$cd ~/log
LOG.015

2026-11-07

Growth: From Knowing How to Make to Knowing How to Judge

Lately I've noticed a change: over the course of a day, I spend less and less time actually making things, and more and more time thinking.

A page, a first draft, a small tool: hand it to AI and you quickly have a version you can look at. I used to get stuck on not being able to make things. Now I rarely get stuck there. The new problem is that when the finished thing is sitting in front of me, I often can't say whether it's actually good, or whether it's worth continuing.

The hard part moved

In the past, most of the effort in doing something went into execution. Whether you could do it, and whether you could finish, almost determined the result. Judgment mattered then too, but it hid behind execution, because doing something once was expensive, and that alone pushed people to do less and do it carefully.

Now that execution costs less, that natural filter is gone too. Anything I think of, I can make a version of, so I end up with more and more things on my hands, but not necessarily a clearer direction. I slowly realized that cheaper execution doesn't make things simpler. The hard part has just moved from my hands to my head.

Choosing what to do

The first thing I felt was that choosing got harder.

With so much I could do, every idea looks feasible. But feasible and worthwhile are two different things. Before I start something now, I ask myself two more questions: if it works, who will use it, and does it solve a problem that actually exists; and does it have anything to do with where I really want to go.

Once I spent an evening making three versions of a page for an idea, going through several rounds of colors, copy and structure. Only when I was done did I realize I had never once thought about who would open it. All three versions looked decent, but comparing them was pointless, because the problem was never in the page.

If I can't answer those two questions, doing things faster is just running in place. When the direction isn't clear, speed doesn't help. It only means I find out later that I've gone off course.

What not to do

Harder than choosing what to do is deciding what not to do.

In the past, I often didn't do something because I couldn't, and that took no resolve. Now almost anything can be tried, so "not doing it" has become a decision I have to make on purpose. I find myself easily persuaded by "it's quick anyway": add this feature while I'm at it, mock up that idea to see how it looks. Each one is no trouble on its own, but together they cut my attention into pieces.

So I started keeping a "not for now" list. Ideas go on it first, and I look again a few days later. Many things that felt like musts at the time no longer seem that important a few days on. The few that remain are the ones I really want to do. The point of the list isn't to remember them; it's to free me from deciding on the spot, and to leave some time for judgment.

When to stop

Another thing that takes judgment is knowing when to stop.

AI can keep revising forever. Every round finds a little problem, and every round makes it a little better. After a while, I sometimes can't tell whether I'm making the thing better or just writing it a different way. When revisions cost nothing, it's hard to call a halt yourself.

My approach now is to picture "good enough" before I start: who is this for, and what level does it need to reach before it can go out. When it gets there, I stop, and leave the remaining problems until someone has actually used it. A single real reaction from putting it out there is usually more useful than ten more rounds of revisions in my own hands.

It works the other way too. Some things, halfway through, already show they're headed the wrong way. What should stop then is the whole thing, not the polishing of details. That step has never been easy for me. The time already invested makes it hard to let go, and I keep thinking a little more work will turn it around. But once execution is cheap, what I've already put in isn't that expensive either, and the cost of letting it go is much smaller than it used to be.

Telling good from bad

Most fundamental of all is judging whether a result is good.

What AI produces usually looks complete on the surface: clear structure, confident tone, nothing obviously wrong. Precisely because everything looks fine, telling which one is really good takes more effort. If I don't understand the subject itself, I can only go by feel and say "it's okay," and "it's okay" is the most dangerous judgment of all, because it lets something mediocre slip through.

For example, if I ask for two versions of the same opening, both read smoothly. In the past I'd pick whichever looked better at first glance. Now I make myself spell out the difference: which one gets to the point faster, which one sounds more like something I'd say, which one makes a reader want to keep going. As I talk it through, I often find that neither works, and the real problem is that I haven't figured out what this part is supposed to say.

So my rule for myself is that when I evaluate a result, I have to give specific reasons: where it's good, where it falls short, and what's missing compared with what I want. If I can't say, it means I don't understand it yet, and I need to go back to the problem itself instead of asking for a few more versions to pick from.

Where judgment comes from

Judgment can't be handed off the way execution can. It comes from understanding a thing, and from looking back after each time you finish.

I now deliberately set aside some time to review decisions I've made: why I chose this, why I dropped that, and where the result differed from what I expected. For more important decisions, I write down a sentence or two of reasoning and expectations beforehand, then compare afterward. If I don't write it down, it's easy to cover for my past self after the fact and believe I saw it coming all along. More and more of the execution goes to tools, but this part I can only do myself. The more I make, the more I need to look back; otherwise speed only brings more things, not a better eye.

More and more things can be made, but the things worth making haven't multiplied. I now spend more of my time telling the two apart.

$ share --x Share on X ↗

© 2026 CYBERCAESAR · cybercaesar.bond