What Happens When You Hand a Half-Formed Idea to a Fleet

Most build stories are told as a straight line. You have the idea, you make the plan, you execute the plan. The story is interesting in proportion to how neatly the ending matches the beginning.

I want to describe one that went sideways on purpose, because the sideways part turned out to be the method.

I had an idea I could not describe. Not a secret one — an unformed one. I could tell you the feeling of it and not the shape. Every instinct I have says that is the moment to sit down and force it into a sentence, and every time I have obeyed that instinct I have produced a sentence that was tidy and slightly wrong, and then spent a month building the tidy wrong thing.

So I did the opposite. I broke the hunch into corners and handed the corners out.

One helper went to find how the thing talks. Another went to find what it looks like. A third went to scout the ground — where this fits, where it doesn’t, who is already standing there. None of them had the whole picture, because there was no whole picture to have. Each one had a corner and permission to wander in it.

I should say plainly what a helper is here, because the word does work it hasn’t earned. A helper is a fresh conversation working out of its own folder. Not a person on payroll. Not one enormous thread trying to hold the entire project in its head and slowly losing the plot. A folder, a corner, and a hunch.

For a while the idea lived nowhere. That is the part I want to defend, because it is the part that feels like failure while it is happening. There was no document with the answer in it. There was no name. If you had asked me that week what I was building I would have given you three different answers depending on which folder I had open last, and none of them would have been a lie.

Then the corners started agreeing with each other.

The signal

Two independent corners, arriving at the same place, having never spoken.

The voice work and the look work and the lay-of-the-land work began pointing the same direction, and they did it without me steering. That is a specific and checkable event, and it is worth naming precisely: two of them landed on the same answer without being told to. Not “it started to feel right.” Two independent corners, arriving at the same place, having never spoken.

That is the signal to consolidate. Not a deadline. Not the day your patience runs out. The moment the form shows up in more than one place at once.

Only then did I pull it into a single home — one name, one site, a calm front page that warms you up with the offer sitting quietly behind it, its own writing, and a small fleet of helpers I can keep handing work to.

I said “a small fleet” rather than a number, and I would like that to be the least impressive sentence in this essay. There is a version of this piece where I tell you the headcount and it sounds like a company. I am not going to write that version. The restraint is not modesty; it is the only thing that makes the rest of it worth reading. If I inflate the part you can’t check, you have no reason to believe the part you can.

Which brings me to what the phrase actually means. “Folders over agents” sounds like an argument for tidiness, and for a long time I half-believed that myself. It isn’t. Tidiness is what you do to a finished thing. This is the opposite: you give each kind of work its own room precisely so that the work can go somewhere you didn’t plan, without knocking over the work next to it. The folders aren’t filing. They’re elbow room.

Here is the move, and it is smaller than it sounds.

Give each kind of work its own folder. A home for the writing. A home for the look. A home for the plan. Not to be neat — so each one has space to explore without stepping on the others.

Hand out hunches, not briefs. A brief is what you write when you already know the answer and want it executed. A hunch is what you have before that, and it is the more useful thing to hand over, because a corner-worker chasing a hunch will bring back something you didn’t ask for. That is the entire point.

Consolidate when the shape arrives, not before. The test is the one above: two corners agreeing without being steered. Until that happens, forcing a through-line is just you writing the tidy wrong sentence again with extra steps.

Refuse to inflate any of it. Say what you have. “A small fleet.” “It found a hold.” Not “we scaled.”

I have been quietly building things on the web for a long time, so the wiring came easily to me, and I want to get out ahead of the obvious objection: the wiring is not the lesson. If it were, this would be a tutorial and it would be useless to anyone whose stack looks different from mine. The lesson is the shape. Spread out to explore. Hand off honestly. Consolidate on evidence rather than on impatience.

That shape scales all the way down.

It works for a team and it works for one person on a slow Tuesday who is willing not to know the answer yet — which is, I think, the actual skill being described here. Not delegation. Tolerance for a period where the thing you are making does not have a name.

So, a real question rather than a rhetorical one: if you handed a half-formed idea to a fleet tomorrow, what is the first folder you would open, or the first helper you would send out to explore?

(The folder shape itself is written up in The Operator’s Handbook. Nightwatch and Claimwise, two working ones, are free and forkable.)