me. “As discovery shifts from search engines to AI, that spend has to find a new home. Every brand on the planet is going to need to figure out how they show up in this new world of AI search.”

Just like how every company eventually needed Salesforce to manage their customers or HubSpot to manage their marketing, James and Dylan said, companies will need Profound to manage where and how their brand shows up in AI search. In a world where you simply ‘talk to the internet’ to decide what to buy (James believes that in five years, $1T of commerce will be driven by ChatGPT alone), a tool that can help make sure your product(s) are getting recommended becomes extremely valuable.

And extremely valuable is important here. Profound will have to break new ground. Semrush, arguably the most popular SEO tool of the last decade, is currently valued at ~$1.8B on the public market. That’s not bad. But Profound was already valued at $1B as of their Series C; the vision is, and has to be, much larger than the SEO tools that preceded it.

So what does it take for Profound to become the most valuable marketing tool on the planet? For the company to become really big (say, $10B, or $100B, or bigger)? I’d say that this success rests roughly on the following series of events:

1. People increasingly rely on AI for buying decisions (note: 45% already do).

2. Profound cements itself as the best tool for getting AI to talk about your brand.

3. Profound drives ~trillions in revenue for companies and exists as far more than just an AI search tool—rather, it becomes the most important tool marketers use bar none. Powered by Profound’s AI agents, which help marketers get more, better work done.

With Profound driving massive amounts of revenue and sitting at the center of most companies in the world, they would have all kinds of options (related and not related to AI search optimization) to expand their product and further grow revenue.

One advantage is that—though they are not the only startup in the space—Profound has made headway quickly and, to many, established themselves as the go-to product for AI search optimization. But another advantage, and the one they will probably need to rely on most if they are really going to win here, is their team.

“I joined Loom after their Series C,” Nick Lafferty, founding marketer at Profound, said. “Loom was about 300 people and the marketing team was about 30, which included our brand team. Profound is about 170 people and our marketing team is about 8. We are way more efficient as a company and as a team, both due to AI and due to the caliber of people we hire… I can confidently say that I have never worked with such a talent-dense [team] in my career, and I’ve been doing this for almost 15 years now.”

The growth

Profound’s co-founders, James and Dylan, launched Profound together in August of 2024. They raised their seed round from Khosla and Saga VC with just a product demo (no pitch deck).

What followed was one of the more aggressive fundraising trajectories in recent SaaS history (perhaps part of why you may have heard of Profound). In June of 2025, Kleiner Perkins led a $20M Series A. Two months later (!) Sequoia led a $35M Series B. And, in February of this year, Lightspeed led a $96M Series C at an impressive $1B valuation.

What’s fueling all of this? Well, in addition to startups and smaller businesses, Profound now works with over 700 enterprises—including more than 12% of the Fortune 500. The customer list includes names like Target, Walmart, U.S. Bank, and MongoDB. They don’t share ARR publicly, but it’s not hard to do some (rather encouraging) napkin math.

Of course, Profound does not do any of this (none of the growth, none of the fundraising, none of the product) without their team. And so, without further ado…

The team and culture

Imagine you walk into one of Profound’s offices in New York, London, San Francisco, or Buenos Aires and you pick a person at random to talk to. You go up to their desk. Say hello. Within minutes, Sharanya Pogaku (ops) told me, “you would be so impressed by them that you would want to work at Profound and you would want to surround yourself with other people like them.”

That is a bold thing to say. But the people (and the associated energy and culture) were the common thread among the conversations I had with the Profound team. “A lot of people say it’s the team,” Profound’s Head of Marketing Trevor Pyle told me, “but we’ve built a real snowball effect. We’ve hired a few people early who were hungry, ego-less, hard workers… They saw the opportunity, got excited, and then helped us hire more people like that.”

Beyond all of the things you might expect to hear from a startup (we work hard, we ship fast, we take ownership), one theme did surface as unique: an intense focus on effectiveness.

One input to this pursuit of effectiveness is “extreme directness and being aggressive about getting results and working well together,” Trevor said. It’s hard to be honest if you’re tiptoeing around your point with buzzwords and fluff.

Another input is efficiency. “There are no calories wasted hemming and hawing, needing to align people.” Joey Loi, Head of Product said. “Just do it.” Meetings are not a big part of the culture. Nor are vanity metrics and other versions of peacocking. Just urgency and accountability.

Even for new arrivals, work at Profound is intense. On his first week, Joey (product) got straight to the whiteboard and started building with design and engineering. Nick (growth) was vibe coding new parts of the website on day two. Erin Rairdan, a product designer, was dropped into a “high-visibility, high-impact” project on her very first day.

The team at Profound is, above all else, a talented and ambitious group of people who work hard, are very honest, and actually like working with each other. They also like spending time together outside of work:

Should you join Profound?

Besides the technical stuff, like compensation and equity, conversations about whether you should join a company generally come down to two pieces: (1) Do you believe in this company’s ability to realize their grand vision, and (2) Are you the right kind of person to work here?

As for the former, I hope this essay has been useful to set the context for what Profound is and what their product does. But you could also research outside of this piece (and probably should, if you are considering joining somewhere full-time).

How would you research? You could read Profound’s website. Read their competitors’ websites. Brush up on current SEO tools and their offerings. See if you can find a way to talk to any of Profound’s customers—there are many. Interview them about the product (and then, if you want to apply, tell Profound about those findings). You could check out sessions or join an event at Zero Click, Profound’s gathering for AI marketing experts. (Past sessions from Zero Click SF in April are linked.)

Regardless of what you learn, you should probably be somewhat convinced that AI is here to stay and/or will get better in the future; these are upstream of the thesis that people are going to rely on AI for purchasing decisions. And you should be convinced that there is room for Profound to become a significantly more valuable company than they are today.

As for what kind of person you should be? Well, I think you should be an honest person who can take direct feedback and criticism. I think you should be willing to work in the office five days a week. I think you should want to work really hard; one employee told me that even if you’re highly intelligent and skilled, Profound is an especially good fit for people who are “currently in an ambitious period of their lives.”

If you do feel like you fit these criteria, then you may want to check out Profound’s open roles. They are hiring for just about every role you can imagine at their offices in New York, London, San Francisco, and Buenos Aires.

“We don’t care about the perfect resume or the traditional path,” Profound’s co-founders said. If you want to send a cold email (or get even more creative), “do it.”

Thanks to Profound for supporting Next Play and making this Should You Join essay possible.

Say you work at a company and you want to try out a hot new AI tool for writing code. What do you do? How do you go from “wanting the tool” to “using the tool?”

In the early days, when everyone at the company can fit into a small office and the ‘engineering team’ is just the technical co-founder, getting the tool would be easy. It might be as simple as asking for it across the office, getting a yes, and pasting some company card details in. Success.

But things change as the company grows. There are more people, more departments. The company has to be more careful. More “stakeholders” are involved. There’s finance. There’s legal. And the process of purchasing something becomes a very official-sounding word…

Procurement. An unsexy (but extremely necessary and frustrating!) part of running a company.

Today, procurement at most companies looks something like this: you file a request on some platform. It sits. Six different stakeholders coordinate across twelve different chats. Legal redlines the contract. Finance has questions. Your boss approves. Then your boss’s boss. Two months after you actually needed the thing, you get it.

And then, just after you get approval, you discover the marketing team actually bought a similar tool eighteen months ago. It auto-renewed last Tuesday. So you’ll be stuck using that instead.

What if there was a better way to run all of this? One clean, unified platform that could make procurement feel as straightforward as it might at a five-person startup? Omnea, the startup we are spotlighting today, has built just that. They’ve managed to 3x revenue in 12 months with customers like Spotify, MongoDB, and Albertsons by building a system of intelligence that will—if they succeed—sit at the center of virtually every company in the world.

Naturally, there are questions to ask. What is the long-term vision here? How does the product work? Who is on the team? And, of course: Should you join Omnea?

The product

To fix procurement, Omnea took every fragmented piece of the process and organized them into one powerful AI-native platform to run procurement from start to finish.

It might sound strange, but very few companies today actually have a clean start-to-finish platform for procurement. This status quo ends up causing problems:

• It takes far too long for people to get approval for important items

• Policies (security, budget, etc.) are not enforced consistently

• Companies end up buying things they don’t need or already have

• Contracts auto-renew at higher prices because no one tracked the deadline

• Employees go around the system entirely, creating shadow IT and shadow spend

So how does the product work?

Imagine a marketing person at MongoDB wants to purchase Resend. In the pre-Omnea version of the world, they would go through the mess I described earlier.

With Omnea, the marketing person could simply open Slack and type in what they need. Omnea AI would ask the right follow-up questions and route the request to all relevant people (e.g. in finance or legal) in parallel and track everything in one place, from start to finish. Everything would move faster. It would require less time from people involved. And it would get done at a higher quality (issues would be flagged, policies enforced, and so on).

This is a big deal for companies and the procurement people within them. “I interviewed 300 procurement leaders,” Ben Freeman, co-founder and CEO of Omnea, said. “[They] all said, ‘our process is hated, our tools are clunky, and we want to do a better job, but the stack holds us back.’” Omnea is, finally, a better stack; in fact, better than a stack, it’s one platform for all of it.

The days of disconnected tools, inconsistency, inefficiency, and being in the dark are gone. Now companies can just use (and are using!) Omnea. This is a great start. But what’s next? How does Omnea win in the long-term?

The strategy

“The category we’re building is what we call AI Supplier Relationship management,” Ben Freeman told me. “Omnea started as a system of record… It then became a system of action… Where we’re headed is a system of intelligence: one where every intake, approval, risk review, and renewal runs through a single, automated workflow.”

“In 20 years, the best companies in the world will be the ones that use procurement and supplier management as a source of competitive advantage. Their businesses will grow faster and more efficiently because they’ll have the right systems in place.”

There certainly is one major factor in Omnea’s favor: the market is huge. Basically every single company on the planet needs to do procurement, and if the stakes are as high as Ben describes them (‘growing faster and more efficiently’), then every company should want to use Omnea. Better yet, procurement is not exactly the hottest market startups are going after; it is more or less ignored. And a huge, unsexy, ignored market is often a good place to be.

But why does Omnea win in particular? Why choose them?

One reason might be that Omnea is taking an AI-native approach rather than bolting AI onto an existing system. “The market right now is about 45% substance and 55% noise when it comes to AI,” Ben Freeman said. “A lot of vendors are selling AI as a marketing stunt… [But] we were built AI-native from the start. Not as a feature bolted on, but as a core part of how the product works.” This approach, Ben says, “changes what’s possible in terms of the workflows we can automate and the speed at which we can move.”

You could also point to Omnea’s obsessiveness about their customers as a differentiator. “People want to use [Omnea] in all kinds of different ways,” Josh Reynolds, a senior PM, told me. “I’ve consistently seen Omneans go above and beyond to ensure that our customers can use Omnea the way they envision.”

The care for customers is “not performative,” Ben said. “We go into customer environments and operate at a level of depth that incumbents and most competitors don’t. We win most competitive RFPs, and I think that reflects this more than anything else.”

The fourth and perhaps biggest reason Omnea believes they will win is their team; both the culture and the talent density. (But we’ll cover that later on in the piece.)

The growth

Omnea has 3x’d their revenue in the past 12 months, driven by enterprise deals with companies like Spotify, Albertsons, and MongoDB. But how did they get there?

The origins were rather humble: cold email emails. “I built some basic outreach sequences in a basic LinkedIn automation tool to figure out who would speak to me. We had no brand, so I just sent concise & direct messages,” Ben said. “From there, it snowballed.”

Early adopters began telling their friends, Omnea built a reputation for doing a good job, and more customers rolled in. The product improved fast. Then the team started attending conferences, hosting events, and hiring full-time people to scale GTM.

Before they knew it they were signing Spotify, Monzo, Gymshark, Albertsons, and others. They were closing their $50M Series B last year led by Insight Partners and Khosla Ventures. They grew to a team of 150 across London and New York (and are now hiring one person a day). Omnea still doesn’t have a huge ‘brand’, Ben said, and growth today still mostly comes from cold outreach and customer referrals. But a referral from a company like Spotify can go a long way. (Omnea now also grows via their monthly events with finance and procurement leaders.)

Besides the obvious, one uniquely encouraging thing about this growth trajectory is that Omnea has signed all kinds of companies. Startups. Massive retailers. Enterprise software companies. Just about everything in between. This goes a long way to alleviate the (reasonable) fears that people often have about startups, fears I’ve heard from founders: are we going to be able to sell to non-tech-startup companies? With Omnea, the answer is a resounding yes.

The team and culture

A company, Ben Freeman told me, “is just a group of people working towards a shared mission, limited by the capability of those people and the tools and systems they have.”

So what about Omnea’s people? What are they like? How do they work?

When I talked to the team, I was told on many occasions that the people at Omnea were the highlight; the big reason you build conviction in the company and decide to join. “Talent is the overwhelming reason Omnea is different,” Josh Reynolds, PM, said. “The colleagues I’ve worked with at Omnea are genuinely in the top percentile of colleagues I’ve worked with across my whole career, and this tends to compound and drive very positive outcomes.”

For some stats: roughly 1 in 10 people on Omnea’s GTM team have founded their own companies (some backed by YC, Sequoia, and Accel). The team are experienced and have cut their teeth at companies like Meta, Slack, Miro, JP Morgan, and McKinsey. Leadership is impressive, too: Omnea’s CEO, CCO, and CFO all previously helped scale Tessian.

When Rebe Tristan, Omnea’s VP of CX, met the team, “[that] sealed it for me. This was a company with real potential to win, and an incredibly exciting journey ahead.”

Perhaps part of the reason Omnea’s team is so effective is the way they’ve approached hiring. The company interviewed more than 10,000 people to make their first 50 hires. Ben, the CEO, still interviews every new hire. And when they hire, they prefer trajectory to credentials. “Almost everyone here is on a steep curve, which creates a compounding effect where the business and the team are accelerating at the same time,” Ben said.

This hiring ethos seems to have seeped its way into the culture, too. Fancy resumes and experience don’t matter much (if at all) when making decisions, I was told. Feedback is direct and honest; it’s not meant to be taken personally. Transparency is a big deal, and the leadership team shares things most companies don’t share company-wide (like pitch decks, valuations, and explanations for losing a deal). Performance and output matter a lot.

And like many great startups, Omnea expects you to be rather autonomous and agentic in the way you approach your work. “Before Omnea I’d only operated as a founder,” Leon Gibson, an AE, said. “I expected joining a fast-growing company to feel like a constraint on that, but it hasn’t. I still have real ownership and a strong sense of agency… [And] there’s nowhere to hide here. Top to bottom, everything is visible. [It’s a high] level of accountability.”

I also heard multiple times that people at Omnea are especially collaborative and focused on problem-solving rather than politics or personal ambition. “It [feels] different from any culture I’ve experienced before,” Arie Barendrecht, a General Manager, told me. “People really care about solving problems together.”

