Eleven ideas. Four started. Three dead, one paying the bills.
That's my ledger, and I'm not even the person this piece is mostly about. But I feel the same pull everyone else feels, and I've watched it closely enough on myself to recognize it out in the world.
For a long time I read those three abandoned starts as a discipline problem — a flaw to fix with better habits, or another productivity system to master. I've stopped reading it that way. The people this pattern visits are not undisciplined. They're some of the sharpest builders I know.
The demo that goes nowhere
You see it at any meetup. Someone opens a laptop. Built it last night. And it's good — real auth, a clean UI, works on the first try in front of a crowd. Three months ago this person had never touched that framework. Now there's a working thing on the screen, a live URL, and an honest smile.
Then you ask what they're doing with it, and the smile disappears. The honest answer is nothing. The repo went up last night. There's already a new idea queued for tonight.
The community chats tell the same story. The just shipped this drops, two per week, each one genuinely impressive, each one the last time that repo is ever mentioned.
A graveyard of small miracles.
Nobody's lying and nobody's lazy. The work is real. The ideas are real. Most of them just never become anything — and we've quietly agreed not to find that strange.
What friction was actually doing
Every one of those demos would have taken a small team a quarter not long ago. This isn't a failure story. It's a pattern story, and the pattern has a shape worth naming.
Building something used to be a single act. You felt the urge, you sat down, and you paid for it: weeks, sometimes months, of unglamorous and frequently ugly work. The urge and the labor were welded together. You couldn't have the first without surviving the second.
AI cut that weld.
Now the urge roams free. You can have the working thing tonight. We keep discussing this as though the only thing that changed is speed — as though we simply got faster at the same activity. We didn't. We took one act and split it into two, and we only talk about the half that feels good.
Here's what those weeks of work were actually doing, the part nobody misses out loud.
They were asking you a question, over and over, the whole way down: do you actually want this to exist? Every hard afternoon was the question again. Most ideas couldn't survive being asked that many times. They died somewhere in week two, quietly, and you never had to decide anything, because the cost decided for you.
The friction wasn't blocking the work. The friction was the filter. It killed the impulses that were only ever impulses, and whatever crawled out the far side was something you had genuinely chosen.
We think we removed an obstacle. What we removed was the filter.
Why "dopamine problem" misses the point
Now the impulse walks straight through to a finished artifact without once being asked whether it should. Nothing dies in week two, because there is no week two. Throwaway ideas and real ones get identical treatment — instant, frictionless, shipped — and they come out looking identical, because at the level of a working demo they are identical.
You can't tell from the artifact which one you actually wanted. The difference between them was never in the code. It only ever showed up in the part that came after: the distribution, the second week of users, the boring maintenance, the thousand small decisions AI still won't make for you. That's where a real thing and a passing mood finally diverge — and it's exactly the stretch we now skip, because there's always something new to build instead.
This is usually where someone invokes dopamine and reaches for the discipline argument. I don't think that's the right frame. The dopamine was always there. The first five percent of any project has always been the best part — nothing's broken yet, the whole thing exists perfect and complete in your head. That hit isn't new. What's new is that nothing stands between you and the next one.
The hit used to be rationed by the cost of earning it. We didn't develop a dopamine problem. We removed the rationing.
Calling it a willpower failure misses the machinery. You're not weak. You're running an engine that used to have a governor on it, the governor is gone, and you're surprised it's burning your legs.
The loop in the closet
Then there are the people who looked at all of this and decided the answer was more throttle.
They're building autonomous loops now — pick a goal, let the model work, feed the output back in, again, again, all night, no hands on the keyboard. The names come from mythology. They run while you sleep. Somewhere there's a VPS humming in a region you've never visited, generating applications nobody asked for and nobody will open.
The pitch, when there is one, is that the moat is tokens per second. Throughput.
Sit with that for a moment, because it's genuinely strange. Generation was never the scarce resource. A frontier model runs at something like fifty tokens per second, already faster than you can read. The bottleneck was always you: your attention, your taste, the slow human act of looking at a thing and deciding whether it's any good. The response to a flood of unexamined output has been to engineer a faster flood, and then leave the room.
But who's reading now? Nobody. That's the whole point of the loop.
They took the two acts AI split apart — the deciding and the doing — and automated away the deciding, not by getting better at it, but by removing the human who was supposed to do it. The loop in the closet is the logical end of the pattern: a machine built specifically so that no one ever has to watch the impulse again. The opposite of patience, manufactured at scale.
Where the filter has to live now
Which tells you exactly where the filter has to go. It can't live in the cost — we deleted that. It can't live in a loop whose entire purpose is to never look. There's one place left. It has to move inside you, on purpose, or it doesn't exist at all.
The method for this is old. Far older than software. It isn't a productivity system. It's closer to a practice.
The impulse arrives. You feel the full pull of it — the certainty that this one is different, this one you have to start tonight, before the feeling fades. And you don't move. You don't open the editor. You let the impulse sit in front of you, and you watch it, and you do nothing.
Most impulses don't like being watched. That's the entire trick. An impulse demanding you act before you could think it through the night was telling you something about itself. Give it a week of being noticed and not obeyed, and it evaporates on its own — the same way it used to evaporate in week two of a build, except now it costs a week of nothing instead of a week of work.
The ones still standing after that, still pulling, still there when you wake up — those aren't impulses anymore. Those are decisions. A decision is an impulse that survived being watched.
The file, the 11 days, and the ones I'm relieved I never started
The pragmatic version of this is almost embarrassingly plain.
I keep a file. An idea isn't allowed to touch an editor until it's lived there for a week. No exceptions, especially not for the ones that feel urgent — urgency is the tell.
Most don't make it out of the file. The one that now pays my bills sat in there for 11 days before I wrote a line, and I resented every one of them. The others that felt every bit as urgent in the moment are still in the file, and I'm relieved I never started them. The ideas that survive the wait are better, and — the odd part — I actually finish them. By the time I begin, the deciding is already done. The work is just the work, which was the part I wanted in the first place.
The more considered version is the same sentence read slowly. There's a gap between the impulse and the hand that obeys it. There always has been. The cost of building used to fill that gap so completely that we never had to look at it. Now it's empty, open, and faintly uncomfortable. That discomfort is the point. It's the space where you get to be the one who chooses, rather than the one who reacts.
A loop in a closet is what it looks like to pave that gap over and call the paving a moat.
The tools got good enough to grant every wish the instant you have it. That leaves exactly one thing standing between you and a hundred half-built monuments to passing moods: whether you can sit with wanting something and not immediately reach for it.
Everything else can be automated now. That can't. And the people trying hardest to automate it are the clearest proof of what it's worth.
I don't always manage it. But I notice now. And noticing, it turns out, was most of what the friction was ever doing for me.
