Technical blogger NotAShelf published an article titled “Taste Is All That's Left” on Aug. 6, arguing that after AI drove the cost of producing software close to zero, the quality-control role once played by “cost” disappeared: when things were expensive, you made only what was worth making; now that costs are nearly zero, the only thing left to act as a gatekeeper is personal “taste”—the judgment to distinguish between what is merely “usable” and what is “worth existing.”
Core Argument of “Taste Is All That's Left”
NotAShelf points out that programming once had a high wall: between an idea and working software stood hours or even weeks of manual testing, documentation reading, and debugging. This wall also served as a filter—when things were expensive, you were forced to make only what was worth making.
Now AI lets you create a usable version within minutes of describing a feature, bringing production costs close to zero. Cost no longer acts as a gatekeeper; personal taste is the only remaining mechanism of quality control. This is precisely why good products have not become more common.
The Source of Taste: Forged Through Failure and Being Forced to Face Bad Work
One of NotAShelf’s core arguments is that taste cannot be learned by consuming excellent work, just as it is impossible to become a chef simply by eating at high-end restaurants. Taste must be honed through several conditions:
Building bad things: Not avoiding the poor-quality results you create yourself
Being forced to live with them: Watching them break and staying with the failure
Etching it into memory: Each failure makes you remember which walls are worth climbing
Friction is the lesson: “Friction has never been an obstacle to learning; friction itself is the lesson”
The lesson of climbing walls: Every wall you cursed at back then was teaching you to recognize what was worth paying the price for
The Beginner’s Dilemma in the AI Era: Fluent Generation Never Fails, So It Cannot Learn What the Walls Taught
NotAShelf points out that beginners today can generate working software fluently from day one and are never forced to live with a broken version of their own work, because these tools freely provide a “usable” version. The walls are gone, and so is what climbing them used to teach.
More cruelly, engineers with taste and engineers without taste currently ship at almost the same speed—the difference in taste is no longer visible in speed, but it still exists in whether the final result is “worth existing.”
Frankfurt’s “Bullshitter” Analogy: Not Caring About Truth or Falsity, Only Generating Usable Output
NotAShelf cites philosopher Harry Frankfurt’s distinction in On Bullshit: “liars” still care about whether something is true, while “bullshitters” do not care at all. He uses this to draw an analogy to “usable code” generated without friction or judgment—the engineering world’s equivalent of bullshit: it is not wrong, but it does not care whether it is right.
As the cost of producing junk has fallen close to zero, genuinely good work now has to stand out in an “ocean” of output that looks identical and seemingly acceptable.
FAQ
What is the core argument of NotAShelf’s article “Taste Is All That's Left”?
According to the article published by NotAShelf on Aug. 6, 2026, the core argument is that after AI drove the cost of producing software close to zero, the quality-control mechanism once provided by “cost” disappeared. The only scarce resource is “taste”—the judgment to distinguish between what is merely “usable” and what is “worth existing.”
How does NotAShelf believe taste is learned?
According to the article, taste cannot be learned by consuming excellent work; it must be honed by building bad things, being forced to face failure, and etching failure into memory. The article argues that “friction has never been an obstacle to learning; friction itself is the lesson.”
How does NotAShelf describe the impact of current AI tools on cultivating engineers’ taste?
According to the article, beginners today can generate code fluently from day one and are never forced to face versions they have made badly, so they cannot learn the judgment that “climbing walls” teaches. Engineers with and without taste currently ship at almost the same speed, and good work must stand out in an “ocean of seemingly acceptable” output.