Should you join Omnea?

It’s often just as helpful to ask who isn’t a fit as who is, so that’s what I did with Omnea. If a talented person joined Omnea and didn’t work out, why would that be?

“It’s rarely a capability issue,” Ben said. “People who don’t find a fit usually find the pace and intensity harder than expected, rather than the work itself.” Omnea, though already at Series B, operates with a lot of intensity. If you’re ambitious and want to improve, this is good.

But not everyone actually wants an environment like that. If someone “was at a certain level of tenure in their career that they didn’t want to come in and be challenged, and if they almost expected to have their opinions overrule others because of tenure, then they would probably not be a fit at Omnea,” Josh Reynolds said.

The third reason somebody might not fit, Arie Barendrecht said, is that they may find the work “not structured enough. It’s a very candid, performance-driven environment with a lot of autonomy and not much scaffolding - which works brilliantly for some people and is hard for others. If you need clearly defined processes and guardrails, you would likely find it difficult here.”

My impression is that Omnea is an exciting place to join for people who care about actually doing the work: in other words, people who care most of all about performance and outcomes and less about all of the other distractions (like titles, credentials, tenure, vanity metrics, and “well we did this at my last company so trust me it’ll work here”).

If that’s you—if you want to roll up your sleeves and join a talented team to help build a product that could theoretically be used by almost every company in the world one day—then you can check out Omnea’s open roles here. They are currently hiring across most departments for their New York and London offices.

Thanks to Omnea for supporting Next Play and making this Spotlight possible.

In the summer of 2024, Ukraine controlled a territory in Kursk that relied on a single logistical route from a city called Sumy (in Ukraine) to a city called Sudzha (in Russia). It was against this bottleneck that Russia began to test a drone technology that has since become a defining weapon on both sides of the war: fiber-optic tethered drones.

Ukraine and Russia have since increased adoption of fiber-optic drones because they have one advantage that traditional drones do not: they don’t rely on wireless communication.

Why? What’s the deal with wireless? Two things, which go far beyond drones.

1. Just about everything in the world relies on wireless connection.

2. The technology to jam or interrupt wireless communications has far outpaced the technology to avoid jamming.

In warfare, for instance, the mere act of transmitting wirelessly—even with encryption—is what enables a company like CX2 to track down and engage drone operators. And it’s what has driven the Russian and Ukrainian militaries, among others, to purchase fiber-optic tethered drones.

But this goes way beyond the battlefield. Any time information travels wirelessly (hospital records, self-driving cars, satellite feeds, infrastructure controls), the signal itself is a vulnerability. And, as we’ll cover, current encryption is not capable of fixing that.

Enter Rampart, the company we are covering today. Its team, which includes ex-NSA PhD scientists, mathematicians and engineers, has developed a brand new technology that makes wireless signals look the same as the background noise. If signals can’t be identified, they can’t be intercepted or jammed. This is revolutionizing the future of wireless communications.

So, what exactly is Rampart working on? What is it like to work there? Why should you consider joining them? I sat down with the team to get answers to those questions and more.

The product

Imagine it is a sunny Friday afternoon and you are sitting down at a neighborhood café with a friend. The conversation turns to a subject you aren’t comfortable speaking about in public. So you suggest to your friend that the two of you start speaking in Spanish, a language you are both fluent in. You hope nobody around you speaks Spanish.

The Spanish-speaking approach is a rough (but useful) analog for modern wireless encryption. The message gets scrambled into something that is hopefully unintelligible. But everyone around you can still see that you’re talking. They can see who you are, who you’re talking to, when the conversation started, and how long it’s lasted. If they wanted to, they could come over and disrupt the conversation. They could even follow you home.

What if you didn’t want people to know you were talking at all?

Then you’d need to figure out some sort of communication system that was undetectable. A way of speaking to each other that made your sounds just like the background noise in the café, a way of speaking that only the two of you could possibly understand. Our hypothetical now fails us, because it seems impossible. And yet it is what Rampart is building.

The core innovation behind Rampart’s products is called UBDM and a related technology known as RSW. These technologies scramble the entire wireless signal so it looks exactly like random background noise. UBDM hides what you say (and then some), and RSW hides that you are saying anything at all. In both cases, there is no chance for an eavesdropper to intercept (recover the data in the signal).

This is a helpful visual:

In other words, with Rampart, your wireless communications are pure background noise to everyone except for you and the person (or device) on the other end of the comms.

Right now, this technology plays out in two distinct products. UBDM Core is the raw innovation itself, sold as a licensable IP block. It’s a piece of logic that can be dropped into any radio platform; the engine that other people build into their own radios and systems.

The second is the StrataWave UAS (Unmanned Aircraft System) Radio. Rampart’s two products both use the UBDM and RSW in different ways. UBDM Core enhances almost any existing system, preventing interception and providing persistent and secure connectivity without needing cables. StrataWave is a drone radio engineered for military UASs operating in contested airspace; it makes drones extremely difficult to detect or jam.

Rampart’s products, to date, have been sold for military applications. Talk to anyone at the company, though, and you will learn that the vision is far more ambitious than the battlefield.

The strategy

Warfare is not the only application for Rampart’s technology. Because, as I mentioned earlier, wireless connectivity is required for most things in the world to work. Hospitals, financial institutions, satellites, power grids, water systems—these are a few other examples of areas where Rampart’s technology could have a big impact.

“Rampart is framing a new field of discovery at the intersection of Physics and Information Theory,” Matt Robinson, Rampart’s founder and Chief Scientist told me. “We are poised for total leadership from the physical layer up, for defining the frontier of communication technology, and for being the global standard for all network engineering,

In this new version of the world where standard wireless and standard encryption is not enough, you can envision Rampart being useful for all wireless communications; perhaps starting with the most sensitive (as they have with the military) and working backwards from there.

One belief driving Rampart’s approach to the business is that “we’ve largely reached the end of easy wins. The low-hanging fruit is picked over, and even most of the medium-sized problems are quickly being solved. What’s left are the ‘hard’ problems - the ones that won’t yield to patchwork solutions or half-measures.”

“If we want the next great leap in history, we have to stop chasing shortcuts and start reinvesting in the hard, slow, substantive work of fundamental engineering, science, and mathematics. Real progress requires real thought, not just harder sprints.”

“We have a fundamentally novel and powerful approach to the next generation of wireless communications,” Rampart’s VP of Growth, Dzu Do, said. “This is the culmination of years of research and development applied towards solving real world problems in both the defense and commercial industries.”

The growth

If the tech startup world is a stereotypical high school classroom, Rampart is not one of the kids raising their hand every two seconds. They are not one of the kids playing video games in the corner. They are not one of the kids scrolling Instagram or debating politics with the professor. No, Rampart feels a little more like the nerdy smart kid sitting quietly in the back, taking notes.

The company doesn’t try much to market to the general public. Their official X account has 0 posts. There are no splashy Series A announcements or big VC names. Rampart has instead raised money behind the scenes through a combination of grants and funding from angel investors; a path more common among defense companies than ‘hot’ software startups.

Rampart’s distribution so far has been mostly through defense ecosystem relationships: SBIR contracts, partnerships with companies like Lockheed Martin, Comtech, and Thales Defense and Security, and direct Pentagon engagement. The company hosted former Under Secretary of Defense Heidi Shyu (the third-highest-ranking civilian in the DoD), Dr. Tom Rondeau (Principal Director for FutureG) and Alex Lovett (recently retired Director of Prototypes and Experiments for Mission Capabilities) at their facility, which is a strong sign of DoD interest in Rampart at the highest levels.

The culture and team

There is one stereotype of startup culture that focuses on MVPs, on pushing to production as fast as possible, on spending as little on development as possible. That is not Rampart.

I talked to Jay Patel, Rampart’s EVP of Product and Engineering, who previously worked on some of the first mobile phones at Motorola and led product teams at Avaya and Vonage. “We invest heavily in true research,” he said. “This is rare to see as many startups focus more on… [just] trying to get a product to market.”

The R&D-first ethos is so strong that Rampart’s CEO, Omar Javaid, thinks it might even be controversial. “We have some of the best mathematicians and physicists on the planet! It’s unpopular because it’s a heavy, long-term investment and most people think short term.”

The team “is unique. They have a deep technical background, were successful academics, and - crucially - applied their research interests at the NSA and in the field with actual operators.”

(The R&D claim isn’t just fluff; Rampart holds 130+ patents!)

You might reasonably assume that working at Rampart alongside some of the ‘best mathematicians and physicists on the planet’ is probably not an easy job. That would be correct. “It’s really challenging to work on fundamental technology where the timing is uncertain. Some people are attracted to uncertainty but find that environment difficult to actually work in.”

The biggest driver of talent to Rampart is the novelty of the technology. It is an opportunity for PhDs to work on revolutionary technology. Also, a common theme amongst all employees is the fundamental goal to keep our servicemembers safe when they are working in contested areas.

Rampart is based in Linthicum Heights, Maryland, and the team works in-person.

Should you join Rampart?

The word “inventor” often sounds like an old-fashioned thing that does not apply to the work people do today. It can feel like lots of tech startups are just building slightly better versions of things that already exist, step-changes that might be useful but are not very exciting.

Rampart is inventing things in the old-school sense of the word. They have “made advances that were previously considered impossible.” Their goal is to become the gravity well for a new era of wireless. If you are a smart person interested in working on a big problem, this probably ranks within the top decile (or higher) or projects you might want to consider.

Matt Robinson, Rampart’s founder, told me that he wants the company to be the place that people “send their best PhD students and professors on sabbatical because this is where the most exciting and groundbreaking work happens.”

The first value Rampart lists on their general job application is: “Privacy is a right.” If you are interested in building a more secure, more private version of wireless for the future, and if you would like to do that alongside some extremely smart and hardworking people, you may want to consider joining Rampart. You can see their open roles here.

Thanks to Rampart for supporting Next Play and making this Spotlight possible

Y Combinator famously offers two big pieces of advice to startups:

1. Write code.

2. Talk to users.

I’d argue that this is good practice for all companies; making a better product and understanding your users is generally upstream of earning more profit.

It also feels like we sort of have the first item on that list covered. Or at least we’re getting close. The comparatively underrated part seems to be that second piece of YC advice: talk to users.

Historically, if you needed to run a study on customer, product, or market, there was no great way to approach it. Research has long been one of the slowest, most expensive, and most unreliable processes in business.

The company we are writing about today, Listen Labs, is transforming that status quo by building a new kind of customer research platform that is faster and more cost-effective without compromising on quality.

The pitch is simple: Listen accelerates your research by designing studies, sourcing participants, conducting interviews, analyzing results, and making them easily presentable. Traditionally, this would be an expensive endeavor that takes months and requires working with multiple legacy solutions like Qualtrics, Gartner, or AlphaSights. Using a carefully trained AI researcher infused with best practices, Listen shortens the research to results loop to mere days.

Listen Labs announced a $69M Series B in January and already counts hundreds of massive enterprises as customers. So we chatted with their team to figure out how it all works: what is the plan for the business? Why do they believe it will win instead of competitors? What makes the culture unique? And should you join?

The product

Market research has been stagnant for decades. “Qualtrics has been running the same legacy-style surveys for years and is still worth $12B,” Alfred Wahlforss, CEO of Listen Labs, told me when asked about the market. “A single Fortune 500 company might spend $10M on Qualtrics and $100M on research agencies.”

Usually, “even on an expedited timeline you’re talking 4 to 6 weeks from true question to insight and action,” said Brian Davia, Head of Consumer and Business Insights at Sweetgreen, a Listen customer.

With the painstaking manual work done with AI, Listen Labs says you can go from a question to a useful answer in as fast as a few hours.

Here is one example of how you might interact with the product. Say you want to know which of two ad concepts resonates better with Gen Z women in tech. You tell that to Listen and their AI turns your plain English question into an effective interview plan that you review and approve. Then Listen sources young women in tech from its pool of over 30M people (or, your own user database if you prefer) and dispatches an AI interviewer to talk to them, simultaneously, in real-time video or audio conversations. Just like a human would, with thoughtful follow-up questions and clarifications.

A few hours later, you have an in-depth analysis of the findings, all traced back to the exact timestamps or transcripts, and reports ready to explore and share.

Ten years ago there was no version of the world where this would work. The big speed bottlenecks (like study creation, interviewing, and reporting) had ceilings. You couldn’t just go out and talk to all of your customers all at once. Now you can, in any language, at any hour, and in any medium, seemingly without compromising on quality, because today’s AI models are finally robust enough to be trained in the right way to ensure quality.

Listen’s traction has been encouraging. Customers range across startups and Fortune 100 companies like Microsoft, Alphabet, Robinhood, P&G, Nestle, NBCUniversal, and Anthropic. Using insights from Listen, Sweetgreen launched its popular Max Protein Bowl, Chubbies iterated on its kid’s swimwear line, and others ran studies on ads for the Super Bowl and Olympics.

In the nine months since they launched, Listen grew annualized revenue 15x, crossing eight figures. They have conducted over one million user interviews and raised $100 million from Sequoia, Ribbit Capital, Conviction, and others—including their $69M Series B announced in January. (Fun fact: Bryan Schreier, the Sequoia partner who led the deal on their side, was an early investor in Qualtrics.)

“Most of our growth now is inbound and word-of-mouth,” Alfred said. Listen’s AEs are back-to-back from 9AM to 9PM. “We’ve just started to build up our go-to-market machine.”

This is an impressive start. But how does Listen Labs win in the long run?

The strategy

Here’s how Alfred, the CEO, laid out the strategy:

“Phase 1: Displace legacy research vendors by providing a faster, better, cheaper way to provide customer research. Then democratize customer research by making it possible for companies to do it in places where they couldn’t before. Phase 2: Expand the platform to act on the insights in addition to collecting them. Phase 3: Become the de facto infrastructure for understanding people across all organizations, bringing forth a world that works in the way people want.

One piece that might help pull this off is a moat that most other AI companies do not have: Listen’s pre-qualified respondent pool of 30M people. This is the kind of thing that takes years and a lot of skill to build; you don’t vibe code your way to a pool of 30 million people in a weekend.

It’s partly this real, almost literal moat that sets Listen apart “from many other AI companies that are relatively thin AI wrappers,” Tobias Schindler, a founding engineer, told me. “We do the end-to-end user research process… That big AI labs would not get into.”

So Listen Labs may be somewhat insulated from one of the big fears around AI: that some upstart competitor could drink your milkshake simply by being clever, or that a company like Anthropic could trivially build their own in-house version and put you out of business. Listen has built a big, physical wall.

The early-mover advantage compounds in this space, too. “Every interview builds reputation scores on participants. We know who gives great answers. Every study makes the next one better,” Alfred said. There’s a clear flywheel for improving the product.

