Taste Is All That's Left: A Case for Judgement in an Age of AI Code

For most of the history of software, the hard part was making the thing exist: between an idea and a working program stood hours or weeks of typing, reading manuals and wrestling APIs. Production was the wall — and that wall is now gone, or rented out, writes the author of notashelf.dev in an essay titled "Taste Is All That's Left".

Good enough is a solvent

The familiar question — are the machines any good? — misses the point. The output is good enough, and "good enough" dissolves the reason to do better. While making things was expensive, cost enforced a floor; with the floor gone, what decides whether something deserves to be kept is no longer the cost of making it but the maker's judgement — the verdict you reach when you look at three plausible versions of the same function and know that two of them are wrong. The author calls it taste, a word engineering is slightly embarrassed to use.

Taste is built from friction

Taste is not decoration or preference, he insists: it is the compressed, wordless verdict reached faster than it can be justified. Nor is it learned by consuming good work — it is built the slow, humiliating way, by making something bad, watching it fail and filing the failure away. That friction was the curriculum. Newcomers who never sit with a bad version they were forced to learn from will look more productive by every metric — while unable to tell whether what they shipped should exist.

The economics are against you

Someone with taste ships at the same speed as someone without it, because taste says "no, again" and the market times both with the same stopwatch. Taste does not show up in the diff: it is unmeasurable, uncreditable and invisible on a dashboard, and the disasters it prevents leave no trace. Borrowing Harry Frankfurt's distinction between the liar and the bullshitter, the author calls slop the bullshit of engineering — not wrong, exactly, but indifferent. The scarce act is no longer making but choosing.

The verdict is the last part that is yours

Invoking Morris and Ruskin, who asked not whether factories could make everything but whether it should be made, the author argues the tools did not devalue the skill — they stripped away everything that was not the skill. The typing and wiring were a toll paid for the privilege of exercising judgement; the discipline that remains is in the deletion, in refusing the plausible version that works and asks only to be let through. The verdict cannot be measured or automated, so nobody can take it from you or sell it back. Everyone can make anything now; almost no one can tell you what is worth making.