How 44 Weekly Reviews Became the Document That Makes All My AI Agents Smarter
Was on a call with Brooks a few months ago. He's my podcast co-host and we were talking about AI personalization. He'd been testing agents for email drafting and kept getting outputs that sounded… fine. Correct tone. Decent structure. But generic. Could've been anyone.
I showed him something. Pulled up a Google Doc. 35 pages long. And told him every single agent I run reads this document before it does anything.
His first question: “How did you build that?”
What's In the Document
It's everything about me. My name, where I live, my time zone. My bio. My current revenue streams and what I'm focusing on this quarter. My core values and principles. How I make decisions. What my brand voice sounds like. Communication preferences. Even the businesses I'm running and what stage each one is at.
It's not a resume. It's more like a really thorough operating manual for “how to be Thanh.” The kind of document that, if you read it carefully, you'd know exactly how I'd respond to any given situation.
I think of these as context files… reusable text documents that encode who you are in a way AI can use. They're assets. You build them once, load them into every workflow, and the output quality goes up across the board.
How I Actually Built It
Here's the part that surprises people. I didn't sit down and write 35 pages from scratch. That would take forever.
I do a weekly review every Sunday. Have been doing it for years. Each review covers what went well, what didn't, what I learned, what I'm focusing on next week. Pretty standard productivity stuff.
I took all 44 weekly reviews from 2025 and pasted them into ChatGPT. Literally just dumped them in and said something like: “Combine everything here into a comprehensive profile of who I am. Include my values, priorities, communication style, business focus, and decision-making patterns.”
35 pages came out the other side.
Then I cleaned it up. Removed things that were outdated. Added a few sections about my brand voice and writing style. But the bulk of it came straight from a year's worth of self-reflection that I'd already done.
What Changes When Agents Have This
The difference is immediate. Without the context profile, AI drafts emails that are technically correct but could be from anyone. With it, the drafts sound like me. My email agent knows I write short, casual messages. It knows I use ellipsis instead of em dashes. It knows I prefer direct language over corporate speak.
My content agent knows which topics I care about right now. It knows my stories. It knows the analogies I tend to reach for.
My meeting prep agent knows my priorities without me re-explaining them. It knows which projects are active, which ones are on hold, and what metrics I track.
All because every agent reads the same 35-page document before it starts working. Centralized context. One source of truth for all of them.
Why Most AI Feels Generic
Here's the insight that clicked for Brooks on that call. He said: “So it's not about the prompt. It's about the input.”
That's exactly it.
Most people spend all their energy crafting the perfect prompt. And the prompt matters… but it's maybe 10% of the output quality. The context you give the AI is the other 90%.
Generic prompt plus no context equals generic output. Decent prompt plus 35 pages of context about who you are, how you think, and what you care about… that equals output that actually sounds like you.
How to Start Building Yours
You don't need 44 weekly reviews. Start with what you have.
If you already do weekly reviews: Upload the last 3-6 months into ChatGPT or Claude. Ask it to synthesize a profile of who you are, what you prioritize, and how you communicate. You'll get a solid starting point.
If you don't do reviews: Start with a brain dump. Spend 30 minutes answering these questions:
- What do you do for work? What are your current projects?
- What are your top 3 priorities this quarter?
- How do you prefer to communicate? Short and direct or detailed?
- What values guide your decisions?
- What does your typical week look like?
Save that as a document. Even 2-3 pages is better than nothing. You can grow it over time.
The key principle: Keep it in a readable format. Plain text or Google Doc. Something any AI tool can ingest. Don't lock your context inside one platform.
The Compounding Effect
Here's what's kind of wild. The document gets better over time. As I do more weekly reviews, I update it. As agents learn new things about my preferences, some of them can actually update the profile themselves.
Six months from now, the document will be even more accurate. A year from now it'll know me better than most people do.
That's the real play. Build this once. Keep feeding it. And every AI tool you use from now on starts with a massive head start on understanding who you are.
Your weekly reviews aren't just for reflection anymore. They're training data for your AI.