A final piece worth considering is the idea that Listen Labs’ success might create a Jevons paradox, where increasing efficiency increases demand. Companies don’t do as much research as they’d like to these days because it’s expensive, slow, and unreliable.

If research could become faster, cheaper, and more effective, though… The market could grow larger still. And that’d be a milkshake worth drinking.

The culture and team

Google had ping pong and foosball. Cursor has the whole “shoes off” thing. Walk into Listen Labs’ SF and NY offices and you will witness fika: a Swedish tradition where you’re meant to relax while enjoying coffee and pastries. Every Friday, the team gets together to welcome new team members, celebrate people who exhibited any of their six values (iterate to greatness, love the details, make it happen), and hear updates across sales, growth, and product. It’s also a time for Alfred and Florian to answer employee questions (they encourage “spicy” questions) that they solicit from the Listen platform.

Fika is the only big company meeting that Listen Labs holds. “Otherwise, meetings are almost never had,” someone told me. Engineers reported one to two meetings a week or less.

There’s another unusual internal communications norm: employees are encouraged to send DMs to one another in a channel called #private-dms. The thinking behind this is to create more transparency, better collaboration, and more accountability. Because these conversations are open and public, it lets others chime in and add context or support where they otherwise wouldn’t be able to in a private DM. This sounds a little insane until you try it, I was told, at which point you realize it helps people move faster and stay up to speed on what’s happening.

Oh, and they take their shoes off in the office, too. For good measure.

The obvious follow-up: “Who will I be enjoying my coffee and pastries with?”

Co-founders Alfred Wahlforss and Florian Juengermann met at Harvard. Florian is a German national champion in competitive programming and has worked at Tesla Autopilot, reporting to Elon; Alfred previously founded a YC-backed healthcare startup.

The broader team has the resumes you might expect (ex-Rippling, ex-Affirm, ex-Twitter, and so on). The team also counts three IOI medalists. When engineer Krish Mehta started on Labor Day, he expected to be alone setting up his laptop. Instead the whole engineering team was in, and Florian was remotely provisioning him access to every system from a trans-Atlantic flight. By day three, Krish was working on major infrastructure refactor and by day five, pitching new platform features.

Alfred and Florian know all too well that to keep accelerating their hockey stick growth, they needed to keep attracting top talent. As a startup, competing with AI labs and Meta on $100 million offer packages is a non-starter, so they’ve found creative ways to stand out. In March, they threw a Berghain-themed techno party to celebrate their Series B, hosted at a multi-story condo in SF, complete with a discerning bouncer. They invited customers, potential candidates, and friends of Listen employees. The goal was primarily to have fun together, but also recruit customers and top talent.

In 2025, they bought a billboard in SF and plastered a challenging engineering puzzle on it with no logo or text. The posts on X went viral, reaching over 3 million impressions. SF Standard explained the challenge: “The test asks participants to design an algorithmic bouncer for the legendary German techno nightclub Berghain…The player whose algorithm rejects the least number of people” won an all-expenses-paid trip to Berlin. 1,000 people attempted the puzzle, 60 solved it, and 1 engineer went to Berlin, on Listen’s dime. It was another way for Listen to build its employer brand and filter for top engineers in an unorthodox way.

Should you join Listen Labs?

Sometimes I find it’s easiest to decide whether you should join by listening to why other people at the company joined. So I asked Erik Bartlett, one of Listen Lab’s founding engineers (who was also one of the first few hundred people to work at Affirm).

“I was about to turn 30, and felt like I might as well take some big strategic swings by going very early somewhere; I could always go back to big tech in a few years if things didn’t work out. A longtime friend who was part of the talent team at Pear VC introduced me to some of his portfolio companies. Listen (then Merlin AI) was the first one I chatted with.”

Listen Labs is later-stage than when Erik joined—they are Series B—but still early enough for you to get meaningful equity and mold something life-changing if things work out.

You might also be a fit if you’re the sort of person who will do anything that helps the company, not just a narrowly defined list of responsibilities. “If you only want to work on your features and don’t debug customer issues, it’s probably not a great fit,” a founding engineer said.

Ria Shah, a growth lead, said Listen optimizes for “moving faster over doing things the ‘right way’. We’re a culture of being very quick to execute. The pace demands your best work.”

Listen Labs is currently hiring across most departments. If you want to get creative with your application, you absolutely can. “One of our recent hires shared ten really amazing ideas, which showed he spent a lot of time thinking about our space and was passionate about joining the team,” Alfred said.

You can read about some of our ideas for getting hired here, if you’d like to join what would be—if successful—a startup that changes how companies make their most important decisions.

Thanks to Listen Labs for supporting Next Play and making this Spotlight possible.

There’s a temptation to believe that just because something can exist, it does exist.

Humans went to the Moon (and are going again). We landed robots on Mars. We have quantum computers. We built the Large Hadron Collider. Inventing miraculous technology is our default.

Surely, then, we must have solved one of the most important things in the world—sleep?

But we haven’t. Or, at least before Eight Sleep, we hadn’t. Sleep is one of the highest-leverage pillars of health, and yet as Eight Sleep’s co-founder and CEO Matteo Franceschetti points out, most of us spend a third of our lives “on a dumb piece of foam.”

Turns out, there was a lot of space to innovate. In the 11 years since their 2015 Indiegogo Launch, Eight Sleep has built one of the most impactful sleep products in the world. They are now a unicorn, sell in 35 countries, and are the sleep system of choice for everyday people to professional athletes (like Charles Leclerc) and tech execs (like Mark Zuckerberg).

Their success so far is impressive. But the vision, Eight Sleep’s co-founder Max Bassi told me, is even bigger than sleep: “Eight Sleep will be the health company of the future.”

I was curious: what does that mean, exactly, and how is Eight Sleep going to get there? What is the product like today? What is the long-term vision? And who should consider joining?

The product

The way I’ve described Eight Sleep so far might imply that their core product is a mattress. But it isn’t. Eight Sleep’s main product, the Pod, is basically everything but the mattress. It’s a sleep system that works with your existing bed to improve your sleep.

Building the Pod meant solving problems no one had solved before: how do you track biometric data accurately without skin contact or a wearable? How do you adjust temperature independently on two sides of a bed, in real time, without waking anyone up? How do you do all of this without changing the comfort of a mattress? The result is a system of highly technical components that works so invisibly most people just call it “the thing that changed my sleep.”

The nice thing is that it actually works. Eight Sleep has conducted over 75 clinical studies to date with their internal research team. Their most recent peer-reviewed research shows that sleeping on the Pod clinically improves deep sleep, REM sleep, resting heart rate, and heart rate variability. And if any of your friends or coworkers are Eight Sleep customers you undoubtedly know about it; not because you asked but because they, unsolicited, will have told you all about how well they sleep. There are few products in the tech startup world that receive customer feedback with such conviction.

It helps to think of the product’s benefits in two buckets: (1) sleep and (2) health more broadly.

The former is the reason most people spend so much money on the Eight Sleep Pod. They want to sleep better. People want this because it is obviously good (sleeping is nice) and also because they know it will have other downstream benefits to their life and health.

The latter—health—is where things get even more interesting. What becomes possible when you sleep on a device that can track important health metrics, like HRV, respiratory rate, sleep stages, and more? What becomes possible when you have a clear picture of these metrics every single night for years on end?

These are the questions that Eight Sleep’s broader vision is aiming to answer.

The strategy

Eight Sleep is already cash flow positive. People love their product and pay plenty of money for it. If they wanted, they could stop at The Pod. The company could survive, even thrive, without another world-changing goal. Better sleep could be world-changing enough.

But the Eight Sleep vision is broader than a mattress upgrade.

“Eight Sleep will be the health company of the future,” Max Bassi, co-founder and CTO, told me. “Today we optimize your sleep. In 20 years, we cover you 24/7 - predicting changes in your body and health, and taking action automatically to help you achieve your goals, keep you healthy, and keep you well rested.”

The Pod, Max said, is the easiest entry point to getting the data you need to achieve this goal. People spend (hopefully!) 7 or 8 hours a day in bed, around the same time of day, every single day. They don’t move too much. Their body is relaxed. It is the perfect place and time to track all sorts of important health metrics; numbers that have wide-reaching consequences.

From this foundation, “we build outward - more sensors, more data, more interventions,” Max said. “Adjusting your temperature, changing your sleep schedule, alerting you before a problem becomes serious. The more nights of data we collect, the smarter the system gets, the more we can do for you. That’s the 20-year bet.”

You may have a separate question: how does Eight Sleep beat all of the other companies that want to play in this space? But the answer is more or less the same: the moat for Eight Sleep is the product they sell today. It is owning the bedroom, the best place to start with tracking many of the metrics that are important for preventative health.

The other differentiator may be Eight Sleep’s AI-powered products. One example is Autopilot, an algorithm that learns your sleeping patterns and proactively prevents sleep disruption. Another idea is AI-driven health guidance based on data that only Eight Sleep has.

And perhaps it is this grand vision that helps explain why Eight Sleep has raised $150M over the past year. Not only as a sleep system, but as a preventative health company.

The growth

The first version of Eight Sleep was a mattress pad on IndieGogo that they launched not long before being accepted into Y Combinator. The founders (Matteo, Alexandra, and Max) enabled the campaign at 2AM just to test that everything worked. A few minutes later, an order came in.

“We assume it was one of our test transactions,” Max said. “But then we looked at the shipping address. It was in France. A real person, on the other side of the world, had found us and bought our product.”

The company has done rather well for itself in the 11 years since. People like Mark Zuckerberg and Elon Musk have, organically and unprompted, talked about how they love the product. Athletes have purchased the product on their own, and then come to Eight Sleep wanting to promote it. “We don’t have to convince people to say they like it,” Max said. “They already do.”

Eight Sleep sells in 35 countries. They have hundreds of thousands of customers. They have collected over a billion hours of biometric sleep data. They are cash flow positive. All the kinds of things that most tech companies dream of.

As you might expect, VCs have been happy to fund Eight Sleep along the way. But the past 12 months have been the most dramatic flurry of fundraising in the company’s history. In August of 2025 the company raised a $100M Series D, led by HSG. And in March this year, they raised an additional $50M, this time at a $1.5B valuation, in a strategic round led by Tether Investments.

Why all of this fundraising, why now? I would bet that investor confidence is powered by a couple of things. One of them is financial fundamentals; Eight Sleep has built a winning product in the (notoriously difficult) consumer hardware space, which has done a lot to de-risk the business.

But likely the bigger reason that investors are excited right now is to help Eight Sleep accelerate towards their big goal: becoming the most important health company of the decade.

The team

What’s it like to work at Eight Sleep? It became clear through my conversations that Eight Sleep operates with a focus on discipline, effectiveness, and ownership.

Among others, I talked to Sam Kang, Eight Sleep’s Chief Product Officer who in the past worked on the Tesla Roadster and Sonos speakers. “There are three things I care about in a job,” he said. “The first is impact. The second is learning and solving interesting problems. The last thing is team… It was clear that the team was very passionate and smart, but I didn’t detect any ego.”

Meetings are scarce and the culture is “very written,” Sam said. “People add comments on agendas and then have discussions in meetings. Rarely are meetings just one-way presentations of information. One saying from our CEO captures it well: a good meeting is one that ends early, and a better meeting is one that does not take place.”

The culture is deeply focused on ownership of outcomes and accountability for your work. Vanity metrics, and the people who like them, don’t tend to do well. Nor do people who need a lot of structure or need to be told what to do. It’s hard work and the hours are long sometimes. Importantly, you need to know what your actual job is. In other words: what result am I being hired to produce? If you’d prefer to be handed a to-do list, Eight Sleep is not your company.

“The good is the company is full of extraordinary people, the mission is real and the product has the potential to improve the lives of millions of people,” Sam said. “On the flip side, expectations are high and the intensity is not for everyone.”

One other thing that sets Eight Sleep apart from other companies, David Q. Sun, VP of AI/ML (ex-Apple) told me, is “how relentless we are in asking, ‘What can be better?’ There is very little complacency, even when something is already working well.”

Should you join Eight Sleep?

There was one theme repeated by everyone I talked to: impact.

“You’re working on a physical product,” Max said. “Something you can actually touch, something you and your friends use every day. That’s rare in software. Most engineers spend their careers building things that live behind a screen. Here, your code changes how someone sleeps tonight.”

It’s true that in tech you do not get many opportunities to work on products that physically exist. It’s even rarer that one of those products is making an obvious, tangibly positive impact on people’s lives every day. Better sleep and better health are two clear positives.

You might also consider that now is a unique time to join Eight Sleep, because the company’s ceiling is not simply sleep optimization. Their goal is to become the most impactful health technology company of the decade, and The Pod is a useful foundation to help them do that. Joining now means building towards a vision with a higher ceiling than “sleep optimization” might at first imply.

If you do want to get hired, Eight Sleep’s CTO told me that he’d “be impressed by someone who sends me a project they built on the Eight Sleep API. It shows they understand the product, they can ship, and they care enough to do the work before anyone asked them to. That tells me more than any resume.” This advice likely applies across all departments at the company; show you can be valuable instead of telling.

The Eight Sleep team works both in-person at their San Francisco, New York, Boston and most recently, Milan, Italy offices as well as remote, depending on the position. See their open roles here.

Thanks to Eight Sleep for supporting Next Play and making this Spotlight possible.

It’s a bit crazy how fast the definition of “impressive” has shifted in software development. Twenty years ago it was impressive if somebody built a website, any website. Then Wordpress and website builders came along and a site was less impressive, but building an app was still a feat.

Now, someone could show you a full-fledged software application built for the enterprise and your reaction might be: “meh. seems vibe coded.”

It seems unlikely we have fully adapted to this new world yet. What are the consequences? What happens when the hottest companies in the world are using AI to write almost all of their code? When companies are shipping 2X, 5X, 100X faster? What problems arise?

One obvious side effect: testing becomes a lot more important.

There is more code than ever. Fewer people than ever are writing the code or even understand how the code itself works. This might be a bad thing. Today’s spotlighted company, Momentic, wrote about this problem in their recent Series A announcement:

“[Shipping faster] is incredible… Until you realize you’re shipping bugs faster too.”

Traditional testing (QA) tools were already brittle and flaky. They could barely keep up even when humans were writing all the code at a human pace. Nobody liked QA before, but it’s simply untenable now in a world where AI is doing a lot of the work.

So is building better testing infrastructure? If you ask Wei-Wei Wu, the answer is Momentic (the company he is CEO of). They raised their $15M Series A in November last year and want to become the “definitive verification layer for modern software.”

That sounds nice. But I wanted to dig deeper: what does Momentic’s product actually do, and in what ways is it really new? What is their trajectory? And should you consider joining?

The product

Historically there have been two ways to test software: one way is you write code (often in Playwright or Selenium) that runs automatically and tests things for you. The other way is you test the product manually, clicking through by hand. In practice most teams do some combination of both.

