← Journal
AI & Agents

The Feedback Loop That Turned Two-Week Tasks into Hour-Long Sessions

11 April 2026·6 min read

Now, I am going to tell you something peculiar. Pay attention, because it is the sort of thing that creeps up on a person very quietly, and by the time you notice it, you are already in rather deep.

It begins, as these things often do, with something that seems entirely wonderful.

The marvellous shrinking fortnight

Not so long ago, building a new feature for this little golf planning contraption of mine, a map with pins and routes, a page that generates its own preview image when you share it, the whole business, would have taken me a solid two weeks of evenings. The work was not especially difficult. Life just kept barging in. You would sit down on a Tuesday, make a bit of progress, and then not touch the thing again until Friday. By Friday, of course, you had entirely forgotten what Tuesday-you was thinking. So you would spend twenty minutes squinting at your own code like a detective examining someone else's crime scene, and by the time you remembered where you were, it was nearly bedtime.

This was annoying. But it was also, in a way I did not appreciate at the time, rather useful. More on that shortly.

Then along came the agents. And the two weeks became two hours. Sometimes less. You describe what you want, the agent builds it, you read through the result, and there it is, working, deployed, done. The gap between having an idea and holding the finished thing in your hands shrank from a fortnight to something closer to a lunch break.

Marvellous. Absolutely marvellous. Except.

The brain that will not sit down

I should mention here, and I mention it not as an excuse but as a relevant piece of the machinery, that my brain is slightly neurodivergent. It has always been the sort of brain that, when it finds something interesting, grabs hold of it the way a huntaway grabs hold of a job and absolutely refuses to let go. You can wave other things at it. You can shout. It does not care. It has found its muster and it is going to work it until dark.

Before the agents, the sheer slowness of building things acted as a kind of governor on the engine. You would hit a wall, a confusing bug, a build that took forever, a dependency that refused to cooperate, and the wall would force you to stop. The wall was maddening, but it was also a rest. A compulsory smoko, the kind the foreman calls whether you want it or not.

The walls are mostly gone now.

And so what happens is this: I finish a feature, and instead of feeling the pleasant tiredness of completion, I feel a pull. A tug toward the next thing. Nobody is asking for it. There is no deadline. The reward just comes so quickly now, describe, build, review, done, that little fizz of satisfaction, that my brain immediately wants another go. Like a kid who has just come off the flying fox at the playground and is already sprinting back around to the top of the hill.

And another go after that.

And another.

The curious matter of the guilt circuit

Now here is the truly sneaky part, the part that would have made Willy Wonka himself raise an eyebrow.

I used to play StarCraft. Hours and hours of it. The macro cycle, expanding your base, scouting the map, queueing up units, microing your army across the fog of war. Same loop, really: challenge, action, reward, repeat. You get that little dopamine hit every time you hold a push or your timing attack lands perfectly. But with StarCraft there was always a little voice. You know the one. It would pipe up around the four hour mark and say, in the mildly disapproving tone of your mum calling from the kitchen, "Shouldn't you be doing something useful?"

That voice does not fire here. Because this looks like useful work. I am shipping features. I am building a real product. I am being, by any reasonable measure, productive. And so the little guilt alarm that would normally tap me on the shoulder and suggest I go to bed stays completely silent, because as far as it can tell, I am being terribly responsible.

It is almost exactly like a StarCraft session, except you feel less guilty because it is "productive work." And that, you see, is precisely what makes it so very difficult to stop.

I will look up from my laptop and discover it is one in the morning. I sat down at nine intending to do one small thing. The one small thing became three small things became a rather ambitious refactor that I definitely did not need to do tonight. And the whole time, the guilt circuit sat there with its feet up, perfectly content, because the work was real.

The question I have not been able to answer

How does one shut off?

The old answer was simple: you shut off because building things was slow enough to create natural stopping points. You would wait for a build. You would get stuck and wander off to put the jug on. You would run out of steam because writing code from scratch for four hours is genuinely tiring in a way that reviewing an agent's output for four hours is not. The friction was the fence around the paddock. It kept you in, but it also kept you from wandering off a bluff in the dark.

When the fence disappears, you have to build your own. And building your own fence requires a particular kind of discipline. Not the "work harder" kind, which I have plenty of, but the "stop working" kind, which I appear to have almost none of. The kind where you close the laptop even though the next feature is sitting right there in your head, fully formed, and you know perfectly well that an agent could have it finished in thirty minutes.

Thirty minutes. Just thirty minutes and it would be done.

You see the problem.

An open ending, because that is all I have

I am not writing this because I have found a solution. I am writing it because I reckon a good number of people using these tools are feeling something similar and haven't said so, because the conversation around AI productivity is almost entirely about how much more you can ship. Nobody seems to be asking what happens to a person's brain when the speed limit is removed and they were already the sort of person who had a fair bit of trouble slowing down.

Perhaps it levels out. Perhaps the novelty fades and the compulsiveness settles into something more sensible. Perhaps I simply need to set a timer and, for once in my life, actually obey it.

I do not know yet. I am still very much in the loop. Sprinting back up the hill, grabbing the flying fox, launching off, landing with a thump, and heading straight back up again. Grinning the whole time.

If you're in it too, I'd be keen to hear how you're managing.