Tactical AI guide for non-technical people

Some practical lessons for using AI to boost your productivity at work

✨ Hey there this is a free edition of Next Play’s newsletter, where we share under-the-radar opportunities to help you figure out what’s next in your journey.

One of the more common questions we see in our inbox and in the friends of next play Slack group (btw you can join here!) is people asking about artificial intelligence.

They want to know if and how they should be using AI: How can I use AI to help me as I build my next company? How can I use AI to assist with my job-search? How can I use AI to be more productive at work?

We thought it’d be useful to try to get you some answers to those questions—so we talked to Next Play community member Linda Mutricy. Linda used to work at Flexport and has spent the past few years building automations for operations teams.

We asked her to write a guide to using AI productively (particularly for not very technical people). It is about 2500 words of tactical, no-bs tips — showing you practical examples of how you can use AI in various settings across data analysis, customer success, content creation, and more. It also includes a sample practice project towards the bottom that you can start experimenting with to test your skills. We hope it’s useful!

---

By the way, if you are looking for job opportunities, here are some lists we recently published:

• 38 fast-growing startups under 25 people

• Sequoia-backed companies hiring remotely

• Companies hiring that recently raised $10-50M rounds

---

Starting with the basics: What is Generative AI?

Generative AI works by training neural networks on vast amounts of data from the internet, creating models that can recognize patterns and generate new content based on those patterns. Think of it as a sophisticated pattern-matching system that can:

• Process human language inputs (your prompts)

• Draw connections between topics based on its training

• Generate responses by predicting what would most likely come next

• Improve with more context from you

While these large language models (LLMs) can produce impressively coherent content, they're essentially making educated guesses rather than truly "understanding" in the human sense.

Their non-deterministic nature helps them shine in situations requiring creativity and with less of an emphasis on a step-by-step process.

Think: an English essay, where there are multiple ways to get an A, and your essay will be different than your classmates’ essays.

LLMs are less good in situations where there is one specific answer. Think: a math test, where there is only one correct answer, and you also need to show your work.

Similarly, in the workplace, LLMs do well when writing marketing copy, sales emails, and customer success responses. In these situations, creating lots of volume can even help us move faster with experimentation. Getting 80 - 90% of the way there, with human revision taking the last 10 - 20% of the steps, is actually pretty useful.

However, LLMs do less well when doing accounting and finance functions. While these roles can be human error-prone, there is little tolerance for error. LLMs struggle here, because they don’t “show their work” well, they don’t follow the same repeated processes every single time, and getting the answer 85% correct means that it’s wrong. (This is why it’s inadvisable to DIY this, and where traditional software automation works better.)

What are some popular basic tools to start with?

You’ve likely heard of 3 major AI chatbots:

1. Claude (claude.ai) - Particularly strong for detailed explanations and code generation

2. ChatGPT (chat.openai.com) - Widely used with strong general capabilities

3. Gemini (gemini.google.com) - Google's offering with strong search integration

The "best" tool depends on your specific needs. The good news: they all offer free tiers, so experiment to find which works best for your particular use cases. Anecdotally, we’ve heard that employing these tools effectively can increase productivity by approximately 30% for individual employees.

Using Generative AI at Work

Content creation and communication

If you haven’t started using LLMs yet, start with these use cases. With simple prompts, LLMs can help you:

1. Draft emails. Ever needed to take a breath before replying to an email? AI can help you generate professional responses and can act as your “anger translator.”

2. Write documents. Create outlines and first drafts of reports and presentations.

3. Optimize content. Improve tone, clarity, and impact of existing content.

4. Prepare for meetings. Generate agendas, talking points, and follow-ups.

5. Generate ad images and ad copy from reference images, moodboards, and inspiration. Here are some ideas for simple prompts – you can include specifics on the text you want in the prompt as well. For example, “Please add a heading, subheading, and Call to Action.” Or, “Please add these benefits on the left-hand side.”

These applications are ideal starting points because they have a high tolerance for minor errors and produce content that you'll review before sharing anyway.

You’ll find that the more context and examples you give LLMs, the better their answers end up being. For example, you can attach brand guidelines, writing samples, and information about your company and Ideal Customer Profile so that it can tailor its responses.

For a deeper dive: Here’s a detailed, step-by-step guide on training AI to write like you.