These approaches are brittle. They are the reason few engineers will tell you that QA is their favorite part of the job. A script that says “find the element with ID submit-btn and click it” breaks the moment a developer renames that element. And in the 22 years since Selenium came out—despite attempts at ‘automatic’ solutions—not much has changed. Testing scripts are still full of Xpath, data-testid, and hardcoded attributes that nobody cares to maintain.

The other problem with the old approaches is that they are slow. It takes time to write the tests, it takes time to run the tests, it takes time to maintain the tests. Far too much time for an era where code generation is automatic and a fast feedback loop is the most important thing to make sure you can actually ship a good product.

Momentic is taking a net new approach: you describe what you want and AI agents interpret your instructions against a website, iOS app, or Android app at runtime. No intermediate code, nothing to maintain.

“We don’t generate test code like Playwright or Appium,” Wei-Wei Wu, CEO, told me. “This is how we’re able to achieve incredible speed (<200ms per interaction) and world-class accuracy and reliability. Notion runs hundreds of thousands of tests a day with a 99.99% pass rate.”

The workflow is rather simple to imagine:

• You have an idea of a test you want to run.

• You write down that idea for Momentic.

• Momentic runs the test.

• You get your results and fix what’s broken (if anything).

That is the basic summary. The obvious follow up is: “Well, is it useful?”

Momentic is small, and early, but the results they’ve published seem good: Retool now releases 8X faster. Quora went from 7 hours of daily QA to 30 minutes. Mutiny reduced production incidents by 85% and sped up their release cycles. GPTZero ran tests much faster (and sped up their release cycles, too).

This is an encouraging start. But how do they become a $1B, or $10B, or $100B company?

The strategy

There are a few key bets that Momentic is making as part of their strategy to become the definitive layer for software verification. The outcomes of these bets will likely help determine whether or not this is a company you look back on in 5 years and think “I should have joined!”

The first bet is the one we touched on earlier: as AI writes more code and teams want to ship faster, testing is going to become a more valuable (and important) industry than ever before. More testing will be required, and increasingly that testing will be performed on software that no human had a hand in actually writing. A successful verification layer in this world is one that both moves faster and is more effective than existing testing infrastructure.

A second, related bet is Momentic’s novel approach to product. The biggest products in testing right now are code-first. Whether via AI or your own hand, you are writing code to test your product. Momentic is making a bet that the future of testing, of AI-native QA, will involve AI testing software at runtime instead of building brittle test code. The idea being that because this is faster and more effective, it will replace the status quo approaches to testing.

And then there is a third bet: that the QA function as it exists today will cease to exist. Instead, it will transform into something more integrated. “Roles across QA, engineering, and product are blending together,” Wei-Wei told me. “QA as a separate role and function from [engineering and product] is inefficient.”

With Momentic, people owning the product will be the ones owning testing. “Quality should be wholly owned by the person who built it,” Wei-Wei said. “It’s the fastest feedback loop and will only become faster and tighter as better AI models and tooling become available.”

To be clear, Momentic does not believe that QA people or their work are useless. Rather, they believe that today’s workflows are inefficient. That testing should be more effective and faster, that the feedback loops should be tighter. That QA should be better integrated.

The company is so opinionated about this shift that they don’t even sell to non-technical teams. “Our ICP is highly technical engineering teams,” Wei-Wei said. “A lot of startups in the space cater to the non-technical persona, which is a mistake. QA as a role won’t exist soon.”

There is logic to this. The current approach to QA at many companies is like if spellcheck didn’t live inside your word processor but you had to instead email it to Grammarly, and in a few days they would get back to you with the corrections. That would be inefficient. In QA, this kind of thing is the status quo.

The future Momentic is aiming for is one where:

• Testing is more important than ever

• Code-first tools aren’t good or fast enough

• Engineers own testing, not a separate QA function

In this world, companies will need a different sort of software—a new layer—to help them confidently ship as fast as they’d like to. If they win, Momentic will be that new layer.

The growth

In 2024, co-founders Wei-Wei Wu and Jeff An were accepted into Y Combinator’s winter batch of companies. Most of their first customers in this era—when things were small—came from launching often and talking about it on social media. A few were personal connections.

This rather humble start snowballed quickly. Customers started coming through inbound. Via Google Search. Via some paid ads. ChatGPT and Claude started recommending Momentic. Customers started referring to other customers. ARR began to grow.

In March last year, not long after graduating from Y Combinator, Momentic announced their $3.7M seed round. And just 8 months after that, they had raised a $15M Series A led by Standard Capital with participation from Dropbox Ventures (and existing investors like Y Combinator, FCVC, Transpose Platform, and Karman Ventures).

Momentic doesn’t share many other metrics, like ARR, publicly. (Wei-Wei told me that they are growing fast, which is perhaps unsurprising considering their two rounds in 2025.) But the customers they feature (Mutiny, GPTZero, Notion, Retool, and others) are impressive, as are the results those customers are purportedly seeing with Momentic.

There is a long way to go, but this feels like as good a start as any.

The team

Sometimes you read about startups whose founders have nothing to do with the problem they are solving. We’re two ambitious college grads who made a few SaaS prototypes and now want to create a brain scanning company! This can be fine. It can work.

But it is often more confidence-inspiring when the founders and the team do seem to have real experience in the thing they are solving. That is the case with Momentic.

Both Wei-Wei Wu (CEO) and Jeff An (CTO) are technical. Wei-Wei was the first employee and founding engineer at Nashi (acquired by Density in 2021) and has been an open-source contributor to Node.js, Storybook, Rollup, and more for years. Jeff built integration test infrastructure at Robinhood, then led enterprise quality at Retool (where he was recruited by Nicole Sgarlato, who is now Momentic’s recruiter).

Now that may sound good, but what would it be like to work at Momentic?

“Our edge is our culture,” Kailen Swain, founding product designer, said. “I’m sure every company says this so it sounds pretty stupid but here’s why: We’ve built a team where we can be completely honest, pointing out flaws and debating ideas because there’s a foundation of deep mutual respect. We win because we’re building with people we actually enjoy being around and who are all-in on the same goal.”

We often hear from startups about how the team trusts each other. But, at Momentic, trust is something that has to be earned. “Wei-Wei is not inherently trustful,” one team member said. “When I first started, he didn’t trust me, and I’ve noticed that he doesn’t trust anyone when they first start. He pushes back a ton in the early days. Once he trusts you, then he lets you run wild.”

I’m told that this early trust-building phase with leadership isn’t intimidating, though. Just the opposite—there is a lot of support as you prove yourself. “I know I’m hard to earn trust from early on,” Wei-Wei said. “So I try to be really explicit about what I’m looking for and give direct feedback fast, so people aren’t left guessing where they stand.”

Like you might expect with a startup of Momentic’s size, working there means you can expect to work on things that are not exactly in your job description. “I spend a lot more time at Momentic thinking about product and business goals [than I have at other jobs],” Henry Haefliger, a software engineer at Momentic, said. Still, though, this is an “engineering-first culture selling to engineers,” he said. “The people at the company are the right ones to build the product.”

The whole team works in-person in San Francisco. The culture is extremely outcome and output-oriented. One of the qualifying attributes for success listed on Momentic’s own career page is simply: “You dislike meetings.”

Should you join Momentic?

For now, Momentic is small. Smaller than most of the companies we spotlight here on Next Play (which tend to be at Series B, C, or beyond). Today there are 14 people on the team. Joining now means you’ll know everyone, probably quite well. You’ll be an important part of the founding team. You’ll help shape the culture, the direction, the roadmap. Being the third or fourth engineer (or marketing hire, or whoever) is different from being the thirtieth.

Perhaps most importantly, I think you need to both agree with and be excited by some of the ideas that (probably) need to be true for Momentic to succeed:

• There is more demand for testing in an AI era

• Current QA tools are frustrating and can’t meet this new demand

• QA as a function separate from engineering is inefficient

• Software is better when engineers own quality

• The best approach to testing is to verify it live

If these ideas ring true for you, the other question is whether Momentic is the right place to go build this. The early customer list—and those customers’ results—would indicate that Momentic is ahead of the pack on truly becoming the new layer for software verification.

“We’re growing fast,” Wei-Wei said, “and some of the best product teams on the planet, like Notion, Quora, Bilt, Webflow, and Retool, use Momentic every single day.”

If you are excited by the trajectory and buy into this vision, there would be no better time to apply to Momentic than right now. You can check out their open roles (all in person!) here.

Thanks to Momentic for supporting Next Play and making this Spotlight possible.

“Hello, IT. Have you tried turning it off and on again?”

This iconic line breached popular culture with the British sitcom IT Crowd in the mid-2000s. But talk to anyone who works in IT and you’ll quickly learn that the show did not invent the line. Rather, it was funny because it was true: IT people have been doing tedious work for decades.

• “How do I reset my password?”

• “Can I get access to [software]?”

• “The internet isn’t working.”

• “I don’t know how to log in to Slack.”

• “Is ChatGPT down?”

• “My new email account isn’t working.”

Requests like these make up a huge number of the tickets an IT team will get. If you need further evidence, read the top comment under the IT Crowd catchphrase video compilation: “After 3 years in IT, this line has resolved 50% of my calls. No joke.”

It’s not that IT people enjoy any of this. A 2017 survey ranked IT as the 10th most boring job on the planet. Nearly half of IT help desk agents quit their jobs every year. Nobody spends four years in college so they can spend the rest of their life resetting passwords.

And come on, it’s 2026! We have technology that can write a functional v1 of an app based on a single-sentence prompt! And yet we still have IT people resetting passwords? It still takes hours of human time to onboard a new hire?

Who’s trying to fix this?

The answer is Serval. Their pitch: the AI platform for employee support, starting with IT work. (But, as we will cover, the vision goes further than that). Serval signed Perplexity, its first customer, in May last year. They announced a $47M Series A in October, and raised a $75M Series B from Sequoia in December. Woah.

I wanted to know: what’s going on here? Why all this traction? And should you join?

The product

You may not spend all your free time researching IT Service Management (ITSM) tools. That might make it easy to underestimate the market. So let me enlighten you.

Have you ever heard of ServiceNow? It’s a software company founded in 2004. Sequoia was one of their early backers. The company grew and went public in 2012. Last year, ServiceNow made ~$13B in revenue. That’s more than Airbnb. More than Dropbox, more than Datadog, more than most of the startups you’ve ever heard about.

ServiceNow is the incumbent ITSM software platform. Yes, the market is huge.

Despite all of this revenue and incumbent success, however, IT workflows remain broken. The existing tools designed to build automations are more trouble than they’re worth. And few of them are AI-native. If you’re an IT person, why spend two weeks in a clunky drag-and-drop automation builder when you could just do the task manually in a few seconds?

Of course, this is not a good long-term plan.

Here is how Serval solves the problem: when someone in IT wants to automate something, they describe it in plain English. Like: reset Okta MFA factors, don’t allow this more than twice in 24 hours, require manager approval first. Serval writes real code from that description, which the IT person reviews and can publish. Then the task is automated for every employee, forever.

With Serval, IT people can write plain-language prompts to automate:

• Onboarding

• Offboarding

• Access requests

• Password resets

• Software license provisioning

• Any request that today is a Slack message to the IT team

A few real-life examples: Perplexity uses the platform to automate all of their onboarding and 50% of IT requests. Together AI automates 95% of just-in-time (meaning, time-sensitive) access requests. Mercor automates contractor onboarding. (These are just highlights; each of these customers automates ~100s of things with Serval.)

The product works either as a standalone ITSM or as a layer on top of existing software (like ServiceNow), a smart play since lots of legacy software is, for now, deeply embedded.

The strategy

Remember how I said the ITSM market is massive? How ServiceNow makes more money than most of the startups you have ever heard about? It’s true. But it’s not enough for Serval.

“We’re not limited to the ITSM or enterprise service management software space,” co-founder and CEO Jake Stauch said in an interview. “Yes, there are tens of billions spent on the software, but there are hundreds of billions spent on IT support and internal employee support… Our mission here is to build the AI platform for employee support.”

It’s not every startup that—even in its early stages—brushes past a tens-of-billions dollar market for something bigger. So Serval is aiming high, which is encouraging. But what’s the plan to get there? How do you actually become the AI platform for all employee support?

You start by solving one set of important problems, in Serval’s case IT. They have done this well and continue to do it well. Once you’re embedded within IT, you expand horizontally across the organization. Sequoia, which invested in ServiceNow decades ago and also led Serval’s Series B, said that “the last time we heard [customer feedback] that supports the thesis of [an] IT system of record empowering horizontal enterprise automation was 16 years ago, when we partnered with ServiceNow. That is why we were so eager to partner with Serval.”

There are clear signs that the AI automation market for IT is hot. Last year, two big acquisitions happened within a month of each other: ServiceNow acquired Moveworks and Automation Anywhere acquired Aisera. Both Moveworks and Aisera build AI agents for automation and could probably have been considered Serval’s two biggest competitors.

Even though their website copy remains IT-focused for now, Serval noted in their Series B announcement that their product is being “rapidly deployed beyond the IT department into HR, legal, and finance.” This is happening naturally; companies just recognize that Serval is useful and want to use it to automate anything they can.

Serval is still early. It would perhaps be too ambitious to make predictions about exactly where this strategy will land them, but one thing is clear to me: the early trajectory is strong.

The growth

The story surrounding Serval over the past few months is the kind of thing that most new startup founders fantasize about. In April of 2025, co-founders Jake Stauch and Alex McLeod still did not have a real customer. “Should we do something different? Should we make some kind of radical change to the business?” Jake remembers himself saying.

Weeks later, in May, they signed their first customer. And it was a big one: Perplexity.

Momentum built fast. June and July produced more customers and contracts, each bigger than the last. In October of 2025, Serval announced their $47M Series A led by Redpoint Ventures with participation from First Round, General Catalyst, and many others.

And then came the home run: on the day of the Series A announcement, a partner at Sequoia texted Jake one sentence: “come to our office.” Jake was in Orlando and couldn’t fly out, so Sequoia came to him, and it wasn’t long before they had a term sheet. In December, Serval announced a $75M Series B led by Sequoia at a $1B valuation.

Less than a year ago, Jake and Alex were in the same position many startup founders find themselves in: is this even going to work? Now their company is a unicorn.

If you are a skeptical reader (perhaps for good reason!) you may start to doubt this story. You may wonder if this is all hype; if investors just have FOMO. But it seems like this investor excitement for Serval is based on actual facts:

• Serval grew revenue 500% between August and their Series B.

• They automate ~50%+ of IT tickets across their current customers.

• Customers are—without being prompted—deploying Serval across the whole org.

It’s true that not all the numbers are there. We don’t know their ARR. We don’t know about any sort of plan for profitability. We don’t know their churn, and so on. Things could certainly still go wrong. But the start is promising and, with $127M total raised at a $1B valuation, it seems like Serval has plenty of fuel to see just how far they can go.

