Every week, more than 10,000 Stripe employees open an internal chatbot called Kai and ask it questions about the company’s code, dashboards, runbooks, and policies. It answers in seconds. Kai isn’t magic. It’s the result of a deliberate engineering playbook that the Stripe team published last week, and the same playbook works at a smaller scale for one-person businesses, freelancers, and small teams drowning in their own notes. You don’t need Stripe’s budget, Stripe’s engineering team, or Stripe’s data center. You need an afternoon and a ChatGPT or Claude account.
Here’s how to actually build your own company brain in 30 minutes, the beginner version of what Stripe built.
What Stripe actually built (and why it works)
Sharadh Krishnamurthy, the engineering manager behind Kai, walked through the system on Lenny’s Podcast this week. A few details stood out.
First, they built it themselves instead of buying it. Stripe had access to plenty of off-the-shelf AI tools, but their context is unique: years of code, internal docs, dashboards, ticket history, customer support transcripts. No vendor was going to package all of that. So they wired the AI directly to their own data layer.
Second, “projects” are a governance mechanism, not a folder. When a Stripe employee opens a project in Kai, the AI is scoped to that project’s data sources and permissions. Switching projects changes what the AI is allowed to see. That’s how 10,000 people can use the same chatbot without accidentally leaking one team’s data to another.
Third, anyone can package a workflow as a “skill.” Two thousand skills exist in Stripe’s catalog now. Marketing has a skill for drafting a launch announcement. Customer support has a skill for summarizing a ticket thread. Engineering has a skill for generating a code-review checklist. The infrastructure is the same across teams, but the workflows themselves are owned by the people who actually use them.
The full Lenny’s interview with Sharadh is worth a listen, because the core idea is portable: bundle your institutional knowledge, give the AI access to it, and scope what it can see per project. Stripe’s version is enterprise-scale; the freelancer version takes 30 minutes.
Why “company brain” matters for a one-person business
If you run a solo practice, a small shop, or any business where you wear every hat, you already have the raw material for a company brain. It’s scattered across Notion, Google Drive, Slack DMs, your inbox, your invoices, and that one folder of “important stuff” you keep meaning to organize.
You probably also find yourself answering the same kinds of questions over and over: what’s our refund policy, did we already book the photographer for the Q4 event, how to phrase this update for clients, what the standard scope is for a 6-week brand engagement, which supplier we use for X.
If those answers live in your head, your business has a bus-factor problem. If they live in scattered docs, your business has a search problem. Either way, an AI that knows your stuff fixes both. You stop being the bottleneck.
The other angle: a company brain lets you onboard help faster. When you eventually hire a contractor, a part-time VA, or your first part-time employee, you hand them a chat link and they can ask questions instead of pinging you at 11 PM. That’s what 10,000 Stripe employees get from Kai. You can get 80% of the same value with 0.1% of the engineering effort.
The beginner version: ChatGPT Projects, Claude Projects, and Custom GPTs
You don’t need to build an internal platform. Three off-the-shelf tools do this for you, and they’re all under $25/month.
ChatGPT Projects (the easiest starting point)
ChatGPT Projects lets you bundle a custom instruction, a stack of files, and a memory into one workspace. Open a project, drop in your SOPs, your FAQ doc, your old proposals, and your brand voice guide. Now when you chat inside that project, the AI uses those files as its source of truth.
For an example: a freelance designer could create a “Client Projects” project, upload the brand guidelines for their top three retainer clients, and ask “draft a launch tweet for the new Acme campaign in Acme’s voice.” The answer comes out already on-brand, no manual context-switching.
Pricing: ChatGPT Plus is $20/month and includes Projects. If you want bigger file uploads and more memory, ChatGPT Pro is $200/month but most freelancers won’t need it.
We already covered ChatGPT’s company knowledge feature in detail; Projects is the consumer-facing version of that idea and it’s the right starting point for most people.
Claude Projects (if your knowledge base is mostly PDFs and docs)
Claude Projects works the same way as ChatGPT Projects — custom instructions, file uploads, persistent context — but with two practical differences. Claude handles longer documents better and tends to be more careful when summarizing dense source material. If your company brain is mostly PDF reports, contracts, and long-form docs, Claude Projects is the friendlier tool.
