Taste Is All That's Left

On the topic of taste, I think of this quote in Susan Sontag's "Notes On 'Camp'" often:

“For taste governs every free — as opposed to rote — human response. Nothing is more decisive. There is taste in people, visual taste, taste in emotion — and there is taste in acts, taste in morality. Intelligence, as well, is really a kind of taste: taste in ideas. (One of the facts to be reckoned with is that taste tends to develop very unevenly. It's rare that the same person has good visual taste and good taste in people and taste in ideas.)

Taste has no system and no proofs. But there is something like a logic of taste: the consistent sensibility which underlies and gives rise to a certain taste. A sensibility is almost, but not quite, ineffable. Any sensibility which can be crammed into the mold of a system, or handled with the rough tools of proof, is no longer a sensibility at all. It has hardened into an idea..."

I don’t like the term taste, but the problem that I have is that LLMs don’t seem to work “good enough”. They seem to be able to solve the immediate problem, but stacking this on the scale of 3-4 devs over 6 months or so doesn’t seem to produce anything.

One thing that I’m particularly frustrated with is the writing quality of LLMs. Like this is the thing that they should be able to do, but I would say almost everything they write has almost no signal.

Over a mid sized AI generated codebase that means I’m reading like 500 words to figure out what a module is even doing.

We've been talking about taste for a year now as it relates to AI, so I'm going to counterpoint it: if competitors can reproduce your features, UX, and visual decisions within days, your taste is not really an advantage. How many apps implemented pull down to refresh once one person did it? How many people even remember (or care) whose taste spawned that UX pattern? [1]

AI is not inaugurating an age of taste. It is shortening the half-life of taste and commoditizing software. What happens when software features, UX, and visual fidelity aren't competitive advantages anymore? If you ask me, it looks like business did in the 90s. We'll probably see the return of department store software in the next few years.

1: for the record https://en.wikipedia.org/wiki/Pull-to-refresh

Generating the first version is almost free now. Understanding why that version is subtly wrong, debugging it in production, maintaining it six months later and deciding whether the abstraction was a mistake are all still expensive...

This article resonated tremendously with me, speaking as someone who has been coding since the 1980s and had to develop my own taste the hard way, through lots and lots of mistakes. Whenever I see a software demo built by someone mostly using agents to do the work, I have to question whether it's any good on the inside - or whether any real intuition or judgment went into its design. On the flip side, maybe none of that matters anymore - as long as it works, who cares how it was built.

Hard disagree on "Good enough", most things are mediocre at best. Despite the massive AI boom I can at best count on one hand "fresh" products made with AI that are meaningful in day-to-day usage. Not to mention most of them seem to be targeting AI-space itself instead of bridging some actual business use case. Just look at all the products on ShowHN, you can "tell" a vibed product in seconds.

Also just a shout out to the author - you're getting a lot of flak here, but your article hit me at the right time in the right way. I trust you're telling the truth about writing it yourself, and thank you.

I've run this thought experiment a number of times with different sections of my (relatively) diverse friend circle. I ask them a version of this question on repeat: "what if AI automates X aspect of your work?", and the final answer (either because of frustration with the loopy nature of the question or a genuine stopping point) arrives somewhere along the lines of "my judgement," "my knack," "my embodied nature," or "my quality of decision making."

I think the core still leads back to human agency and the ability to consider aspects that wouldn't fit into an LLMs limited context window or be able to be vectorized into a DB, including ultra-long-term consequences (especially those with great thinking abilities).

I still happen to think that AI/LLMs will never be able to "fully" replace humans, because the evolutionary process that led to our cognitive abilities and the way we train LLMs are vastly different thanks to different pressures, but maybe that's just me defending the last bastions of our collective humanity as a human; I don't know what else to root for.

Taste seems something that should be more related to a craft but when doing engineering we don't talk about taste, we talk about theory, systems, practices, skill, intuition, experimentation, teaching and learning. So the debate on whether taste matters or not might come down to whether creating software can be called craft or engineering. In my opinion, maybe programming was more of a craft before LLMs but it looks like it is being squished between the engineering aspects of producing and maintaining software going ahead.