Thinking out loud · Product · Site only, for now

Clay first, chisel later

AI score 2.8/4 me. Idea 2/2 · Words 0.8/2. The idea, the examples and the argument are mine. Most of the sentences are AI-drafted from what I said, and I haven't rewritten them all yet

Start with a little more than you need, roughed out as a whole. Then cut. AI changed what that costs.

I built this site in about two days, and it has more in it than it should. A version of the first piece for my kids. A placeholder for a version for my mom. A score for how much of each piece is me versus AI. A changelog, a charter, an ad for airlines. A tug-of-war animation on the homepage.

That was on purpose. Mostly.

The idea

A sculptor working in clay doesn't start with the nose. They rough out the whole figure first, a little bigger than it needs to be, close enough that you can tell what it is. Then they carve back and refine.

That's how I've started working. At the beginning, I overstep the scope on purpose. I build a rough version of the whole thing, plus the edges around it: the connections to other things, the second reader, the parts I'm not sure belong. Then I chisel back.

It isn't throwing noodles at the wall to see what sticks. Every piece of the clay is there for a reason I can explain. It's a considered first draft of the whole, not a pile of guesses.

Standing on two good ideas

This isn't new, and credit is due. Alistair Cockburn's walking skeleton says to get a thin version of the whole system working end to end before you flesh out any one part. Kent Beck's make it work, make it right, make it fast says the same thing about order: don't polish what doesn't work yet. I love both.

Clay first shares their goal: see the whole thing early. Where it differs is the starting size. A skeleton is lean on purpose, with just enough to connect the ends. Clay first starts fat, and that's the point. The periphery is where I tend to miss things: the second reader, the thing it connects to, the question someone asks on day three. A lean start doesn't show you those. A fat one does, and then you get to decide what stays.

One more thing on credit. If this turns out to be someone else's idea that I read years ago and absorbed, I'd like to know. I've looked and haven't found it framed this way yet, but that's not the same as it not existing. I don't want to be the person who says they heard the band first, or who announces they just came up with "prime the pump." If someone got here before me, tell me. I'll credit them here, and happily walk back anything I claimed.

Why now

Lean starts made sense when building was the expensive part. If a rough version of a feature took a team a few weeks, building three you'd throw away was real waste.

AI changed that math, at least for the first rough version. I can get a considered draft of the whole thing, edges included, in an afternoon. When building gets cheap, missing something becomes the expensive mistake: the connection you didn't see, the reader you didn't think of, the piece you have to bolt on later because it wasn't in the shape.

So I'd rather start fat now. Not because lean was wrong. Because the cost of starting fat went down.

What chiseling looked like here

In two days on this site, the chisel took off:

• An AI-drafted version for my mom. I built it, read it, and pulled it. That one's personal, and I'm writing it myself.
• A comic for my kids about a castle and robots. Replaced by one about clothes and towels on the floor, because that's their actual life.
• The old services pages, a policy generator, a terms page. Gone.
• A line in the airline ad that read like I was looking for a job. Rewritten. I'm not. I want the front of the plane.

Each of those was worth building. I found out what was wrong by seeing it whole. The changelog has the rest, with the fast calls marked.

Where it goes wrong

Here's the honest catch. Building got cheaper. Attention didn't. Every feature on this site is something a reader has to get past and something I have to keep up. The danger with fat first is that you never pick up the chisel, because cutting something you built feels like losing it.

So the chisel needs a date. Mine is roughly four to six posts in. After that, it's one change at a time, a guess about what it'll do, and a check on whether it did. It's written into the charter, so I can't quietly skip it.

Where does this break?

What have you built too big and never cut? Or cut too early and wished you hadn't? I'd like to hear where this falls apart, especially from people who've lived with a walking skeleton for years.