Customer success automation

Customer-facing roles can also benefit from AI-powered automation:

1. Create response templates: Generate customized response templates for common customer inquiries. Save these response templates in your Gmail.

2. Create Standard Operating Procedures: Record yourself on Google Meet, with transcript enabled, walking through a process step-by-step, making sure to narrate as you go. Upload the transcript to your LLM with a prompt to create Standard Operating Procedures. You’ll now have a video walkthrough to share with new team members and a written SOP.

3. Simple notifications: Set up automated reminders for task completion or follow-ups

Data analysis and reporting workflows

LLMs can help streamline data workflows:

1. Basic report generation: Create simple KPI dashboards or data summaries by uploading the CSV along with your prompt asking for what you’d need.

2. Analysis assistance: Help interpret trends and patterns in your data. LLMs are specifically excellent at interpreting “open-text” or “short-answer” survey responses, in formats such as customer feedback surveys, product reviews, and support tickets. Upload these responses in a CSV, and give it a try. Make sure to scrub personally-identifying information before you do!

3. SQL query, Google Sheets / Excel formula generation: Convert your plain English questions into database queries, or write prompts to brainstorm the Google Sheets or Excel formulas you’d like to write.

Note that large datasets will max out the LLM’s context window, so it might instead opt to write you some code to run the analysis on your own. In that case, you can easily ask the LLM to guide you through the setup.

Learn to code with “vibe coding”

AI is still constrained today with what it can automate. It’s fair to think about LLMs as magnifying your abilities into superpowers, but it can’t give you new abilities.

However, one hack you can use AI for is treating it like a personal tutor for whatever you want to learn, such as learning how to code.

Enter “vibe coding.”

“Vibe coding”, a term coined by Andrej Karpathy, AI researcher and a co-founder of OpenAI, refers to the practice of using AI to generate code from natural English descriptions.

Even without traditional programming knowledge, you can create custom software:

• Describe your desired outcome in detail, and in plain English, by prompting LLMs

• Have AI generate the code to accomplish it, and give you the step-by-step instructions to run the code

• Make adjustments and improvements through conversation

This approach focuses on outcomes rather than technical perfection, making it ideal for rapid prototyping and solving immediate problems.

Expect to get 85-90% of the way there with AI-generated code. Though “vibe coding” is fun, “vibe debugging” to close the 10-15% remaining gap is painful.

Expect to learn coding concepts gradually as you vibe code by developing a basic understanding of:

• How to write clear product requirements

• How APIs work

• Fundamental coding concepts like data structures and algorithms

• Security and version control best practices

The best vibe coders today are software engineers who already understand all of the fundamentals.

Implementation without technical expertise: things you shouldn’t vibe code right now

Note that your newfound vibe coding skills are best used for proof-of-concepts or prototypes, communicating your vision, and automating your personal life.

At work, make sure that you get your leadership on board with your plan before you try to improve any company processes without the prerequisite software engineering experience. And especially pay attention to this section below!

While AI can help with many tasks, some areas still require professional software engineering expertise now:

1. Anything requiring integrations
   Complex integrations between multiple systems often require robust error handling and security considerations beyond what AI-generated code can reliably provide.

   1. Try to instead get started with a no-code platform like Zapier, Make, or n8n. The major drawback here is that LLMs are not as well-trained to tutor you on using these platforms as compared to popular programming languages like Python and Javascript. Additionally, no-code platforms are constrained in what they are able to help you accomplish, as compared to writing code. You might find that learning to code with AI ends up taking less time than getting to a similar proficiency level on these no-code platforms.

2. Anything involving financial transactions
   These require zero tolerance for errors. For example, if AI hallucinates or misreads an invoice for 20 units as 200 units, financial consequences could be significant.

3. Business reporting and data analysis workflows
   For critical business decisions, rely on established tools or professional development.

4. Document processing and contract management
   While basic document handling can be automated, sensitive legal document processing should involve human oversight.

Case study: Vibe code your first AI-enabled automation project

Here's a practical example you can implement today. It requires $0 in additional software (uses your existing Google account), minimal technical knowledge, and about 15 minutes to set up.

It’s important to do this experimentation on your personal accounts, and not on your work account.

If you want to view the video version, I’ve created a Loom narrating this.

The prompt + Claude-generated output is here, linked as a Claude artifact.

