The Last Human Job Is Judgment

Two earlier posts on this site turn out to be the same post. The Ego Left the Codebase watched pride migrate from the lines to the taste. Everyone Said No First argued a rejection reflects the rejecter’s judgment, not the idea’s merit. Different domains, same shape: when the making gets cheap, the deciding is what’s left. The pattern kept showing up until it stopped looking like coincidence, so this is the essay that names it.

Run the thesis through every domain this site has touched and it’s the same mechanism each time.

Code. Execution is abundant now. A model writes the endpoint faster than a team, and roughly as well for the intern as for the architect. The constraint moved to knowing what to build, which of five working versions to keep, and recognizing slop when it compiles. The way I work now, one person’s judgment amplified by abundant execution, is this thesis practiced daily.

Capital. Opinions about companies are abundant and costless. Every investor has one, and the no costs its holder a calendar slot. Conviction, judgment you pay to hold, stays scarce, and telling it from delusion is the founder’s actual job. The money was never the scarce input. The willingness to be accountably wrong was.

Media. Content is going abundant, video last of all, and the business is repricing around what can’t be generated: attention, trust, and curation. The editor’s judgment outlives the editor’s toolchain.

And intelligence itself. As raw capability commoditizes across model providers, every one of them selling roughly the same tokens at collapsing prices, the premium moves to knowing what to ask, what to accept, and what to want. The judgment layer sits above every model and ships with none of them.

So the definitional question can’t stay soft: what is judgment, exactly? Not intelligence. The models have that, in the measurable senses, and more arriving quarterly. Not information, which is abundant to the point of being the problem. Not confidence, which is worthless as signal, since the deluded and the visionary report identical certainty. Judgment is accountable choice under uncertainty. Deciding with stakes, owning outcomes, updating on consequences. The accountability isn’t decoration. It may be the load-bearing part. A model can rank options. It cannot own one.

The steelman deserves its own section, because it’s the whole ballgame: maybe judgment is only the last human job so far. Every “AI can’t do X” claim has had a shelf life, and models already critique, rank, and choose plausibly. Why should this capability be the one that holds?

I know two honest responses, and neither fully reassures me. The first is the accountability argument: judgment without ownership of consequences is just ranking, and ownership is a social fact about persons, not a capability. We hold people accountable because they can be harmed, praised, fired, jailed. That holds right up until society decides to assign ownership to systems, which is a choice, not a law of nature, and choices get made badly all the time. The second is the regress argument: even a world of superb machine judgment needs someone to decide which judgments to delegate, and that deciding is itself judgment. That holds until it too gets delegated, at which point the question stops being economic and becomes something older.

I’m not going to resolve that here, because I can’t, and pretending otherwise would be exactly the confident forecast I distrust. The optimism I’ve argued for elsewhere, that we can craft our future rather than be subjected to it, enters here as a stance rather than a proof. Judgment remains ours as long as we insist on keeping it, and the insisting is itself the job.

Every prior technology ate a human task and left the deciding to us, and each time, the deciding got promoted to being called the real work. Farming, typesetting, spreadsheets, chess. Maybe judgment is genuinely different, the thing that was always underneath all of it. Or maybe this essay reads in ten years the way “computers will never play chess” reads now.

Judgment is the last human job today. We don’t get to know for how long. What we build in the meantime is the answer we’re giving.