ai didn't make me faster. it made me throw away more.

Everyone wants AI to be the thing that made me faster. It's the clean story. Robot writes the boring parts, human ships twice as much, line goes up. People ask about it like that, expecting a number. How much faster. And the honest answer disappoints them every time, so let me just say the true thing first and explain it after.

AI didn't make me faster. It made me throw away more.

That sounds like a downgrade. It isn't. It's the whole point, and almost nobody who sells you AI tooling will say it out loud, because "you'll generate three times as much garbage" is a hard line to put on a pricing page.

Here's the mechanism, the actual physics of it. The cost of building the wrong thing used to be high. Not money-high — time-high, and ego-high. If a layout took me a day, I'd lived inside that day. I'd argued for it in my head. I'd defended it before anyone else got to see it. So when it turned out wrong — too clever, too cold, technically fine and emotionally dead — I didn't kill it. I fixed it. I nursed it. I shoved good hours after bad because the bad hours were already spent and I couldn't stand to watch them die. Everyone does this. It's called being precious, and it's the most expensive habit in the trade.

Now an iteration costs almost nothing. I can have the wrong version standing in front of me in twenty minutes. And a strange thing happens when the wrong version is cheap: you stop defending it. You look at it cold. You go, no, that's not it, and you delete the whole thing without a funeral, because there was nothing to bury. You didn't bleed for it. You can build the second wrong version before lunch. And the third.

the edge is in the killing

Three wrong versions before lunch isn't a productivity flex. It's the opposite of one. Two of those three are dead by 1pm. What I bought wasn't speed — it was cheap death. The ability to be wrong out loud, fast, on the screen instead of in my head, and to find out it's wrong by looking at the real thing instead of arguing about the idea.

Because here's what I actually believe, the position underneath all of it: you cannot think your way to the right version. You can only recognize it once it's in front of you. The brain is a terrible judge of an idea and a pretty good judge of a thing. Describe a homepage to me and I'll nod. Build it and put it on a black screen at 1am and I'll know in four seconds whether it's alive. The whole job is getting from the idea to the thing as many times as possible, as cheaply as possible, so my eyes can do the part my brain was never any good at.

AI collapsed the distance between idea and thing. That's it. That's the entire gift. Not "it writes my code." It lets me be wrong eleven times on the way to right, where I used to get two tries and had to make the second one count.

So yes — most of what I generate is garbage. I want to be specific about this because the people selling the dream are allergic to it. I throw away more than I keep. Not 10% more. Most of it. The ratio of dead drafts to shipped pixels has gone up, not down, and that is not a failure state. That is the machine working exactly as intended. The garbage isn't the bug. The garbage is the search. Every dead version told me something the alive one couldn't have told me on its own.

People hear "throws most of it away" and assume waste. It's the inverse. The waste was the old way — pouring real, irreplaceable hours into the first plausible direction because backing out felt like admitting something. The new way wastes electricity and saves the only thing I can't make more of, which is the part of me that has good taste at 3am and a finite number of those nights left.

what this looks like in practice

It looks quiet, honestly. Small builds. Two people. A site goes out in a couple of weeks and from the outside it looks like we just sat down and knew exactly what to make. We didn't. Nobody does. What actually happened is we made the wrong one four times where the old version of this work would've made it once and shipped it tired.

The two-weeks number isn't the brag. The four-dead-versions-you-never-saw is the brag. The polish you do see is just the last one standing after the others got killed early, before they could cost anything, before anyone got attached.

That's the part I'd defend to anyone. The speed is incidental. The killing is the craft. Learn to kill things while they're cheap and you're free in a way no amount of raw output will ever make you. Refuse to, and the machine just helps you fall in love with bad directions faster.

I'd rather throw away more. It means I'm still looking.