Filling the gaps

For a while now I've been prompting and reviewing code over remote access — through Grok, through Claude running on the Mac mini via Grok, and straight from my phone to Claude on the Mac mini. What used to be reels, posts, and tweets got replaced by something else: write, wait, look, fix, again.

At the start I felt very excited. Then I noticed the day was narrowing. In every spare gap I was trying to make something, put it in front of someone, then feel a little relief. For a long time I mistook that for deep work. I was really just running the same loop.

The problem isn't bad output. It's output that arrives almost ready. The app comes up but a button is off. Half the tests are green. I wrote the prompt, so it feels like one more round is owed. My Cyprus days came back to me: in the casinos, the table isn't trying to make you rich. It's trying to keep you there for one more round. Nobody there calls that work. Here, people call it work for themselves.

Habits don't break on willpower. Atomic Habits had an example that fits this exactly: change the environment, don't trust willpower, it said. As long as the environment keeps calling you back, “I'll write fewer today” doesn't do much. I'm not the weak link. The loop is solid. The metric isn't the jackpot. It's time spent there.

Knowing that doesn't mean the loop only has a bad side. After a long stretch away, what pulled me back into building was the same thing. This time the output was a demo, a prototype, something that ran. Same mechanism. Different job.

Because this is a build era. Intelligence is getting cheap. The tools are in everyone's hands. While the big players write their own rules, small teams can still ship. Working at this pace here isn't only falling into a trap. It's walking onto a field whose gate hasn't closed yet.

My claim is this: the loop is dangerous because it holds you. The loop is valuable because it brings you back to making, after a long time away from it. I have to hold both at once.

Everyone can spin up a prototype with the same tools. A product that reaches nobody is a job left half-done. If I spend the loop only on “one more feature,” I stay stuck inside. If I turn the same energy toward distribution, the work opens outward.

There's a lot to do:

That's why I spend time on local AI. Control. Not dependency — insurance. Same knot on the rope since Rhizio: don't get locked in, own your own head.

I'm not quitting. This isn't an age without tools. But I'm keeping the line clear.

I tie the reward to an outcome verified outside, not to how many rounds I ran: a user, a payment, a decision. I don't count work stuck at the threshold as success. I name why I'm picking up the phone. I put hours where the tool is off. I keep authority revocable. And I deliberately put part of the energy toward distribution: for every “one more feature” round, one “reach one more person” round.

The work isn't measuring how long you stayed on the screen. It's doing the thing while knowing what you're doing, how long you're doing it, and who it reaches — and opening that work outward, to real demand.