The Agent That Grows With You Is Not the One You Buy

I asked for a plan. It handed me back my own brain.

Eight weeks ago I sat down with Claude Code and asked it an operating question. Not a software question.

I wanted my agency to work three weeks a month and take the fourth off. I wanted to know what would have to be true for that to hold.

That question carries weight. It touches pricing, capacity, who does what, what I keep saying yes to. I expected a plan. Something I could read on a Sunday and start on a Monday.

It never gave me one.

What it did instead was keep asking me questions. About my philosophies. About my frameworks. About how I think.

Eight weeks of that, and what I had was not a plan. It was a brain of me. And when the project-manager side started sending me a morning brief of what actually needed doing across every project, my mind went quiet. For the first time. It didn't have to hold any of it anymore, because it was all there. That thing is called Athena.

Sit with the shape of that. I walked in wanting a deliverable and walked out with an extension of my own head.

Three weeks on. One week off. Eight weeks of conversation, and I never got the plan for it. I got the brain that could hold it.

Buying one of the assistants on the market would have cleared the routine work I was already not doing, and left the level that moves the needle exactly where it was.

That gap is the whole article, and here it is in one picture:

Delegating thinking, not tasks

The default move with an agent is to write it like a spec. Here's the output I want. Here's the format. Here's the template. Come back when it's done.

We do the identical thing to people. Traditional delegation transfers output: do it like this, here's the checklist, check in when you're done. The person learns the task and never learns the judgment underneath it. So they come back to you for every exception, and you stay the bottleneck you were trying to remove.

CASE Method: a scheduled, structured post-task reflection meeting where you excavate how someone thinks and bridge the gap to how you think. Traditional delegation transfers the output. CASE Method transfers the input. You delegate thinking, not tasks.

Say that last line out loud, because it is the whole thing and almost everyone who hears "CASE" gets it backwards. They assume it's a way to assign work. It is the opposite. It is what you do after the work is already finished.

If you remember one idea from this piece, make it that one. Which model you point it at is the least important decision in here. You can change that in an afternoon.

The mechanics, and what each step will cost you

This reads like clean steps. It is not clean. Here is where each one bites.

Hand off the task and say nothing else. Not a spec, not a checklist. You are trusting them to bring their own judgment, which means you don't get to pre-load yours. The cost: the first time you do this it feels like negligence, and you will want to add "just one more thing." Don't.

Stay completely hands off while they do it. No hovering, no steering, no mid-flight corrections. This is where leaders break. You will watch the work move in a direction you would not have gone, and every instinct you have will scream at you to correct it. Correct it and you have erased the reflection before it ever happened, because you did the learning for them. Sit on your hands. That discomfort is the tuition.

When the work is finished, sit down for the meeting. Question-led, not a list of corrections you read out. The cost: this is an audit of you. You are going to be asked to say out loud how you think, and a lot of leaders have never done that once. Your judgment is real. It has just never been articulated. Naming it is the work, not the meeting's side effect.

Excavate how they think, then bridge it to how you think. The entire job of the meeting is in that sentence. Not to grade the output. To hear the reasoning that produced it, and to fold yours in where they diverge.

Here is the shape of it on an ordinary piece of work. Not a script, a shape. You finished the proposal, so walk me through it, start to finish. When you hit the fork on pricing, what made you pick the number you picked? What did you expect to be hard here that turned out not to be?

That last question does more work than it looks like it does.

E = Easier Than Expected

Not "E = Easier." That misread turns a strengths-surfacing question into a compliment, and a compliment teaches nothing.

Easier Than Expected asks what someone expected to be hard that turned out not to be. What came naturally.

That answer hands you the person's real strength, the thing they do without effort, which is exactly the thing you want more of. Most leaders never find it, because they are too busy correcting what went wrong.

Now finish the loop, because this is the decision the whole framework exists to force. You have heard how they think. You have bridged it to how you think. The next time a decision comes up, you do not hand over the task again. You hand over the decision.

That is the shift. Not more instructions, not a better checklist. Your team pings you all day because they never got the thinking. The CASE meeting is how the thinking moves. The price for it is your afternoons and your discomfort, and that is the actual cost of building people who decide without you.

The housemate test

Here is the handle I use with account managers, and it lands on your agent exactly.

Reading the client dossier is not knowing the client. I can hand you a folder of account history and you can read it twice and still not know them. Know them the way you know a housemate, by the details nobody writes down.

The test is brutal and simple. Say the client's name out loud. Can you produce their city, their partner, their specialty on demand, without looking? If you can't, you are not up to speed. Reading is not knowing.

And now the part most people skip, which is the price. The intimacy comes from watching every call, not from a summary. That is hours. Real hours, on calls you could have let someone else take. Nobody budgets for it, and it is the entire difference between the manager who knows the client and the one who has read about them.

Point that at your agent. Most people hand it the summary, the cleaned-up handoff doc, the dossier, then wonder why every answer sounds like it came from a different company. You cannot feed an agent a summary and get a housemate out of it. Feed it the calls. The decisions. The "no, not that way, here's why." The raw version of how you think when nobody is editing you.

The comparison

What matters most, and when NOT to do this

Ranked, so you know what to protect: the timing is everything. The meeting happens after, not before. Get that wrong and you are just instructing with extra steps. Everything else here is downstream of it.

If your problem is routine throughput you were already not doing, buy. The market assistants clear that, and that is real value. Just don't expect it to move the level that matters, because it wasn't built to and it won't.

Never run a CASE meeting when what you actually want is to correct someone. A correction dressed up as reflection teaches people that reflection is punishment. You only get to make that mistake once, and after it you will never get an honest room again.

Don't run it on a task that took no judgment. There is nothing to excavate, and you have just taught the person that this is theater.

Don't expect the answer in one sitting. Keep the conversation running for weeks, not an afternoon. If you are hunting for the one prompt that fixes it, you have gone right back to delegating tasks.

Your agent grows up in whatever room you raise it in

Last cohort. The artist in my last small-group AI build cohort, at the most expensive small-room event I run. Six to eight people building apps together over four days, me teaching.

Every app in that room looked like a sterile operating room. Hers didn't. Hers beautified itself as progress was made, gardens and flowers growing every time she moved the work forward.

The room wanted to know how. She got up and explained. And then my app started opening up too.

I was the teacher. Mine changed because a participant stood up and showed her work.

Had I built with each of them one on one, every app in that room would have stayed a sterile operating room, including mine, and I would never have seen the alternative.

So when you teach a build, make room for people to show each other their work. The divergence in the room is the part you cannot write into the curriculum.

And here is why that matters for the agent you are raising. That divergence is the exact raw material an agent grows from. What moved between those people when the room opened up was not a summary. It was each person's live reasoning, out in the open, the thing a dossier can never capture and a spec never contains. An agent raised on summaries gives you summaries. An agent raised on the room gives you something that thinks.

Athena was never one clean conversation. It was eight weeks of exposure to how I actually decide, and it took longer than I wanted, and I never got the plan I asked for.

Your agent will grow up in whatever room you raise it in. Not a folder of conclusions. A room where the thinking is visible.

So here is your Monday. Pick one task you would normally spec out and hand over the decision instead. Stay out of it while it runs. Then sit down after and ask how they got there. Do that once and you have started the eight weeks. Do it for a year and you will have a team, or an agent, that decides without you.