Learning to Judge When You Used to Produce

There's a transition happening in a lot of jobs that nobody announced and few people have named: the work has shifted from making the thing to assessing the thing.

It's usually framed as a promotion in disguise, or as the good news buried in the AI story. Partly true. What's missing from that framing is that producing and judging are different skills, and being excellent at the first does not make you good at the second — it can actively get in the way.

Anyone who's moved from doing to managing knows this. The difference now is that it's happening to people who haven't changed role, and nobody has told them it's happening.

Why being good at producing gets in the way

You judge against how you'd have done it. The most common failure. A different-but-fine approach gets marked down for being different, and you spend your energy converting good work into your own style rather than finding what's actually wrong.

You over-index on craft. Producers notice craft, because craft is what they spent years on. Phrasing, structure, elegance. Those are visible and usually not what determines whether the work succeeds — which is more often whether it answers the right question.

You can't see the omission. Making something means holding the whole thing in your head, so you know what you left out. Judging something means you only see what's there, and the most consequential failure in AI output is the thing that isn't on the page.

You're tired in a new way. Producing has flow; judging doesn't. It's a sequence of discrete decisions with no absorption available, which is more draining per hour even though it looks lighter from outside.

The four things that transfer

Decide the standard before you look. The single highest-return habit. Two or three checkable properties, written before the output exists, turns judgement into comparison. Undecided, you're forming a view while reading something engineered to be persuasive.

Read for what's absent. Practise deliberately, because it's unnatural. After reading any output, ask what a knowledgeable person would have expected to see that isn't here. Omissions are where the expensive failures live and they leave no trace on the page.

Separate wrong from different. Two categories, and only one of them needs your attention. Wrong means it fails the standard. Different means it isn't how you'd have done it — and letting that through is the skill, not the compromise.

Batch it. Eight outputs in one sitting is markedly less draining than eight scattered through a day, because you pay the context-switch cost once and the standard stays loaded. This is the most effective practical change available.

Judge the process when the output looks fine

Once anything is recurring, output-level review has a ceiling. The higher-value question is whether the process still fits: is the brief still right, have the inputs changed, is the standard still what we'd set today?

This is what catches drift — the gradual failure where everything is being done correctly against a specification that stopped being correct. No amount of careful reading catches it, because each individual output passes.

Keep producing something

Not nostalgia, and not only for the recovery — though that's real. It's that judgement decays without production.

Your standard for good comes from having made good things, and it dulls if you stop. Keep one thing you make properly, by hand, on a regular basis. It doesn't have to be efficient. It's what keeps your assessment calibrated to something other than the last twenty outputs you saw.

Sinéad — the difference she was correcting

Sinéad is an editorial director who spent fifteen years writing. Her reviews of AI-assisted drafts were thorough and produced heavy revision.

Reviewing her own edits, she found most were style rather than substance — restructuring things into her preferred shape, changing phrasing that was already fine.

She now runs one pass against three criteria and stops. Her observation: she'd been doing the writing job on someone else's draft and calling it review, which was slower than writing and less satisfying than either.

Wojciech — the question he added at the end

Wojciech is a technical pre-sales lead. His AI-assisted proposals were well-reviewed and occasionally missing something a client raised later.

He added one step: after reading a proposal, before sending, write down what an experienced person would expect to see here that isn't here.

It takes two minutes and it works because it's a different reading mode — searching the absence rather than checking the presence. In six months it has caught four omissions, all of which had passed a careful read.

The one thing

Producing and judging are different skills, and excellence at the first brings specific liabilities to the second: judging against your own approach, over-weighting craft, and missing what isn't there.

Set the standard before you look, read for absence, separate wrong from different, batch the judging — and keep making one thing properly, because the standard you judge against came from producing and fades without it.