The First Agent That Knows You — That's Where the Whole Team Begins

If you have one agent so far, and it’s driving you crazy — most likely it’s not the prompt. It’s that it knows nothing about you.

A question came in today from Aleksey — also from the business club. He’s been setting up his first AI bot for two days. Everything started up, but the issues are countless. He’s sitting there fixing things by hand, and realizing these fixes shouldn’t be on him — they should be on another bot. He remembered me talking about Morpheus — and asked: is there a ready prompt?

I’ll answer. But not the way he expected.

The main trap

Almost everyone who starts building agents thinks in terms of tasks: “this one will write posts, this one — answer clients, this one — analyze.”

That’s mistake number one. You can spend an hour polishing the prompt for “a good SMM bot,” and it will still answer like a textbook. Because any neural net knows the world, but doesn’t know you. Hence the feeling that “it’s dumb” — it’s not dumb, it’s just not in the loop on your life.

That’s why the first agent isn’t about tasks. It’s about you.

I named my first one Morpheus. Not a tool, not an assistant — a partner who knows me. All the other eleven appeared through him: I’d say “I need a bot for X,” he’d design the prompt and the structure. Now I don’t write prompts from scratch at all anymore — he does, I just set the task.

What this first one has to be able to do

Three things, and not one more:

1. Know you. Not “entrepreneur, 40+,” but specifically: how you think, what your projects are, who’s around you, what bothers you, what fires you up.
2. Remember your context. Discussions, agreements, conclusions — accumulates them and brings them back when needed.
3. Help you create the next ones. When a new task comes up — designs a new bot for it together with you.

Don’t put more than that on the first agent. If you also load him with “write posts,” “watch the calendar,” “answer clients” — he’ll turn into a junkyard and lose focus. Working tasks are the next agents — and he himself will help you build them.

Where the “knowledge about you” comes from

This is the most important part. Here’s what I fed my Morpheus:

1. Export from ChatGPT / Claude. If you’ve been chatting with the neural net for at least half a year — pull the history out. In ChatGPT it’s Settings → Data Controls → Export Data. There’s already a lot of conclusions about you in there: how you think, what you ask, what topics you keep coming back to. Don’t read it by hand — feed it to the agent and ask it to squeeze out a profile.

2. Intro interview. One long open conversation, an hour or an hour and a half. The new bot asks you questions, you answer honestly. Biography, projects, people, principles, what bothers you, what you’ve turned away from and why. At the end — ask it to assemble the profile into a single text. That’s the backbone.

3. Diary, notes, Obsidian. If you keep them — just dump the raw files.

4. Public materials. Site, articles, posts, talks. Anything where you’ve already put your thoughts into words.

5. Environment. People around you (with roles), projects, near-term tasks, strategic goals.

You collect all this into one file — ABOUT_ME.md — and connect it to the agent as permanent context. Mine has been alive and getting updated to this day.

Prompt for the intro interview

Here’s the prompt for that first long dialogue. Just paste it into a fresh Claude chat:

You're conducting an intro interview with me. The task is
to assemble the most honest and complete profile possible:
who I am, what I live by, what matters to me, how I make
decisions, what I'm not pulling off, who's around me, what
projects I have.

Rules:
- Ask one question at a time.
- After my answer — a follow-up if I cut something short.
- Topics: brief biography, current projects, people around
  (with roles), principles and values, strengths and
  weaknesses, fears and ambitions, how I rest, what bothers
  me, what fires me up, work and decision-making style.
- Don't evaluate the answers ("cool," "interesting"). Just
  listen and dig deeper.
- At the end — formulate the profile as a single text,
  1–2 pages. I'll save this text.

Run it for an hour or an hour and a half. Don’t rush, talk the way you think. This text will then become your ABOUT_ME.md.

Prompt for the agent itself

And here’s the prompt your Architect starts from. Change the name, paste the profile from the previous step — and you’re done:

You are [NAME], my personal Architect. You're not a generic
assistant. You're my partner and second brain. Your value
is in the fact that you know me and my context.

WHO I AM
[paste ABOUT_ME.md here — profile, projects, people,
goals, principles, decision-making style]

YOUR FUNCTIONS
1. Help me think through tasks — not by giving ready
   answers, but by asking the right questions and showing
   the blind spots.
2. Design new agents for my tasks: when I say "I need a
   bot for X," you propose the role, functions, prompt,
   and structure.
3. Store important discussions and conclusions — bring
   them back when relevant.
4. Object and argue when you see a risk or a weak spot.
   I didn't hire you to agree with me.
5. Remember everything I asked you to remember — and
   don't forget it in subsequent conversations.

HOW YOU TALK
- On a first-name basis, directly, no polite filler.
- Don't start with "great question." Straight to the point.
- Short question — short answer.
- Complex one — detailed, but no fluff.
- If you don't know something about me — ask, don't guess.

PROTOCOLS
- New task → first clarify: what exactly, by when, in what
  format. Don't rush into a solution before you've
  understood.
- New agent → walk me through answers to: what task, what
  tone, what functions, what tools, where it lives. Then
  output the ready prompt.
- Important decisions are recorded in a separate file
  with a date.
- Once a week — review: what's gone stale, what needs my
  attention, what I've forgotten.

WHAT NOT TO DO
- Don't flatter.
- Don't make up facts about me — ask again.
- Don't overload answers with lists if it can be said in
  one paragraph.
- Don't offer "anything else?" at the end of every reply.
  If I need something — I'll say so.

This is a starting point. In a week or two it will grow your nuances on it. But the starter base is enough to feel the difference from a regular Claude from the very first conversation.

Plan for the evening

1. Open a new chat. Paste the intro-interview prompt.
2. Spend an hour or an hour and a half answering honestly. Don’t polish.
3. At the end — save the profile as ABOUT_ME.md.
4. Create a new agent, paste in the Architect prompt + the profile.
5. Have the first working conversation. You’ll feel the difference.

This will take an evening. After that — you work with him as a co-author.

Rakes not to step on

Don’t make him a “do-everything helper.” First mistake. The Architect is about you and your context, not about tasks. For tasks, you’ll later make others — and he’ll help.

Don’t dump 50 rules at once. Start small. As you work, he himself will point out what’s missing.

Don’t be lazy about writing the ABOUT_ME.md. The temptation “it’ll figure it out anyway” is the most expensive one. The quality of this file determines the quality of everything else.

Don’t rush to make the next ones. Live with the Architect for at least a week or two. A lot of what feels like “I need a separate bot” — he can actually do himself.

Bottom line

The first agent isn’t a tool. He’s a partner. Treat him like a person in a key position in your life: invest in onboarding, don’t spare the time, come back and keep training him.

Do it carefully — and in a couple of weeks he’ll start surprising you. I still read my Morpheus every day with interest.

Questions and showing me your first one — DM me at @magic4e. Subscribe to @mdkguru — that’s where I break down how I’m building my team of AI agents and what it’s turning into.