AI removed the friction that once taught programmers taste
For most of software history, the hard part was making something exist at all: the distance between an idea and a working program was filled with hours, sometimes weeks, of typing, misreading manuals, and slowly correcting a wrong first attempt. The essay argues that distance has now collapsed. AI tools let a person describe a program and receive a plausible version of it faster than they could type the first function by hand.
The essay's central claim is that this collapse did not remove value from the craft, it moved the value elsewhere. When production was expensive, the expense acted as an invisible filter: nothing shipped unless it survived the cost of being made, so bad ideas rarely reached the world in bulk. Once that floor drops out, the deciding factor is no longer what something costs to make but a person's judgment about which version, among many equally plausible ones, is actually right. The author calls this judgment taste and treats it as the only part of engineering that was never mechanical.
The essay borrows Robert Pirsig's account of Quality from a book Pirsig built around the word while deliberately refusing to define it, on the theory that a definition would kill it; Pirsig argued that people recognize Quality before they can explain it, that the recognition arrives first and the reasons, if they arrive at all, come later. The essay treats taste the same way: a compressed, wordless verdict reached faster than it can be justified.
The essay's most pointed claim is that taste is downstream of friction: it is built only by making something bad, being forced to live with its failure, and letting that sting sit until the mistake gets filed away. On this account, someone who starts out generating fluently with AI tools from day one never goes through that apprenticeship. They will be measurably more productive at the same career stage than an unassisted predecessor, but they will not have climbed the walls that used to teach judgment, because a competent version is handed to them for free from the start.
The essay argues this creates an economic problem for anyone who does have taste: they still ship at the same rate as someone who does not, because taste is slow, it sends the plausible thing back and asks for the right thing, and the market times both people with the same stopwatch without being able to see the difference. It invokes Harry Frankfurt's distinction between a liar, who respects the truth enough to work against it, and a bullshitter, who is simply indifferent to it either way, and calls low-effort AI output the bullshit of engineering: not wrong, just indifferent, and now the most abundant substance in the field.
To argue that the underlying ratio of good to bad work has not changed, the essay cites Sturgeon's dictum that ninety percent of everything is crap, made decades ago in defense of science fiction against a critic. Its point is that this ninety percent used to be throttled at the source because making it cost something: a bad novel still took a year, a bad program still took a month. With that cost removed, the same ninety percent of an effectively infinite output is still infinite, so the scarce act stops being making and becomes choosing what, out of that ocean, deserves to exist.
The essay draws a parallel to the Industrial Revolution, when factories first made mass, cheap, uniform production possible: it says a handful of people, Morris and Ruskin among them, looked at the flood of identical manufactured goods and asked not whether something could be made but whether it should be made, and made that way, by no one, for no other reason than that a machine could. It says Morris and Ruskin lost the economic argument of their time but were right that once making becomes free, choosing becomes the craft.
The essay concludes that the tools did not devalue the underlying skill, they stripped away everything that was not the skill: what used to look like the work, the typing and the wiring, turns out to have been the toll paid for the right to exercise judgment, and with that toll near zero, judgment is what remains exposed. Its closing argument calls for refusing the good-enough plausible output anyway, not out of nostalgia for the old friction, but because the verdict is described as the one part of the work that stays unmeasurable, unautomatable, and therefore genuinely the maker's own.
The piece closes with a postscript responding to readers who told the author, in comments, that the essay itself reads like AI-generated writing. The author denies that any large language model wrote, storyboarded, or reviewed the post, and says the short sentences, the reversals and the one-word lines are something the author writes only sometimes, not usually, and that happens to resemble machine output because both were shaped by reading the same kind of essays.
Key facts
- The essay argues that when producing software was expensive, that cost acted as an invisible filter keeping low-quality work from shipping in bulk; now that AI has made production cheap, judgment ("taste") is the only remaining filter.
- It defines taste through Robert Pirsig's concept of Quality, from a book built around a word Pirsig deliberately left undefined, arguing that recognition of quality precedes the ability to explain it.
- It claims taste is "downstream of friction," built only by making bad work, living with its failure, and learning from the sting, a process AI-assisted beginners now skip entirely.
- It cites Sturgeon's law that "ninety percent of everything is crap" and argues AI removed the cost that used to throttle that ninety percent at the source, so it is now free noise that rises without limit.
- It compares the moment to the Industrial Revolution, when Morris and Ruskin asked not whether factories could mass-produce goods but whether they should, arguing that once making becomes free, choosing becomes the craft.
Why it matters
The essay reframes an ongoing debate about whether AI coding tools produce reliable output; it argues the interesting question is not reliability but what happens once the cost of production drops to near zero. That shift removes a filter that used to be invisible: the expense of making something guaranteed that whatever shipped had survived a minimum bar, and losing that filter changes what actually separates good engineers from bad ones, from the ability to produce something to the ability to judge what is worth producing.
Who it affects
The argument is aimed at software engineers, particularly ones who start their careers using AI generation tools from day one, versus engineers who built judgment the slow way, by shipping bad versions and living with the consequences before such tools existed. The essay suggests the newer group will be measurably more productive but may never develop the same taste, because they are never forced to sit with a failure long enough for it to teach them anything.
How to use it
There is no product here, only a stance the essay urges readers to take: keep rejecting the good-enough, plausible version even when nothing external rewards the extra effort. That means sending a functional but not-quite-right result back for another pass rather than shipping it the moment it passes.
How solid is it
This is a personal essay, not a study or a data-driven argument. It rests on philosophical and literary references: Pirsig's Quality, Frankfurt's liar-versus-bullshitter distinction, Sturgeon's law, and the objections of Morris and Ruskin to industrial mass production, rather than on measurements of code quality or of what AI-assisted engineers actually learn over time. The essay names no specific AI product, model, or company as the cause of the shift it describes, referring only to "the tools" and "the machine" in general terms.
Risks and caveats
The author appended a postscript acknowledging that some readers accused the essay itself of reading like AI-generated slop, given its short sentences, reversals, and one-word lines. The author denies that any large language model wrote, storyboarded, or reviewed the piece, while conceding the criticism is fair on its face. The essay's claim about what beginners will or will not learn from AI-assisted work is presented as the author's own view, not backed by data on outcomes.
“Taste did not become less valuable. It became the only thing that was ever scarce. We just could not see it, because it was buried under all the labour it used to take to get to it.”
— the essay's author (no name given in the source)