The team

Serval’s story—and some of their founding team—starts at Verkada.

Co-founders Alex and Jake, along with founding engineer Sebastien deGroot, worked together at Verkada doing all sorts of automation (among other things). “When Jake and Alex left,” Sebastien said, “the cultural shift was obvious. We stayed in touch, and about six months later I visited them in a small WeWork to see what they were building. It already felt like a novel product with real potential.” Serval’s COO, Tatiana Birgisson, did not work at Verkada but previously led growth at Rippling. It’s certainly a confidence-inspiring lineup.

When I talked with the rest of the team (who seem equally impressive) about work at Serval, it quickly became obvious that this is an opinionated company where you can expect to work really hard, even by startup standards.

For one thing, there are no product managers, Sebastien told me, with Serval opting for the Palantir-esque model of hiring forward-deployed engineers that build product through direct customer feedback. This is part of why, as engineer Derek Zheng said, “customers can expect turn-arounds on feature feedback within the same week and, often, the same day.”

Of course, working at a place where customers can expect turnarounds on feedback within the same day is an intense experience. Spencer Lang, an FDE at Serval, told me that his first week he was “immediately thrown into customer calls and shipped core features like emails from our workflow builder and our Crowdstrike integration.” The second week of work at Serval, he said, the company became a unicorn.

Practically-speaking, you can expect to work long days and—according to a few people I talked to—sometimes weekends. “It is NOT a traditional 9-5,” Helen Taylor, a mid-market AE, said. “Anyone who needs structure and concrete processes may not enjoy working here.”

Everyone I talked to said they worked for at least 10 hours a day, and the most dramatic answer was pretty succinct: “I work 15 hours a day because I love it.”

This isn’t to say that you will work 15 hours a day and weekends if you join Serval. But I doubt it’s the right place to go if you are in a four-hour work week era of your life. You will work hard here!

Luckily, you will work hard alongside some truly talented people.

Should you join Serval?

“During my last interview,” one engineer told me, “I asked one of our founding engineers why [he joined] Serval. He told me that if he thought he were better than the founders, he would start his own company. This really stuck with me because I thought he was one of the strongest engineers I’d ever met.”

There’s a lot of love at Serval for Jake and Alex and the team they’ve hired so far, and so one reason you might join is because you feel like the engineer in Spencer’s quote: I trust these people to run a great company more than I’d trust myself.

You might also join because enterprise automation (despite sounding like a buzzword) is actually quite important. On the human level you are saving IT people and legal people and HR people from doing tedious manual work. And on the business level you are helping companies pay people to do more valuable, higher-leverage work, which helps them grow faster.

If you are convinced (or semi-convinced) that Serval is going to achieve even a fraction of its grand vision, then now is the time to join. There is no better time to get equity in Serval than now—a phrase that will be true for as long as the company continues doing well—and they are early enough that an exit or liquidity could result in life-changing money for you. (This is often less true for joining a startup at ~Series D and beyond unless something crazy happens.)

Finally, I think you might join so that you can help realize a new version of that quote:

“Hello, IT. Don’t worry. Serval will take care of it.”

You can browse Serval’s open roles here.

Thanks to Serval for supporting Next Play and making this Spotlight possible.

Here is a T.V. show pitch for you: we reboot Sherlock. Bring Benedict Cumberbatch and Martin Freeman back. Rebuild the old sets, hire the old writers. It’s detective time.

There is one catch, however. The first eight episodes are just Sherlock Holmes sitting behind a desk doing paperwork. That’s it. Each episode is one single shot; the camera doesn’t move.

Then, in the last couple of episodes, Sherlock does some real deducing and solves the case.

What do you think? Riveting, no? Maybe I’ll send a cold email to HBO.

Of course, this T.V. show would be bad. You wouldn’t watch it. If I got HBO to make this show, it would be ridiculed as the biggest waste of money since the final season of Game of Thrones. Sherlock Holmes is a detective, and he’s supposed to be doing smart detective things—not tedious manual paperwork that anyone could do. I am sure you agree.

Now what if I told you there is a real version of this T.V. show happening every day?

It’s not Sherlock Holmes. It’s financial crime investigations. Every day, thousands of smart analysts working at financial institutions from Robinhood to J.P. Morgan wake up, open their computers, and spend the majority of their time doing repetitive pattern-matching. Only a small fraction of their day is used for actual deduction about whether crime has happened or not.

Financial crime is a huge deal and it’s seriously underrated relative to how much of a role it plays at financial institutions. In a 2024 letter to the FDIC, the popular fintech startup Mercury said that 1 in 4 people at Mercury work in compliance (much of which is about financial crime). And that’s not out of the ordinary; North American companies spend $60B+ per year on financial crime compliance.

The world may be a much better place if the following things happened:

1. Financial crime investigations become far more effective (less fraud)

2. Financial crime investigations become far more efficient (humans focus on better work)

3. Financial crime investigations become far more flexible (labor cost scales to workload)

If you pitched this thesis to Will Lawrence, founder and CEO of Bretton AI, he might tell you that the solution is simple: have every financial institution use Bretton AI as their infrastructure layer for financial crime investigations.

Last month, Bretton AI (formerly Greenlite AI) raised a $75M Series B to help financial institutions automate financial crime investigations with AI agents. Below I’ll share more about how it works, what the long-term play is, and what kinds of people should consider joining.

The product

Startup founders often enjoy describing their companies as the “AWS of [X].” I have heard about AWS for the internet, AWS for healthcare, AWS for AI. Perhaps we are not too far off from a future where some founder pitches the AWS for AWS. It wouldn’t surprise me.

And so when founder and CEO Will Lawrence told me that Bretton AI is like “AWS, but for compliance workforces,” I raised my eyebrows. But as I dug further into what Bretton AI does and what the world looks like if they succeed, the comparison felt apt.

The product’s outcome is simple: it automates the bulk of financial crime investigation work for financial institutions. Existing customers include fintechs like Mercury and Robinhood to larger, FDIC-regulated banks; really any company that has to deal with financial crime compliance.

So how does it actually work?

When Bretton AI graduated Y Combinator in the summer of 2023, it had a narrower scope: it was an AI tool to automate sanctions screening reviews. Sanctions screening is what you have to do when a customer’s name matches against a government watchlist. You, the person doing the review, have to check and see if the alert is correct or not.

Will Lawrence and his team knew they wanted to tackle all of financial crime and compliance, but sanctions screening was the place to start. It is fairly straightforward (relative to some other kinds of compliance work), but it is also important and requires a lot of manual labor. And the competency of AI models in 2023 was enough to automate this effectively.

Fast forward to 2026 and Bretton AI has hit an inflection point. “First,” Will told me, “LLMs crossed the threshold where they can reliably handle complex, judgement-intensive compliance work. Second, regulators are increasing pressure… So compliance teams are drowning. Third, the talent market for compliance analysts is brutally tight and expensive. Banks are spending billions on compliance staffing and still falling short.”

This is the perfect storm for Bretton AI’s current product, which operates as a more comprehensive suite of investigation-ready AI agents that can tackle almost all portions of financial crime compliance. Two of the biggest things it does today are AML (anti-money laundering) and conducting enhanced due diligence (like KYC and KYB), both of which today require tons of tedious, repetitive manual work.

With Bretton AI, a company’s compliance analysts can focus on the Sherlock Holmes-y part of the job, not the database searching/copy-and-pasting/repetitive pattern-matching part.

The AWS comparison slots in nicely via the mechanics of how companies use and pay for Bretton AI. Instead of flexing analysts up and down with demand (complicated), companies can scale capacity up and down immediately (easier). Like how you’d scale compute.

And, to be clear, the product works. In their fundraising announcement last month, Bretton AI shared that:

• A $15B+ institution cut BPO spend by $5.35M in Year 1.

• A Fortune 500 reduced time to onboard institutional clients by 50%.

• An FDIC-regulated bank reduced time to originate loans by 90%.

This is a promising trajectory. So where is it all headed?

The strategy

If Bretton AI works out, then in twenty years “it becomes the infrastructure layer for staffing in regulated industries,” Will Lawrence said. Bretton AI is capitalizing on a theory that many startups (and analysts, and journalists, and others) espouse: regulated industries are filled with a ton of tedious work humans shouldn’t have to do.

Nobody wants to watch the ‘Sherlock Holmes does repetitive paperwork’ show the same way nobody wants to be the Sherlock Holmes doing repetitive paperwork. Financial crime is interesting and important, but people want to spend their time doing the tasks that really require a human; the complex tasks and the deduction. The Sherlock stuff.

But Bretton AI is starting with arguably the most difficult regulated industry first. Financial crime compliance is so important to financial institutions—and the consequences so large—that convincing an FDIC-regulated bank to use your product is different from selling to a small startup.

That’s why Bretton AI started with fintechs. Mercury, Robinhood, Gusto, and many others; these startups were willing to take a chance, and luckily, the tools worked. Now the question is: how does Bretton AI move upmarket?

Two things stand out. First is what Bretton AI calls Trust Infrastructure, a governance layer that embeds specific federal banking guidelines (like OCC and NYDFS requirements) into every agent, with full audit trails and explainable outputs. This is great because banking execs don’t need convincing that AI could help, what they want is proof it won’t screw things up in front of a regulator. Bretton AI’s Trust Infrastructure goes a long way towards doing that.

Second, Bretton AI is pushing into more complex investigations, and they are doing so faster than their competitors. Just last week they announced Horizon, a product designed to tackle what’s called L2 work: longer, judgement-intensive investigations that today take hours. “Our competitors have not cracked this,” one employee told me. “We want AI to not only automate low-risk work but also to build the best UI possible for [the complex stuff].” If they pull it off, that’s a meaningful moat.

The growth

It took 2 years between Bretton AI’s acceptance into Y Combinator and their $15M Series A. And it took just 9 months between that Series A and last month’s $75M Series B, led by Sapphire Ventures with participation from Greylock, Y Combinator, and others.

It’s not hard to see why Bretton AI is having success with VCs. Their average contract size has increased to $201k, up from $85k in May last year and $25k at their seed round. The total market cap of customers is now over $1T. Customers are no longer other startups but also include OCC-, FDIC-, and Federal Reserve-regulated banks. Bretton AI’s AI agents have now completed more than 1.2 million L1 and L2 financial crime investigations.

There are some people who say that AI is a bubble. These people may point to the less-exciting stats about AI products, like the MIT report last year that claimed 95% of AI pilots were failing. Perhaps some of these claims are true; perhaps there are a lot of AI tools that promise a lot but can’t deliver. Or tools that end up being a distraction, costing more than the value they deliver.

All the hype (and the backlash to the hype) makes it difficult to see through the mist and figure out which AI companies are actually helping their customers make more revenue. But Bretton AI’s numbers are about as positive as any that their product is delivering actual value.

And if Bretton AI ends up realizing its vision to become the infrastructure layer for all compliance in regulated industries—then the company has plenty more room to grow.

The team

I sat down with the team (which includes ex-Uber, Palantir, SpaceX, Stripe, and Airbnb employees) to learn more about their culture and who might fit in. What’s the vibe at Bretton AI? How does work get done?

Well, if you apply, you can probably expect to work in-person (though there are some limited remote roles). “You don’t realize how much gets lost in async until you’re back in person full time,” Nitya Mummalaneni, AI ops specialist at Bretton AI, said. “Being together every day means conversations don’t get lost in Slack threads, decisions are made faster, and connections feel more authentic. It’s life-changing for sure.”

Bingqing Wu, a product engineering lead, said she was excited to join because things are moving quickly. “I was especially struck by how much people are encouraged to think creatively and tackle challenges from different angles. One thing that might be ‘unpopular’ in the broader tech community is how much we value being practical over hype. The culture here leans toward substance over flash. We focus on solving real customer problems.”

People who work at Bretton AI are expected to “own outcomes,” and working at Bretton AI is not a “chill job,” James Rogers, a product manager at the company, said. But, for the right person, that’s a good thing. James started working at Bretton AI while wrapping up university and “ended up just skipping almost all my classes to work on the product.”

I also asked: who isn’t a fit? What would a smart person who quit Bretton AI say? Someone told me that you might quit because “there’s a lot of ambiguity and not a ton of process.” This is true of a lot of startups (especially earlier-stage) but it is worth reinforcing. Bretton AI is not a calm place to walk in and expect a well-structured, pre-defined list of everything you will be doing this week or this quarter. You’re going to have to try harder than that.

Should you join Bretton AI?

Perhaps Bretton AI isn’t too unrelated from Sherlock Holmes after all: 17 of the 56 original stories by Arthur Conan Doyle feature some kind of financial crime. After all, if you want to run a successful criminal operation at any scale, you probably need to do some illegal money stuff.

So, if you’re looking to be pitched on the mission: Bretton AI helps reduce real-world crime.

If you zoom all the way in, working at Bretton AI is in part about making financial investigations faster and more effective. You may play an important role in reducing the amount of financial crime in the world. You may also help to make compliance work more fair, a useful goal in an industry where discrimination and bias is still somewhat baked-in.

If you zoom all the way out, then working at Bretton AI is—as CEO Will Lawrence told me—about building an ‘AWS for compliance.’ About building a brand new infrastructure layer in one of the most important industries in the world: finance. If the company ends up becoming an embedded infrastructure layer in regulated industries, then joining Bretton AI now means joining a startup that leaves a real footprint on the world for decades to come.

The money could be nice, too. Bretton AI is at Series B, a good stage for a startup to be at if you’re interested in long-term equity (the risk-reward tradeoff is often favorable). They are on a fast upward trajectory. And if the company succeeds with everything it would like to, you would probably walk away with a lot of money. I’m sure you wouldn’t be upset about that.

If you would like to join Bretton AI at their San Francisco office or look for a remote role, you can do so here. And feel free to send Will (or someone else) a cold email. “If you cold email me with a POV on how we should think about a compliance workflow, or you’ve built something that shows you understand the problem, that goes straight to the top of the pile,” Will said.

(You can read some creative ideas for getting hired in this essay.)

Thanks to Bretton AI for supporting Next Play and making this Spotlight possible.

All industries are disrupted, but some industries are more disrupted than others.

You could even say that some industries are too disrupted. Product management. Productivity. Workplace communication. Documentation. Great new companies will probably crop up in these spaces in the future, but they aren’t where I would send the world’s most promising startup founders to go build.

Then there are other industries—unsexy but enormous—that have seen comparatively little disruption. We’ve covered industries like this before: like with our spotlight on Hearth, a company solving revenue for home service contractors, or Fulcrum, software for insurance brokers.

Today, we are covering a company that deliberately chose the hardest version of this thesis. Valon started in mortgage servicing, a $25B+ market encumbered by regulatory complexity, where the casual innovator is quickly deterred by the barriers. Six years of learning the regulatory complexity firsthand has compounded into real scale.

