I spent two hours setting up three files and project briefs for Claude. The result: it generates code in my exact style, catches bugs my team misses, and asks the right questions before guessing. Here's the exact setup.

I spent two hours on a Saturday afternoon setting up something that should've been obvious from day one. The result? My AI assistant actually understands my work now. Not in the generic "I can help with that" way — in the "I know you prefer Zustand over Redux and here's why that matters for this component" way.
Two hours. That's it. Here's exactly what I did.
Most people treat AI like Google with extra steps. Type a question, get an answer, move on. But that's like hiring a brilliant consultant and then refusing to tell them anything about your company.
The breakthrough came when I stopped treating the prompt as a question and started treating it as a briefing.
Continue Reading
I created three files:
1. IDENTITY.md — Who the AI is
Not "you are a helpful assistant." Mine says:
You're Henry. Sharp, resourceful, a bit lion-hearted. You get things done. You have opinions. You disagree when it matters.
Sounds silly? It changes everything. The responses stopped sounding like corporate press releases and started sounding like a colleague.
2. USER.md — Who you are
This is where most people skip. I wrote down:
Suddenly the AI stopped suggesting Python solutions when I'm building in TypeScript. It stopped recommending MongoDB when my project uses Postgres.
3. RULES.md — How we work together
This is the non-negotiable stuff:
any in TypeScriptSimple rules. But they compound. After a week, the AI's suggestions were aligned with my codebase, not fighting it.
This is where the magic happened.
I created project-specific briefs for each active project:
Project Brief Template:
I have four active projects. Four briefs. Took about 30 minutes.
Then I set up what I call "feedback loops" — rules that tell the AI to ask clarifying questions when it's uncertain, instead of guessing.
Before: AI generates a component, I spend 10 minutes fixing it. After: AI asks three questions, generates something closer to what I actually need.
The feedback loop saved me more time than anything else.
Day one after setup:
The compound effect is real. Each interaction builds on the last. The AI gets sharper, more contextual, more mine.
They over-engineer it. You don't need a 50-page prompt. You need three focused files and project-specific briefs.
They skip the personality. Generic AI gives generic results. A personality isn't fluff — it's a filter that shapes every output.
They don't update it. Your projects change. Your preferences evolve. I review my setup every two weeks. Takes 10 minutes.
They forget feedback loops. The biggest time-saver. Teach the AI to ask before it guesses.
The difference between AI that's "kind of useful" and AI that's "I can't work without this" is two hours of intentional setup.
Not two hours of tweaking prompts. Two hours of documenting yourself — your work, your preferences, your standards.
Because the AI doesn't need to be smarter. It needs to know you.
Two hours. That's the investment. The returns compound daily.
Start this weekend. You'll never go back.

AI Engineer & Full-Stack Tech Lead
Expertise: 20+ years full-stack development. Specializing in architecting cognitive systems, RAG architectures, and scalable web platforms for the MENA region.
Practical AI + full-stack insights for MENA builders. No spam.