Problem: You manually save your email attachments (invoices, receipts, contracts) to Google Drive folders, wasting valuable time.

Solution: Create a Google Apps Script bot that:

1. Checks unread emails for PDF attachments

2. Categorizes them based on keywords (e.g., "invoice" vs "contract")

3. Saves them to appropriate Google Drive folders

4. Logs actions in a Google Sheet for tracking

Implementation approach: Instead of coding this workflow from scratch, I wrote a single prompt to Claude with detailed requirements. Claude then wrote the code and gave me instructions on how to set this project up.

Prompt I used:

<role>

You are a software engineer who helps non-technical people like myself automate admin tasks by creating integrations. Assume that I know nothing about coding, so you must explain to me what I need to do at every step.

</role>

<goal>

Your goal is to help me create a Google Apps Script project to automatically save business document files from my email to my Google Drive folders. This script should be attached to a Google Sheets spreadsheet so that I have a dashboard summary of every item that was saved into each folder.

</goal>

<requirements>

I receive many important documents as PDF email attachments. They can be categorized in a few different ways.

Invoices and receipts will generally have the word “invoice” or “receipt” in the subject line, the body, or the attachment filename. This script should create a folder named “bookkeeping” and save these files into the folder. Contracts are everything else that’s a PDF attachment. This script should create a folder named “contracts” and save these files into the folder.

I want this Google Apps Script project to be attached to a Google Sheet to be viewed as a dashboard. The script should run every 5 minutes. I should have the option to disable it. Every time an email comes in with the attachment, I want to record it in the spreadsheet with the date received, the sender name and email, a link to view the email, and the attachment link after being saved in my Google Drive folder. If an email comes in with multiple attachments, I want to save them all.

</requirements>

Including XML tags really helps the quality of your LLM outputs. More here.

Output: Here’s the Claude output. It includes the prompt and the code that Claude generated. To implement this for yourself, follow the instructions step-by-step.

Note: For better or for worse, this bot WILL save all of your unread PDF attachments to your Google Drive. (Designed for inbox zero people!)

In these situations, Claude’s step-by-step instructions may be dated, but directionally correct (e.g. menu labels may have changed in the software you’re using).

And lastly, no two outputs, even with the same prompt, will be the same (try copy-and-pasting this prompt into your own chat window to test it out). This is the nature of generative AI.

Modifications: It’s likely that this bot, in its current state, doesn’t work for you.

To tweak it to your needs (and get started with vibe coding!):

1. Download a text editor like Visual Studio Code or Cursor.

2. Copy and paste the Claude-generated code into a new file.

3. Name and save the file with a Javascript .js file extension (e.g. email_bot.js).

4. Draft a prompt with the requirements you need, using the below prompt as a baseline. Upload the Javascript file with the prompt.

<role>

You are a software engineer...

</role>

<goals>

I want to achieve this...

</goals>

<requirements>

I want to modify this attached code file to do...

</requirements>

Tools to get started with vibe coding

Start with Claude; the Sonnet 3.5 and 3.7 models especially are great at generating code. Try creating simple workflow automations within your Google Workspace account, using Google Apps Script.

Google Apps Script is actually Javascript, which is a great first language to understand. It’s a popular language, which is ideal from the perspective of using LLMs as a tutor to learn to code.

A popular language means more community support and discussion about it on the internet, which means more training data that has been fed into the LLM, and since LLMs are non-deterministic and make predictions based on their training data, that increases your chances of getting relevant code on the first try.

Likewise, “vibe coding” in Google Apps Script poses relatively lower risk, because your code is hosted at script.google.com and you don’t have to worry about setting up your own server.

Another popular language for getting started coding is Python. It’s specifically helpful for data analysis. Both Javascript and Python are also used in AI software development.

Once you get the hang of vibe coding with Claude, you can branch out into using other tools, depending on your needs. Some popular options are Cursor and Windsurf for AI-assisted code generation.

Other tools that help with generating design-forward prototypes (especially useful if you are a UX designer or product manager) are Lovable, Bolt, Replit, and V0.

Was this guide helpful? Do you want to learn more? Join Linda’s 30-minute Maven Lightning Lesson on April 30, where she will present a few in-depth use cases for AI and automation for people in non-technical roles. Every participant will also receive a free prompt templates document filled with prompts that get results.