Today, Valon has:

• Built a top 10 mortgage servicer in the US, with over $110B of mortgages serviced on their platform

• 3 of the 10 biggest servicers signed up to deploy ValonOS

• Raised more than $230M from a16z, WestCap, and others

• Team of 200+ talented people

If you have any interest in learning about (and maybe even joining) quietly dominant startups disrupting unsexy but important industries, well, this spotlight is for you. I sat down with the team at Valon to understand how they built such an impressive business and where it is headed. (As a hint: Valon’s mission is to build the AI-native operating system for all of institutional finance, not just mortgages.)

I also asked the questions that you may be especially interested in: What is it like to work at Valon? And who, specifically, should consider applying?

The most common loan in America

You probably don’t need much convincing that mortgages are a big deal. Most Americans own a home, and most homeowners have a mortgage; more than $13T of consumer debt is underpinned by mortgages. And the way that mortgages are distributed and run can have major impacts on both regular people and the broader economy as a whole (see: 2008).

So, in some sense, you may know a lot about mortgages. But there are a couple of things you may not know that you will need to understand if you want to get a grasp on Valon’s business.

The first thing is that lenders are often not the ones who manage or ‘service’ a mortgage. Lenders often hand off the work of actually running a mortgage (payments, compliance, communication, etc.) to a kind of company called a “mortgage servicer.” Valon built their own (creatively named “Valon Mortgage”).

The second thing is that servicing mortgages has become significantly more complicated in recent years. The aftermath of 2008 brought a wave of regulatory crackdowns, which is part of why the cost of servicing a mortgage since then has tripled; we’re talking about a $25B market.

How did those loans get serviced before Valon? “With technology that was invented before the internet was created,” co-founder and COO Linda Du said in an interview. The clunky servicing software and high servicing costs were, naturally, problems being passed straight onto everyday Americans.

This reality, and the mission to fix it, deeply resonates with the people I talked to on the Valon team. “Mortgage servicing is what banking was before banks had mobile apps or mobile check deposits,”. “We’re rebuilding the entire core infrastructure in a way that is aligned with where the world is going: AI-native, automated, and with safety nets.”

If you were living in the pre-Valon world and wanted to build a startup, the facts would start to add up:

• Real problem that affects real people in painful ways

• Untouched industry with stone-age software

• $25B market that powers homeownership in America and continues to grow

What’s not to like?

From friends-and-family loans to $110B in mortgages

Back in 2019, co-founders Andrew Wang, Jon Hsu, and Linda Du saw a lot to like in the mortgage market. The origin story is classic startup lore: Andrew, who had been working on mortgage problems from the other side of the table at Soros, kept running into how messy and inelegant the whole system was. He wanted a better solution to exist, badly enough that he sketched out the idea on the back of a napkin with Brian McGrath, who was an early investor with Jefferies.

And they started with the same question many startup founders find themselves asking: Why can’t we just build a better piece of software to solve this?

What they discovered next was both humbling and, in retrospect, clarifying. A couple of impressive résumés and fancy degrees don’t conjure a perfect piece of software into existence — and the existing mortgage servicing companies, slow-moving by nature, weren’t exactly lining up to pilot something new. But there was a more interesting revelation buried in the difficulty: nobody else had tried to solve this problem not because it was a bad idea, but because it was so extraordinarily hard. The opportunity was sitting right there in plain sight, and the whole industry had collectively shrugged and walked away. That complexity wasn’t a warning sign — it was the moat.

So, instead of wasting time building unproven software nobody would want to adopt, Valon’s founding team asked a different question: What if we just build a mortgage servicing company instead? Most startups pursue a wedge strategy, starting small and expanding outward. Valon intentionally did the reverse.

Things started scrappy. In order to get a license to operate as a mortgage servicer, Andrew said in an interview, you’re required to have a track record with credit agencies. This was something of a paradox, so the team “found which states had exemptions for small loans, then went out and gave loans to a bunch of friends in order to be able to report 100 loans.” All of this just to get licensed.

The business snowballed, adding billions of dollars in loans. And as they grew, Valon built out the software they had envisioned from the start; effective internal software for servicing loans as well as great consumer-facing software for managing those mortgages.

The team calls this the “anti-wedge” approach and it’s the only way to deploy AI into regulated industries. Owning the operating system means every workflow and data trail is structured, inspectable, and programmable, making it the perfect surface for AI.

Fortune, or perhaps just a lot of hard work, has favored Valon. Valon’s mortgage operating company is now one of the top 10 mortgage servicers in the United States and has serviced over $110B in mortgages via its platform.

In some ways, things have come full-circle. Now that Valon has built a proven servicing software solution, they are also able to work on realizing the idea they started with: licensing their software to other companies. The broader vision is even bigger: “Valon is building the AI-native operating system for regulated finance—starting with mortgage servicing.” While the business is currently focused on mortgage-related products and has not publicly launched anything in other areas of regulated finance, this is worth noting; Valon’s ceiling is not the size of the mortgage market. It is much higher than that.

If you were searching for somewhere to work, all of this information is (probably) music to your ears. There is still a pending item, though: What is it actually like to work at Valon?

What it’s actually like to work at Valon

Ask most people at a startup what the culture is like, and they will give you aspirational truisms (we work hard, we move fast). These are not helpful if you want to know what it is actually like to work somewhere.

So what is it like at Valon?

Valon has offices in New York and San Francisco (though there is flexibility for some remote roles). As for the vibe? Understanding the culture is easiest by learning how people actually work. “Even as an intern I worked on meaningful projects with real ownership. The internship project I led generated roughly $1M ARR, and I was treated like any other engineer,” one employee said.“Leadership, engineers, product, biz ops, and analytics are all visible and approachable. Oh, and people bring treats from their travels.” One product manager brings loaves of bread on Friday because the team has to ‘get this bread.’

If you worked at Valon, you would find the kind of fun, intense energy that you may have had “back in college working with friends on a really hard problem late at night, bouncing ideas off of each other, furiously marking up a whiteboard.”

“Your coworkers are deeply bought in. There’s no coasting, no political nonsense. People take real ownership and hold a high bar for themselves and each other.”

Every Tuesday, you could expect laptops-down “family-style dinners.” “The week of Thanksgiving, we did a mini Valon Thanksgiving dinner that was a ton of fun.” But the culture of access extends beyond the dinner table. Andrew holds regular office hours — with a particular focus on AI adoption — opening his calendar to anyone in the company who wants in. It’s a small thing that signals something bigger: even as Valon has scaled, it operates with the accessibility of a much smaller startup. Anyone can get time with the CEO. And when the CEO is literally in the weeds with his team on AI, the whole company moves faster because of it.

This was refreshing to hear from the Valon team—it’s not too often a startup grows like Valon has and is able to retain this kind of atmosphere. “If you think you’re moving fast, know that you can always find ways to move faster.” “Fast > perfect.” Like at most (good) startups, at Valon you are expected to perform at a high level and to put your all into the work. This isn’t some sort of 996 model where everyone’s in the office on Saturdays, but you will be asked to meet high standards.

Valon is also not the kind of startup where teams work on pre-established playbooks; where anything you might do is defined six months in advance. Not long ago, some folks on their recruiting team built custom Monte Carlo simulations for pipeline forecasting and a Chrome extension for resume screening. Not because anyone told them to, but because they saw a problem and solved it. If you work at Valon, you can expect to be proactive and build things because they are useful and generate value—regardless of whether someone told you to do it.

“We aggressively kill meetings that are not useful.” Meetings at Valon are usually designed either for alignment or debate. If a meeting accomplishes neither, it gets cut.

One other theme I heard repeatedly: Valon feels like an asymmetric bet. The downside is genuinely capped (profitable, real product market fit, and a very strong balance sheet), but the upside is massive. Valon has resisted the hype cycle of chasing the latest AI fad and the inflated valuations that come with it. Employees joining now can still capture meaningful upside at multiples that are impossible at companies where growth is already priced in..

Two of the co-founders (Andrew and Linda) come from investing backgrounds, which is partly why they got to see what happens when founders and CEOs are chasing the latest hype cycle instead of thinking long-term. “Andrew and Linda are both very focused on creating long-term value. From my experience, they are better execs than most of the public CEOs I met when trading stocks at Millennium (a top tier hedge fund).”

“I was coming from a company with an unproven business model (and frankly no PMF). It was so refreshing to talk to a founder and company with a clear identification of the 3, 5, and 10 year strategies and numerous back-up plans to make money and win the market.

“It was nice to know I had to focus less on the strategic direction of the company and could pour my focus into the pieces I was personally responsible for.”

Should you consider joining Valon?

Valon is currently hiring for dozens of roles here. (There are two branches of the company—Valon Mortgage and Valon Tech—but if you are reading this you should probably look at Valon Tech.)

I’d start by asking yourself whether you’re the type of person who gets excited by untangling complexity that scares everyone else off.

“The competition is outdated, slow, and stuck on airport-terminal GUIs and we’re building something dramatically better in a massive, unsexy-but-enormous market,” one person said.

Valon told me that people who do well at the company tend to share three qualities:

• Turbo: relentlessly outcome-oriented, with the genuine enthusiasm to go 10 layers deep on a problem

• Anti-hype: motivated by creating actual value rather than chasing shiny things; comfortable playing the long game

• Precise: holding an extremely high bar for technical accuracy in systems where trust is a one-way door

So, should you consider joining Valon?

It may be easiest to answer that question by defining who shouldn’t apply. Don’t apply if you’re joining for a quick payout or a flashy headline; the payoff here will be massive, but the tradeoff is that it won’t be quick. Don’t apply if you need prebuilt systems and clear playbooks. No one here is a cog executing the machine. The job is to find the most valuable thing to do, and then go do it.. Don’t apply if your solutions tend to break under pressure; at Valon, systems have to handle zero downtime, strict compliance and edge cases that would collapse most software.

From a numbers standpoint, Valon is in a sweet spot: growth is inflecting but the company is also mature enough that you have a good chance at liquidity with the equity you get, but not so late that your equity may underperform the market. “The downside is genuinely capped: we’re profitable, we have real product-market fit, a strong balance sheet, and customer demand that’s already contracted or imminently landing. This isn’t a science project or a hope-and-pray startup. But the upside is massive,” Ariel Brito said.

“Valon offers an exceptional opportunity to grow quickly in your career, the company is at a pivotal stage, preparing to scale rapidly across many customers. If you want true hands-on experience in scaling a business, this is the ideal place to learn.”

If you do apply, what should you expect?

“The interview process was rigorous but fair,” a software engineer told me. “I did technical interviews focused on practical coding and system design, and I also had conversations with leadership about the company missions and goals. I really appreciated that people took time to answer my questions and make sure it was the right fit on both sides.”

You can find Valon Tech jobs here. Don’t be scared of reaching out directly if you see somewhere you think you can add a lot of value. If you do, consider reading our guide to getting a job with cold email before you do so to increase the odds your email is a good one.

Thank you to Valon for supporting Next Play and making this spotlight possible.

If you are a person who has been in regular meetings for the past few years, you have probably heard of Granola. And if you have heard of Granola, you may know the general consensus about it: Granola is the best AI notetaking tool for meetings. It transcribes your meetings, writes helpful notes, enhances your own notes… The sort of stuff you would’ve needed a real assistant for.

There is one version of the world where Granola is truly just that: an AI notetaker. If co-founders Chris Pedregal and Sam Stephenson wanted to, they could probably thin down the team to ten people or so, never raise another round, and run Granola as a profitable lifestyle business that owns its niche.

So it may surprise you when I say that Granola’s vision (and the reason they raised a $43M Series B from Nat Friedman, NFDG, and others last year) is far bigger than AI notes. Co-founder Sam told me:

“We want Granola to be the primary interface where knowledge workers do most of their work.”

If you are looking for interesting startups to join, this sentence should probably make Granola significantly more interesting. Now we’re talking about a many-billion dollar market. Now we’re talking about a trajectory that would position Granola as one of the most impactful startups of this decade.

Then again, you probably just know Granola as a notetaker.

So… how? What is the plan here? And should you join to help Granola accomplish this mission?

Don’t underestimate the scribe

Every software engineer knows that modern code editors (IDEs) are spectacular. They can help engineers work far faster and make fewer mistakes. And the AI versions these days could be the best yet.

On their pre-launch 2024 landing page, Granola wondered:

Why is there no great IDE—no great workspace—for knowledge work?

Today’s knowledge work is “fragmented across dozens of different tools,” co-founder Sam Stephenson said. “You’re constantly switching between email, Slack, your CRM, task management, [and so on]. We believe AI agents will eventually handle much of that context-switching and updating for you.”

You may have seen some early attempts at this in the market. Email software like Gmail and Superhuman these days can try (and often fail) to write good emails for you, as one example. Anthropic launched Claude Cowork back in January. Companies are aware that AI could integrate into everyday work.

Granola has had this in mind ever since they began working on their product back in 2023. The key question, though, was how to realize this vision in a way that would be actually useful. How do we make AI feel like a valuable extension of yourself and not a context-less slop machine?

Sam says the answer was simple: “For [this] to work, you need something that understands your context deeply. And meetings are where the most important context lives.”

If you were a medieval ruler with a royal court, you would want your advisors and scribes to be present for everything. You would want them to have the same information you do so that they could help you make good decisions. And you would trust them over a know-it-all who rode in from afar with no information.

Starting as a humble meeting scribe, then, was the smart place for Granola to begin.

The Cambrian explosion

Publicly, all was quiet for the first year that Chris, Sam, and their small team worked on Granola.

Internally, the product was undergoing an iteration process that Jim Fisher (software engineer) describes as a Cambrian explosion approach. The traditional v1, v2, v3 path was ditched for an iterative process in which they would onboard users one-by-one and ship all kinds of different permutations of the product; different features, different styling, different everything. As 2024 approached, they honed in.

And then, when Granola launched in the spring of 2024, they removed 50% of their features. “If you add 50 buttons in there with new features,” Chris Pedregal said, “you kind of kill the golden goose.”

This focus on restraint, taste, and what I would call “actual usefulness” paid off. Word-of-mouth took off and Granola was quickly recognized as the most reliable AI notetaker around. And people stuck around.

“From very early on, we had this retention curve that just flattened out at a really high level. People who did one meeting with Granola kept coming back. We didn’t have to growth hack our way to virality - once people tried it once, they just naturally kept using it for all of their meetings,” Sam said. 70% of people stuck around for a week. 50% of people stuck around for ten.

You know it’s good because an AI notetaker isn’t exactly an impossible product to build. (This isn’t powerful military tech.) I bet you could go spend a few hours with Claude or ChatGPT and vibe code a meeting notetaker this afternoon, and maybe it would even work. As a result there are 100s of legitimate competitors for Granola; the idea feels simple and easy to replicate.

