Tech blogger NotAShelf published a new essay on Aug. 6 titled Taste Is All That’s Left, arguing that as AI strips away much of the old difficulty of writing software, the scarce skill is no longer coding itself. What still matters, he says, is taste: the judgment to tell what is merely usable from what deserves to exist.
The wall between an idea and working software has dropped
NotAShelf writes that programming used to come with a high wall. Turning an idea into a functioning program often meant hours, or even weeks, of manual testing, reading documentation, and fixing errors.
AI has changed that process. Now a developer can describe a feature and get back a version that appears workable within minutes, often faster than writing the first function by hand.
Better tools have not automatically produced more good products
The article asks why strong products do not seem to have become more common even as software generation gets faster. NotAShelf’s answer is that cost once acted as a filter. When making something was expensive, people were pushed to build only what was worth building.
With that cost now moving closer to zero, the filtering role that cost once played has weakened. What remains, in his view, is the builder’s own taste.
He says taste is earned through friction, not absorbed by exposure alone
On where taste comes from, NotAShelf argues that it is not something people get simply by consuming excellent work. Reading 100 strong codebases, he says, will not automatically produce judgment, just as eating at high-end restaurants will not turn someone into a chef.
He describes taste as something formed in a clumsy, uncomfortable way: by making bad things, being forced to live with them, watching them break, and carrying those failures forward in memory. Friction, in this telling, is not the obstacle to learning. It is the lesson itself. The walls people once complained about were also teaching them which walls were worth climbing.
AI lets beginners move smoothly, but may remove part of the learning path
NotAShelf says beginners now start with tools that can generate code smoothly from day one. They are no longer always forced to sit with the consequences of a broken version they built themselves, because the tool hands them a usable version for free. If the wall is gone, he argues, some of what people learn from climbing it disappears too.
His point is not just about speed. It is about what happens when errors, delays, and rework stop shaping the way judgment develops.
Engineers with taste and those without it can now ship at almost the same speed
He adds that this is the harsher part of the shift: engineers with taste and engineers without taste can now ship at nearly the same pace.
To explain that idea, NotAShelf borrows from philosopher Harry Frankfurt’s categories in On Bullshit. A bullshitter, he notes, differs from a liar because the liar still cares about whether something is true or false, while the bullshitter does not care. Under that framing, “usable” code produced with little friction and little judgment becomes the engineering version of that behavior: it may not be outright wrong, but it is not concerned with being right either.
When junk becomes cheap to make, good work gets harder to see
The essay closes with a broader observation. As the cost of producing junk falls close to zero, genuinely strong work has to swim through a vast ocean of output that looks similar on the surface and appears acceptable at a glance before it can be noticed.
The core claim of the piece is simple: AI has not made “being able to produce” the rare thing. It has made judgment over what should be produced more valuable.