Two ways of misjudging the size of a thing

27 Jun, 2026

The towers from the garden steps in Pacific Fair, Gold Coast.

Two small observations have been rattling around in me this week, picked up from two unrelated quarters of my life. I want to resist the reflex to hurry them into a single tidy law. They may be cousins; they may just be two things that happened to the same person in the same seven days. I will show you both and leave the welding alone.

One. The asking got cheap

There is a rule I picked up years ago, off the internet somewhere, and then ran for the better part of a decade: if a task takes only a couple of minutes, like washing the one cup or sending the one reply, do it now. Don't log it, don't schedule it, don't dignify it with a place on your to-do list. Just do it on the spot. It is clean, frictionless counsel, and for a long time it served me honestly. I had picked it up as a three-minute rule, which I now know was my own loosening of it; the version with a name runs to two minutes, from David Allen's Getting Things Done, and James Clear later borrowed those same two minutes for Atomic Habits, turning a rule for clearing small chores into one for starting small habits.¹

The rule rests on an assumption I never thought to check. It treats the time a task takes to start as a fair guess at the time it takes to finish. Reply to the email and it is sent; for a small task the starting and the finishing sit almost on top of each other, so the guess holds and the rule works.

Then I began routing a good deal of my work through Claude, and the proxy quietly came apart.

I write software for a living, and more of it each month now happens beside a model. A task lands and, in the privacy of my head, it is nothing but a prompt, a few sentences, a small act of asking. It reads as two minutes, because the only part I can see, the asking, really is two minutes. So the rule fires. Do it now. And I do. And because the asking costs me almost nothing, I seldom stop at one. I fire off the next before the first has come back, and another behind that, until three or four are in the air at once. That, more than the work itself, is what has tripped me up, more than a few times this year. There is some cold comfort in not being alone in it: when researchers actually timed experienced developers on real tasks from their own backlogs, the developers came out about a fifth slower with the AI tools, and were sure they had come out faster.²

And the model is good, and that is the trouble. It hands back something that runs: a function that turns the test green, an answer that compiles on the first attempt, a paragraph that says very nearly what I meant. The asking took a minute, the result looks finished, and by the old arithmetic the matter is closed. Nothing I worked with before could hand me a thing this finished-looking this fast.

But the prompt was never the work. The prompt is the cheapest fraction of it. The model will carry the first seventy per cent of a job in one pass, the scaffolding and the obvious patterns; the last thirty, the edge cases and the seams where the new thing has to join everything already running, costs about what it always did.³ What follows the asking is the reading, the checking, the slow judgement of whether I would put my name to what came back, and beneath that, whether I have any real grasp of why it works. Hand-work has a natural floor of effort, and, just as important, a ceiling. Steering a model knocks the ceiling out. The two-minute job swells into twenty, and no alarm ever sounds, because the asking stayed short from the first keystroke to the last.

The rule never had to price this part, because before the tool it never came up. When I built a thing by hand, finishing it and understanding it were one act; you could not reach the end without having walked every step, so comprehension came bundled in, paid for whether I wanted it or not. A model unbundles them. It will hand me the finished thing without the walk, and I can take it, ship it, and move on, only to find a week later that it works and I cannot quite say how. The code is done; I am not. Last week I worried the same knot from the other side, how easily a confident answer passes for understanding. This is not only my own weakness. In one study, people set to write essays with and without a chatbot were wired up as they worked, and the assisted writers showed the faintest mental engagement and the weakest sense of owning what they had written, many of them unable to quote back a line they had handed in minutes before.⁴ The understanding that used to come for free now has to be bought, and I am the only one who can pay for it, slowly, by hand, with my own attention.

There is a second trap folded inside the first. Because each asking is so cheap, nothing stops me from starting the next one, and the next, until I am steering eight or ten of these loops at once, every one quick to open and slow to close. The expensive part, though, does not run in parallel. Verifying, judging, understanding: these are paid for in attention, and attention is stubbornly serial, able to sit with one thing at a time. The open loops do not get faster for running together. They only get shallower, each one half-read, half-trusted, half-understood. So the real discipline is the dull one I keep having to relearn: take a single thread, carry it all the way to the point where I would vouch for it, and only then open the next. Doing one thing at a time is the only speed at which understanding happens at all.

So the rule had been measuring the wrong end of the thing. Time-to-start was once a serviceable stand-in for time-to-finish, and the tool severed the link between them. The truer question now is: how long is the loop in which I have to verify what came back, and understand it, before I will stand behind it? If I cannot accept the result without reading it closely, it was never a two-minute job, however curt the instruction that set it going.

The asking got cheap. The understanding never did.

Two. The empty boat

The second observation came out of a bad week, and from inside my own skull rather than from any tool.

I had made a mistake. Minor, recoverable, the sort that gets absorbed and quietly handled. And yet for days afterward the room ran cold. Every idle moment arrived like a verdict. I caught myself plotting an exit, casting about for somewhere that did not feel like this, an old reflex of mine. Then I put a plain question to myself: was anyone actually treating me any differently, or was I only flinching from a judgement that no one was issuing?

It was the second.

No one was standing there. I had assembled a figure out of my own dread and then recoiled from the thing I had built. The hostility was not incoming. It was outgoing, manufactured in me, aimed first at myself, then flung outward onto the world so I would not have to keep holding it.⁵

Epictetus put it plainly, two thousand years ago: men are disturbed not by things, but by their judgements about things.⁶ The mistake itself did not disturb me. What disturbed me was a sentence I imagined being passed on me, one that in fact no one had pronounced. And the mechanism, caught in the act, is almost a piece of clockwork. You err. Shame announces that you are not enough. The feeling is too scalding to hold barehanded, so the mind does the one thing that brings relief: it relocates the verdict outward. I think I failed becomes they think I failed. But the temperature had changed in me, not in the room. Psychologists even have a name for the inflation at work: the spotlight effect, our standing habit of overestimating how closely others are watching and weighing us.⁷

There is a parable in the Zhuangzi I keep returning to. A man is poling a boat across a river in heavy fog. Another craft looms out of the mist and strikes his. He wheels round, ready to curse the careless boatman, and the boat is empty. It had merely slipped its mooring and drifted down on the current. There is no one to shout at, and so his anger, finding nowhere to land, drains away.⁸

The boat that struck me was empty. The judgement I had braced against had no one at the oars.

What helps, when I catch myself mid-flinch, is to name the thing flatly, I am bracing for a judgement that is not being made, which is enough to move me from inside the feeling to standing just beside it. And not to withdraw, because going quiet is the one move that does make me seem off, and so summons the very reaction I feared. And to let the resolution be the end of it. Rumination feels like responsibility, like I am still working the problem. I am not. It is only interest, paid and re-paid, on a debt already settled.

Two, not one

Writing them out, I feel the old pull to tie the two together and call them one idea, that I keep mistaking the cheap front of a thing for the real weight behind it. It is a tempting symmetry, and I half believe it. But I have a weakness for the neat bow, and a week is rarely one idea in two costumes; it is just a week, with several unrelated things loose in it. So I will let them stand as they are. Two small misjudgements, a few days apart. If they prove to be cousins, you saw them before I forced the resemblance.

See you next week.

If you would like the next one to land in your inbox, you can subscribe here. One letter a week, and easy to leave the moment it stops earning the space.

#ai #attention #judgement #zhuangzi