TJ Larkin

AI workflows

How do I give AI context about my business without re-explaining it every time?

Short answer

Have the AI interview you once. Ask Claude or ChatGPT to ask you about 40 questions about you, your business, your processes and workflows, then turn the answers into a context document. Put that file in your project folders or upload it to any chat. For more than one brand, keep a separate business bible for each.

The fastest onboarding for any business owner new to AI takes about an hour and one prompt. You don't write the context document. The AI interviews you and writes it.

Why let the AI interview you instead of just telling it what to do?

Because AI is an eager intern. The second you say something, it runs out of your office to go do it. Say "go pick me up a sandwich" and it's off to Subway for a crappy sandwich. It won't ask what kind unless you tell it to ask.

The best advice I heard from an executive coach was exactly that: have AI interview you about your problem. Don't dive in and start telling it what to do. Tell it to ask you what you want first.

What prompt do I use?

This is the one I give business owners, close to word for word:

I haven't used AI much. I want to build a foundational document for context.
Ask me 40 questions about me, my business, all my processes and workflows.
Then give me a document and an HTML file of everything I do.

Answer out loud if you can. Dictation gets more context into your answers: why I talk instead of type. Talk to it like a person, not like Google with keywords and short sentences. When someone tells me ChatGPT sucks at writing, I ask what prompt they gave it. Usually it's one line with no context, so they got the most generic output. Give it tons of detail and you get a detailed answer back.

What do I do with the document?

"Then you have a context document you can put in those folders, or upload to a chat, and you never have to give it context again."

  • In Claude Cowork, Claude Code or Codex, drop it in the project folder. Those tools read files, so the context is always there.
  • In regular chat, upload it at the start of anything that matters. Projects let you attach files, too. It's more manual, but it works.
  • For your team, share it. Everyone's AI starts from the same picture of the business.

Then flip how you ask. Instead of telling it what to do, give it your scenario and your goal, list everything you have, and ask: how can you help me optimize everything we're doing? What can you do to get us to extremely high value quickly? I forget to do this too. Most people only ever give it instructions.

Can one interview build a whole set of files?

Yes, and this is where it gets good. Another AI trainer described a setup I started testing the same night: the AI interviews the owner, then a Claude plug-in builds the whole document set from that interview. It cascades. It builds one document, then references that document and the interview to build the next, and so on down the chain. Each step is smarter about the business than the last.

After one call, you'd have company information, how you like to write, your tone and your rules all in one folder. From the same interview, you can tell the AI to build every document you need for a company brain for AI coding and Cowork. Everything in my folders, you could do in an hour as a non-technical person.

What if I run more than one brand?

Give each brand its own business bible. When I start a new project, I talk through the plan out loud with the AI listening and have it write a bible: what we're doing, why, the goals, the plans, the target audience and how we'll grow it. That file goes into a Claude project my teammates share, so everyone starts from the same context.

Without that separation, voices leak. I need a social media style, a WilCo Guide style, a greetform style and a Local Media HQ style. Social formatting can't leak into articles, and WilCo Guide's casual first-person voice can't show up in greetform copy. The AI should know which project it's in and load that brand's skill and bible automatically. Different brands may even need different agents.

How do I use this with clients or new hires?

I've built it into my sales process. After a client pays, they get an intake form. Once they fill it out, the system produces a styled HTML company overview and a markdown context file their employees load into their own AI chats.

It's also the best demo trick I know. Back in 2024, my recipe was to have a prospect spend about 10 minutes answering questions, a paragraph each, then load that business profile into custom AI assistants built for their industry. When they type "create me some social content" and see their own brand and city come out instantly, it lands far harder than a generic demo. You don't need 20 or 30 use cases. About five gets them asking "what about this?"

For my own operators, I go further. A new operator gets a cloned copy of my whole operating environment: instruction files, brand guidelines and project context, set up to work in Codex or Claude Code. They act as an extension of me with fewer permissions.

Why does living context beat a chat that starts from zero?

A chat tool starts every conversation knowing nothing. Even in early 2025, my line between a beginner and an advanced user was whether they understood saved custom assistants: uploading a knowledge base and writing instructions for the assistant's job so you don't repeat them every time. Turn each repeated use case, like writing memes, into its own assistant, and nobody on the team needs to know prompting.

A properly configured agent with living context, files it reads every time and updates as things change, goes further still. It will understand your business better than a chat ever can. That's the logic behind a shared company second brain and good agent instruction files.

Answered by TJ Larkin. From AI training sessions, client and team calls, and my own notes to my AI agents, 2024-2026. Last updated .

Related answers

TJ Larkin / Answers / AI workflows