And, yet, Granola is winning. They are winning a lot. This says something about the team’s approach: the restraint, the taste, the knowing how to build and what to build and when.

There’s a long way to go still, and “[the founders] don’t feel like we’re there yet: they want to keep listening to users and keep trying to think about how to make the product better,” Vicky Firth said.

Sam and Chris are “down-to-earth, optimistic, and visionary,” someone on the team said. “They are super transparent - a lot is shared about how they’re feeling, what investors are saying, how the company is doing. They have strong opinions about the product but place high trust in the team.”

Traction? Check. Smart, measured founders? Check. What about the culture?

Merge first, ask questions later

Granola’s offices in London and San Francisco, where everyone works in-person, are not safe spaces for complainers. “There is a huge amount of positivity and very little complaining at Granola,” Vicky told me.

“This covers everything from big product changes all the way to which drinks are in the fridge. There’s a buzz here that you don’t see everywhere.”

There is a “just give it a go” energy at Granola. You’re trusted to do a great job and, if you try something but it doesn’t work (like when Jim, an engineer, deleted half of the prod database), the team is not going to bury you for it but rather solve the problem and move on (and maybe ask you to drink ‘OMG whisky’.)

“There are no pre-code merge reviews,” an engineer said. “You just merge and iterate! Many other companies I’ve worked at would balk at this.” But it works at Granola, and this general philosophy—just do a good job and we’ll iterate as we go—seems to be present on every team.

How long should you expect to work? Everyone I talked to said they worked about 9 or 10 hours per day, but they also stressed that hours can be flexible. “You’ll often find me dashing off at 5pm for another commitment,” someone said. My overall impression is that, while the team at Granola works hard, this is not a startup that will ask you to come in on weekends; they will not force you to stay up until 2AM working on some feature (though I’m sure you could do it if you wanted to).

Shreman Shrestha, who works on various business and growth projects, had three points to sum it up:

“1. Highest density of talent I’ve ever encountered.
2. Founders you meet once and feel they’re ‘different.’
3. Everyone is genuinely lovely, generous with their time and knowledge, and fun to be around.”

Top-shelf talent, great founders, and perhaps a cheeky pint after work. What more could you ask for?

Should you consider joining Granola?

If you have read this spotlight and feel like Granola is the right company and right team to build the primary interface for knowledge work, then the basic filter is: Do you live in London or San Francisco? If you don’t, are you willing and able to go live in London or SF? Perhaps you could get hired as a contractor somewhere else, but the entire full-time team at Granola works in-person.

Beyond that, “if acting with high caution or constantly trying to play devil’s advocate… are your default modes, Granola probably isn’t the place for you.” You should be the sort of person who assumes things can work, approaches your job with optimism, and can take a lot of autonomy.

If you want to increase the chances that your application gets looked at, you could put in a little more effort than usual. “We love to see side projects, or signs to show you care,” co-founder Sam said. (As it happens, we recently wrote an essay with some advice on doing just that.)

Most of the people I talked to described having a one- or two-week work trial in the office. “I felt I was able to have impact quickly,” Vicky, CX lead, said of her work trial. “Thankfully they felt the same!”

A few of the folks I spoke to told me that they had been rejected for jobs at Granola they’d originally applied for, but that the team had then reached out in the future about other roles. Another encouraging sign that—if the company seems interesting—you should let them know how you can add value.

Granola has a long way to go if they want to become the interface where most knowledge work gets done. But, with AI notetaking, they have already won the first game. Why not go help them win the next?

Thank you to Granola for supporting Next Play and making this spotlight possible.

Seven years ago, Brandon Hill’s parents came home with a stack of wholesale catalogues; thick papery booklets listing every product a grocery store could order.

“Wow,” Brandon said, curious, “Is this how they used to place orders back in the day?”

“No,” his father replied. “This is what they do TODAY.”

So began Brandon’s (and then co-founders Tre Kirkman’s and Robert Pinkerton’s) journey towards founding Vori: the startup building a modern operating system for grocery stores. Because although not every grocery store is still placing orders via paper catalogues, existing software is hardly better. “The grocery industry is one of the most under-digitized industries on earth,” Brandon said.

It is a nonobvious place to build. Maybe today is the first time you are reading the phrase ‘grocery store startup’. But the problem is real, the market is massive, and Vori already has encouraging traction:

• Processed $350M+ in payment volume in 2025

• Revenue scaling 5x year-over-year

• Generated $22M+ in incremental sales for independent grocers in 2025

• 577K+ supplier cost changes auto-detected by Vori AI, protecting retail margins

• Launched in 55 new cities in 2025, now operating coast-to-coast

• Customers (incl. Mollie Stone’s Markets) have become investors

(Go watch this video if you want to see a visual recap of Vori’s progress so far.)

Most startups you hear about are helping theoretical people you will never be on theoretical problems you will never experience yourself. Vori, on the other hand, is aiming to transform something that every single person reading this is familiar with: going to the grocery store.

What does the market look like? What is Vori’s plan for success? And should you join now?

Arming the rebels

Running a grocery store is a harder job than you might think. If you are a company like Amazon (which owns Whole Foods) or Walmart, you can smooth over some of that difficulty with billions of dollars that you use to build your own back-office teams and develop your own internal software.

But if you are an independent grocer—not a national behemoth—then you are in a tricky situation. You have to do all the same things Amazon does (ordering, sales, accounting, inventory management, etc.), and you have to compete with the big boys, but you have only a fraction of the resources.

So what do you do? Well, in the past, the best answer might have been: “suffer.”

Growing up, Brandon Hill (Vori’s CEO) watched both of his parents in the industry deal with systems that were built in the ‘90s and probably should have stayed there. If you run a grocery store today, there is no great operating system you can use. There is no Stripe, no Shopify. You are out there on your own.

And being out on your own is not a recipe for success. Especially when, as Brandon told me, “wages have stagnated, margins have compressed, and scale retailers [like Walmart] wield a significant pricing and technology advantage. At the same time, the industry is undergoing a massive intergenerational transfer of ownership: thousands of owners are handing the keys to a next generation that doesn’t want to run a business on paper, fax machines, and tribal knowledge.”

You often hear startup founders talk about ‘underserved industries’ that are ‘ripe for innovation.’ When you see ten different startups in the same industry building same-y products for that ‘underserved’ market, though, you start to wonder: is it really underserved? Or is this just to farm goodwill with investors?

Here with grocery, it feels as though Brandon, Tre, and Robert have encountered a market that is the kind of underserved, prime-for-disruption market that so many startup founders say they are attacking.

And starting with the $350B independent grocery industry is a carefully-calculated decision.

Bear hugs in the aisle

If Vori is successful, co-founder Tre Kirkman told me, the company will become “the clearing house for the global food supply chain. Every basket from Oakland to Auckland will be powered by Vori.”

That’s ambitious. But you have to start somewhere, and Vori started where it made the most sense: by building reordering software for independent grocers. Reordering software because it is one of the more tedious (and important!) things a grocer needs to do; independent grocers because they have fewer embedded systems and lower inertia preventing a switch to new software. (Also because, as it has become clear, the Vori team has a real passion for great independent grocers.)

Over time, Vori has built out their software into a full operating system for grocery stores: POS, payments, back office, price changes, and more; an operating system that has now facilitated more than $350M in payments and generated more than $22M in incremental sales for independent grocers in 2025.

Vori is an AI-native OS; the company is building solutions (actually useful ones!) with AI to let stores run on “full self-driving mode,” Brandon said. This is not a requirement—stores are not forced to give up the reins to AI—but the products are effective. To date, more than 577,000 supplier cost changes have been auto-detected by AI to protect retail margins for grocers. This is perhaps one of the few examples of AI tools in the market today where there is an obvious line between using the AI and increasing profit.

Then there are the intangibles. “Jaime, the dairy clerk at the Market at Edgewood, gave [co-founder] Rob a massive bear hug in the dairy aisle where he gave [Vori] a test drive. ‘I have goose bumps,’ he told us. This is when we realized the potential for life-changing impact,” Brandon said.

“The founders are close-knit and supportive of one another,” Klaire Korver, growth marketing lead, told me. All three are Stanford graduates. Brandon and Tre launched YC-backed social app Greo in 2017; Robert was building rockets at SpaceX and working on Lyft’s first fleet of self-driving cars.

Move fast, don’t break things

I do not think I have talked to a more grocery-obsessed group of people than the team at Vori. This is a group of people that cares deeply about grocery stores, about what they mean, about the value of independent grocery stores, about the problems grocery store owners face daily.

Vori is “customer-centric in a very literal way. Slack is full of customer conversation recaps — install learnings, post-live feedback. These stories directly fuel what we prioritize and what we build next.”

If you can’t empathize with the plight of the independent grocery store, well, I’m not sure if Vori will be a fit for you. You need to love (or be able to love) grocery stores and their importance. They are the customer!

“One of our core values,” Engineering Manager Clinton Blackburn told me, “is don’t mess with the dairy order. You mess that up and you’ve ruined a shopper’s visit to the store and potentially lost their trust. We move fast, but don’t break things because we know that the things we might break can ruin a business and someone’s livelihood. We keep this in mind with everything we do.”

It stood out to me just how emphatic people at Vori were about the no-politics culture. “I’ve worked in tech for 13 years across seed through public companies… And Vori is the least political company that I’ve ever worked at. Everyone’s ideas are valued,” Klaire Korver said.

“I don’t feel that there are any company politics to deal with here,” someone else told me. “Everyone is approachable and motivated and there’s no sense of work being above or below someone’s paygrade.”

This low-drama approach to things means that anyone—even a brand new junior hire—can make a big impact at Vori. “Titles aren’t a big thing here,” someone said. “Flat structure / real ownership. Everyone has meaningful scope, and that’s what motivates people.”

People work the hours they need to; some people told me they worked 8 or 9 per day, while Clinton Blackburn told me he works 12. “Did I mention I need more engineers to spread the workload?”

Should you join Vori?

We’ve established that you should care about grocery stores. But that’s easy enough to do by simply paying more attention the next time you shop. People 150 years ago had none of this.

Why else should you consider joining?

Jesse Lopez, Head of Product Marketing, had thoughts: “Over my career I’ve come to learn that what you do is really only 15-20% of what makes for an enjoyable job. Certainly, you want your work to be engaging and impactful, but in reality, who you work with, and how you work together, contributes way more to your enjoyment of a role. Now that we’re expanding our team, we often talk about carving out the most interesting work for our new joiners. Better to give the exciting new projects to the new hires so they can be easily engaged while they build rapport with their teammates.”

People who do best at Vori tend to be “highly curious, high ownership, customer-obsessed, and low ego. The last one is really important. We like to live by the proverb ‘If you want to go fast go alone, if you want to go far, go together.”

There is also the equity and growth trajectory perspective. Vori is growing fast, and it would not be surprising to see a Series B sooner than later (and then a C not long after that, if all goes well). This is a stage at which you can still get pretty meaningful equity (that could end up being very valuable if Vori ends up succeeding long-term). This is also a stage where you can make a big impact—your individual decisions could help shape the company’s direction—and be part of the reason Vori wins.

Vori is hiring across engineering, sales, marketing, design—just about everything. If the company sounds interesting, give them a closer look. And if you do reach out, you could try being unconventional: “We are down,” Brandon told me. “Shows creative hustle.”

Thank you to Vori for supporting Next Play and making this Spotlight possible.

Here is a seemingly basic question: What allows most modern war technology to work? What allows the drones, the autonomous vehicles, the GPS-guided missiles, the communication systems—and frankly, almost everything useful on a battlefield—to function properly?

The answer is the thing that the startup we are spotlighting today, CX2, is focused on: the electromagnetic spectrum.

Over the past few decades, billions of dollars have been poured into funding new defense startups and building advanced tech that is useful on the battlefield. This is good. Yet it is also true that almost none of these things work without connectivity. Disrupt that and everything stops working. Which is why electronic warfare has quietly become the most important thing to solve if you want to win.

“The electromagnetic spectrum is the base of the 21st century defense tech stack,” Porter Smith, CX2 co-founder and ex-Army commander, said. “If we don’t dominate it, none of the other fancy stuff works.”

CX2 is building the technology the West needs to dominate that spectrum. And I have a hard time imagining a better founding team for the job:

• CEO Nathan Mintz and President Mark Trefgarne are both former founders with successful exits.

• Lee Thompson was a SpaceX team lead for RF engineering.

• Porter Smith was a helicopter pilot in the Army for a decade and invested for a16z before joining.

The strong level of talent and military experience extends beyond the founding team, too: a third of the team at CX2 are veterans.

Better yet—if you are someone looking to make a big impact—CX2 is early. They raised a $31M Series A in 2025 led by Point72 ventures (participation from a16z, 8VC, Pax Venture, and others). They are currently hiring for their relatively small team at their El Segundo office; this is a unique opportunity for people, especially engineers, who want to work on hard problems with real stakes.

Over the rest of this essay I will share a little more from my conversations with the CX2 team. What is the market here? What are they building? And what would it be like to work there?

Invisible warfare

In the summer of 2024, Porter Smith was with troops in Ukraine when he noticed something odd.

“Ukrainian infantry guys take spectrum analyzers [devices to analyze the RF spectrum] with them everywhere,” he said.

Why were the Ukrainian troops doing this? Because ground forces were dealing with Russian jammers; devices that interfere with the RF spectrum. These jammers “affect your ability to maneuver on the battlefield. Because if you have large jammers, you don’t have GPS-guided munitions, you don’t have the ability to fly your own ISR assets, you don’t have the ability to fly your own logistics efforts.”

This obsessive focus on the RF spectrum is not the way war has always worked.

“Historically, [what the Ukrainians were doing] is a very niche electronic warfare task in the United States. And their version of 11 Bravos, standard infantry guys, were doing this.”

If the RF spectrum is upstream of most modern battlefield leverage, well, it is only natural that your adversary will want to disrupt it. This reality means that winning on the invisible, electronic battlefield is now the key to winning on the physical one.

So how would you go about this—practically speaking?

CX2 describes their approach simply: “Hunt the archer, not the arrow.” The thesis goes that the best way to dominate the RF spectrum is to identify enemy emitters (like jammers, or drone pilots) and engage them directly. “We think the 21st century kill chain begins in the RF spectrum,” Porter Smith said.

Hunting the archer

So far, CX2 has launched two products: Vadris and Wraith.

Both of them identify RF signatures on the battlefield. But they have different purposes.

Vadris is a small payload you strap onto an FPV drone that identifies adversary drone operators so you can engage the operator directly—not just their drones. “The counter-UAS market is flooded with drone intercept approaches,” Lee Thompson, co-founder and Head of Engineering, said. “We have flipped that on its head to instead hunt for the drone operator.”