Pricing: Claude Pro is $20/month with Projects included. If you’re already paying for ChatGPT Plus, you can use both: ChatGPT for everyday chat, Claude for deep document review.
Custom GPTs (for repeat customer-facing answers)
Custom GPTs are different. They’re public-facing AI assistants you can publish and share. If your business has a public FAQ — a yoga studio with class descriptions, an e-commerce shop with shipping policies, a software startup with developer docs — a Custom GPT trained on those docs lets customers ask questions in plain English and get accurate answers from your actual content.
Pricing: requires ChatGPT Plus ($20/month). Custom GPTs can be free for users, or you can put them behind a paywall inside ChatGPT itself.
For a related workflow, our guide on building reusable AI assistants with Gemini Gems walks through the Google-side equivalent.
A 30-minute setup that actually works
Here’s the recipe.
Step 1: Pick your “brain” workspace
Decide where this will live. For most beginners, ChatGPT Projects is the simplest starting point. Sign in, click Projects, create a new one called “Company Brain” or something specific to your business. ChatGPT Projects is more flexible about file types than most alternatives and stays with you across devices.
Step 2: Gather your materials
Open a folder. Drop in anything that contains answers your business needs to remember:
- Top 10 customer questions and your standard responses
- Pricing page and what’s included / not included
- Service descriptions and standard scope of work
- Brand voice guide (tone, do-and-don’t words, sample copy)
- Refund and cancellation policies
- Standard emails you send (welcome, follow-up, post-project)
- Old project briefs you can mine for reusable structures
- Anything else you’d want a new hire to read on day one
Don’t overthink this. Ten messy files is better than zero perfectly-organized files. You can refine later.
Step 3: Write the project instructions
This is the part most people skip, and it’s the most important. Inside your Project, write a short instruction (200-400 words) that explains to the AI how to behave in this workspace. Include:
- Who you are and what your business does (2-3 sentences)
- Who your typical customer is (1-2 sentences)
- How the AI should respond (voice, length, format)
- What it should always do
- What it should never do (e.g., “never invent pricing we haven’t published”)
- A few example questions and your preferred answers
This is the equivalent of what Stripe does with their project-scoped governance. The AI isn’t guessing how to behave in this workspace — you told it.
Step 4: Test it with a real question
Now ask your new company brain three real questions you actually had this week. Check the answers. If it gets them right, you’re done. If it gets them wrong, you have two options: add more files (more context) or sharpen your instructions (clearer rules).
Do not skip this step. The whole point of a company brain is that it answers your real questions. Test it like you’d test a new hire.
What you should NOT do
A few traps beginners fall into.
Don’t dump every file you’ve ever created into one project. Projects work best when they’re scoped: a “Client Projects” project, a “Marketing” project, a “Personal” project. Scope is what makes the AI useful — too much data and the AI starts hallucinating connections that aren’t there.
Don’t use the company brain as the only source of truth for high-stakes decisions. AI is a brilliant first draft and a terrible final authority for medical, legal, tax, or safety questions. Pair it with a human expert for anything you’d sue someone over.
Don’t pay for ten different AI tools at once. Pick one (ChatGPT or Claude), set up Projects, run it for a month, and only add a second tool when you’ve hit a real limit.
Don’t expect it to organize your files for you. The brain works only as well as what you put into it. Spend 30 minutes on day one loading it up. Maintain it monthly. That’s the entire workflow.
The takeaway
Stripe’s internal AI agent is just a big version of what you can build this afternoon. Bundle your docs, write tight instructions, scope by project, test with real questions. That’s it. Ten thousand Stripe employees save hours every week using Kai. A freelancer with a Project full of client docs can save the same hours without writing a line of code.
If you’ve been drowning in scattered notes and re-answering the same questions, this is the cheapest productivity upgrade you’ll make all year. Thirty minutes of setup, $20 a month, and your business finally has a brain that isn’t just you.