Wraith is an autonomous recon drone that creates a heatmap of the RF spectrum and geolocates hostile emitters (like jammers and radars), even in GPS-denied environments. So you can dominate the spectrum even when connectivity is under attack (which is now the norm). If the 21st century kill chain begins in the RF spectrum, as CX2’s Porter Smith believes, then “Wraith is the spark for that kill chain.”

These two products are effective expressions of the same idea: start treating the electromagnetic spectrum as the battlefield itself. Because you need to, if you want anything else to work.

As you might expect, building these products has required a skilled team with a lot of drive and a lot of experience—you can’t really spend a weekend with ChatGPT vibe coding an instrument of war.

(Author’s note: I want to mention that while CX2’s market is currently focused on providing military technology to the West, there are other future applications for this too. Mapping and dominating the RF spectrum is useful if you want to keep drones away from arenas and airports, as one example.)

The team for the job

The team at CX2 is not your typical team; the environment is not your typical environment.

Usually, when you hear about impressive coworkers at your company, you hear people say things like “they were early at Stripe.” At CX2, you may be slightly more likely to hear “they were a commander in the U.S. Army,” or “they built RF systems at SpaceX.” The team at CX2 has a combined experience of more than 22 tours of experience in war zones. They know the mission because they’ve lived it.

As for the environment, “if you prefer to stay behind a desk… and you don’t enjoying being outside,” someone said, then you probably wouldn’t fit in at CX2. They test their products frequently in the Mojave Desert, at military bases, and at test ranges. In total, the team was outside testing for at least half of last year. “And 2026 will likely be even more than that,” I was told.

The culture at CX2 is rather intense. The stakes and standards are higher than at the average startup. The problems are more complex. “We have a few core values,” Nathan Mintz (CEO) said. “Show don’t tell; always remember who the end customer is; and to be open and transparent with feedback.”

Porter Smith described working at CX2 like driving in Formula 1. “Quick decision-making, rapid prototyping, and swift action are our DNA. An honest mistake caused by a sense of urgency is better than lethargic indecision or action. We prioritize and focus ruthlessly,” he said.

I do not get the sense there is much room for bullshit, or for puttering around, at CX2. And my favorite quote about the way they view work is just three words: “Get shit done.”

Nathan’s goal as leader is not to burn people out (“we take vacations and believe in weekends,” he said), but the “concept of work/life balance is certainly more biased towards a fast-paced startup environment. [This is not] for someone who is just looking for an 8-5 job.”

Amid all of this work on what is self-evidently an ambitious and important mission, however, I do get the sense that the team at CX2 is having fun. The company’s office is just a 7-minute drive from the beach, where one employee at the company says he surfs most mornings before work.

Responses about favorite stories, moments, and non-work pieces were intriguing:

• “Polish train rides, nights in the desert, and Drone Jesus.”

• “Going out to Lucerne to film our first product videos a month into starting.”

• “Our team spends a lot of time in the Mojave desert doing testing and we’ve developed a certain set of rituals and terms for the various things we see there. One time, one of our Airbnbs was supposedly haunted and someone found some sketchy letters from one of the previous tenants in a public drawer. It’s become part of our recurring lore.”

• “Don’t forget how close we are to the beaches of South Bay!”

Should you consider joining CX2?

CX2 is not your typical startup, and working at CX2 is not your typical startup job experience. And I think that, for the right person, CX2 could be an excellent fit.

Here is how I might describe the right kind of person:

• You want to solve highly technical problems; RF is a difficult domain and adversaries are working on their own solutions, too

• You find defense and systems engineering fascinating

• You want to build products that ship to real-world environments where the stakes are high

• You want to be outside testing products often

• The idea of defining how 21-st century warfare is fought sounds interesting to you

• You want to have a big impact with lots of ownership at an early-stage company

You could also ask yourself this question, which Nathan Mintz shared with me: “Are you interested in joining a top-shelf team that is reshaping how the West fights electronic warfare?

If this sounds like a problem worth working on, and a team worth working with, you can view open roles at CX2 here. The company is hiring at their El Segundo office and, right now, is mostly looking for engineers (both software and hardware).

Thanks to CX2 for supporting Next Play and making this spotlight possible.

2025 was a strange year for the public perception of AI-generated video.

As the technology to produce realistic-looking video on command (e.g. Sora 2 and Nano Banana) became widespread, so did the number of startups aiming to capitalize on this new capability. This was fun at first, but fatigue set in quickly. Really, another AI video generation app/website/product?

The word “slop,” now mostly used to refer to meaningless AI-generated content, was Merriam-Webster’s word of the year in 2025. Google Search interest for ‘AI video generator’ peaked in the fall of 2025 but has since begun to wane. Not everyone wants to see another wacky AI video of…Whatever.

There is a twist, though, there must be a twist, or we would not be writing this essay.

Amidst the graveyard of already-failed and currently-failing and soon-to-fail AI video generation applications there is at least one company that is doing better than ever: Higgsfield.

“On fire” would be an understatement to describe Higgsfield’s arc over the past year:

• Higgsfield publicly launched its video generation tools on March 31st, 2025.

• Since then, they have gathered more than 15M users.

• They currently report $200M run rate (doubled from $100M just 2 months ago).

And, in perhaps the biggest news, Higgsfield last week announced their $130M Series A funding round at a $1.3B valuation with participation from Accel, AI Capital Partners (Alpha Intelligence Capital’s US-based fund), and Menlo Ventures and others. Yes, those numbers are correct. For context, Anysphere (Cursor) raised $60M on a $400M valuation for their Series A; Anthropic raised $124M on $461M.

Higgsfield’s latest round is much like their growth over the past year—historic.

If you join Higgsfield to work under founders Alex Mashrabov, Yerzat Dulat, Mahi de Silva & co., you are not joining because you think this early-stage startup may one day become a unicorn. It already is. No, you are joining Higgsfield because you believe this story ends with Higgsfield growing to $10B, or $50B, or $100B, or more. You are joining because you believe Higgsfield can become one of the most important tech startups of the 2020s. Whether that will happen is anyone’s guess, but I hope the following piece can explain a little about what Higgsfield is doing, why they are winning, and whether you should join.

Not all AI video is created equal

Just five years ago, the idea that you might be able to generate a beautiful video with AI sounded like science fiction. It would have been outlandish to tell a social media marketer that they would be able to create video ads, user-generated content (UGC), and organic content, by typing a prompt into a computer.

In 2026, however, this is the reality of video generation. Models like OpenAI’s Sora 2 and Google’s Nano Banana have gone viral; anyone can now generate a video of almost anything. This should be exciting. It felt exciting at first to me, and probably to most people who tried creating their own videos.

But as I spent more time with the videos people were creating, doubt crept into the back of my mind. I had one thought: How much of this is actually valuable? A video of flaming giraffes jumping off a train Mission Impossible-style is cool, I guess. But what is the point? Where is the value?

What is the market here?

If the first step to AI video is simply having the raw energy—the magic—accessible, then the second step is figuring out a way to harness that magic. This is, as I learned quickly when talking to Alex Mashrabov, co-founder and CEO of Higgsfield, exactly what Higgsfield has done.

Most of Higgsfield’s run-up to $200M and 15M users thus far has been driven by marketing teams—especially on social media. “In the near future, half of all social media video will be generated with AI, and most of that AI-generated content is going to be created with Higgsfield,” Alex said. Jeff Herbst, an investor, described the social media market as bigger than Hollywood (another potential marker for Higgsfield).

To Alex and his team, the timing feels perfect. “First, generative AI has moved from being a research experiment to a practical, mass-market technology,” he said. “Second, the demand side exploded. Social platforms today require constant content. We are not trying to invent artificial demand; the problem already exists at a massive scale. Third, we are in the middle of a very strong global trend. Venture-backed startups succeed when they ride a wave, not when they fight against the current. And small teams [like Higgsfield] suddenly have an unfair advantage [because of] AI-native tools and workflows.”

Social media has (primarily) fueled Higgsfield’s ascent, but the potential market is much bigger than that. You could imagine a world where Higgsfield is the default creative tool for anyone who wants to create any video or create any visual effects. This is a market in the hundreds of billions annually, if not bigger.

“In 10 or 20 years, we see a company with billions of users, hundreds of millions of commercial videos, a thriving professional ecosystem, and billion-dollar revenues. A true platform of the industry, setting the standard for the industry,” Alex said.

Co-founder Yerzat Dulat is more laconic: “[we are] building the biggest media company ever.”

Building the biggest media company ever

Only a month after Higgsfield’s public launch, software engineers had a meltdown on Slack.

“I think we’re under a massive DDoS attack,” frantic messages read. “Servers are down. Everything is down. Traffic just spiked 20x.” Engineers worked through the night, calling AWS and Cloudflare, attempting to understand what was going on. Could it be real, they wondered? But DevOps experts all had the same opinion: “This cannot be real traffic. Nobody grows this fast.”

“And then,” Yerzat said. “Cloudflare confirmed. It wasn’t an attack. It was real people. Millions of real users trying to use our product at the same time.”

Thus marked the beginning of an eventful growth trajectory for Higgsfield that has not slowed down since. After launching their generative video tools, Higgsfield has become one of the fastest-growing startups in the world. The platform has attracted the attention of prominent cultural figures, including Madonna, Snoop Dogg, and Will Smith. As videos went viral, users poured in. Topline went up and up and up, and is now at $200M run-rate.

How? Well, you don’t need to look further than the team.

Alex Mashrabov, CEO, sold his former company AI Factory to Snap for $166M and went on to lead Generative AI at Snap. Yerzat Dulat, CTO, is a top AI researcher and author of popular open-source libraries on GitHub. Mahi de Silva, CSO, has decades of leadership experience (Triller, Amplify, Verisign). He was also a senior engineer at Apple in the late ‘80s and early ‘90s—how often do you hear that?

Look down the resumes of anyone else at Higgsfield and your reaction may be similar: impressive.

“Honestly, this is the most concentrated pool of top-tier talent I have ever seen in the market,” said Dalida Yerkuliyeva, Higgsfield’s VP of Talent. That is certainly the kind of thing you want to hear from someone like Dalida, who has spent most of her career helping startups hire great people.

“What first got me interested in the company was its fast growth - I have never seen anything like it,” Ruchita Gupta, VP of Finance, said. “PMF, traction, team are all unreal. This is a once-in-a-lifetime company and opportunity.”

To start, hiring was mostly in-house. “The first 50 people we brought on board were colleagues and experts that Yerzat, Mahi and I knew personally and trusted deeply,” Alex said. Higgsfield’s team has since expanded to well over 100 people. But, when I talked with the folks about the working culture, it seems like that early founding team environment still lives on today.

How hard can you work?

If I asked you to describe the “classic startup work experience,” you might describe a few things:

• Intense, long days

• A bunch of smart people all deeply invested in the problem

• Constant pressure, shifts, wins, and losses

• An expectation to ship extremely fast

• Huddled around laptops in a low-rent garage somewhere

Higgsfield doesn’t work in a garage, but the rest of it is basically accurate.

When new people join, Dalida told me, “you can see the hesitation in their eyes. These are talented people who [are giving up] a lot: comfort, family time. But the moment they see the team in action, working together, everything changes. You can feel the energy, the ambition… and it’s so contagious.”

“I think many outsiders assume that once a startup reaches our size, it becomes more corporate and comfortable,” Alex said. “In our case, it’s the opposite: we scaled the team, but we kept the founder mentality. At most startups the extreme ‘grinder mode’ usually only applies to the very first people - the founding team. We are no longer a tiny startup, and yet we still operate with that same mindset.”

Kamazhay Kospan, the marketing lead at Higgsfield, told me that although the work is intense, “it is hard to find the right word for [the founders’] remarkable trust in us. They really let us own our work and learn along the way. I’ve always thought: wow, Yerzat, Alex, & Mahi really believe in us and listen to us.”

“The magic of Higgsfield is our ability to execute quickly and thoughtfully,” said Matt Stratton, Higgsfield’s General Counsel. “Where some companies spin their wheels in an uncertain environment, we find a way to move forward.”

“Even when I give 200%,” she said, “I feel I can still do more. I don’t get burned out.”

Expectations are high, and they are consistent. “Yesterday’s success is just yesterday’s success. You can be extremely smart and skilled, but if you don’t keep delivering every single day, it’s easy to feel out of sync,” Dalida said. “[The intensity at Higgsfield] isn’t for everyone.”

So who’s a good fit? “Someone who doesn’t sit and wait for instructions,” Kamazhay said. “People who catch questions that are floating in the air, and who don’t believe that something is impossible. You need to be the type who dives in, figures things out, and trusts your ability to solve problems.”

If you “don’t need to wait for things to be fully defined before acting,” Jasmyn Jarnigan, VP of B2B Strategic Parternships, said, “and enjoy operating at the intersection of creative intuition and technical systems,” then you might be a good fit.

“The company is very flat,” Ruchita Gupta said. “If you want a fancy title or to avoid doing ‘analyst’ tasks, then this is definitely not the place for you.” Everyone is expected to do real work.

It helps to contextualize this within the reality that Higgsfield is one of the fastest-growing startups on the planet. The people who work there are talented and smart and believe they are building a generational company; they have already reached some historic milestones. If you are this kind of person, and perhaps you are, why wouldn’t you want to work hard?

Should you consider applying to Higgsfield?

There is one expected path for startups at Higgsfield’s stage: a modest $20M Series A, a few million dollars in ARR, and a few hard years of work to hopefully one day reach unicorn status and beyond. This approach sometimes works. Loom, for example—a name that most people in tech know and respect—peaked when it raised a Series C at $1.5B; this was viewed as a huge win.

With Higgsfield I think you have something of a different path. Something closer to a Cursor or an Anthropic or a Perplexity. You have an insanely fast-growing company that is raising way more money than companies usually raise; that is valued far higher than most companies at their stage are; that, if successful, could be one of the most valuable companies of its generation.

The other reason you join, I think, is because you crave something that few startups today can provide: an exhilarating startup experience alongside a lot of talented people. You want to work a 10-12-hour day when necessary. You want to be up late at night, solving problems together as a team. You want to become better at your craft.

And you want to be early at a company which could, if it completes its ambitious and admittedly difficult mission, become one of the most successful startups of the century.

As for how to apply? You can look for careers at Higgsfield here; the company is based in the San Francisco bay area and they’re hiring for both remote and in-person roles. They also have a presence in Kazakhstan. I asked Alex how he feels about cold emails; he’s open to them, though prefers thoughtful LinkedIn messages. (People often reach out with ideas that Higgsfield is already trying, he said, but you are welcome to reach out. Cold outreach is one of the best ways for candidates to get across Alex’s desk.)

You may expect a quick interview process (Dalida said it took “less than 10 days from my first interview to receiving an offer”) and onboarding is “extremely short. You don’t wait to learn everything first.”

If Higgsfield sounds like the kind of place for you, well, I’d encourage you to apply and/or reach out.

Thanks to Higgsfield for supporting Next Play and making this Spotlight possible.