Taste Is All That's Left
As large language models make it cheap to generate code, prose and interfaces, many engineers argue that “taste” — judgment about what’s worth building and how it should feel and behave — is becoming the main human differentiator. Commenters debate whether this is comforting or delusional: some see taste, experience and long-term maintainability as the new moat in a world of rapid cloning and “vibe-coded” software, while others counter that markets reward “good enough” slop and that AI itself will eventually learn average human taste. Underneath are broader worries about enshittification, declining craftsmanship, and whether reliance on AI will erode the very muscles — judgment, rigor, creativity — that produced good software in the first place.
What “taste” is and why it matters
- Many see “taste” as compressed, tacit judgment built from friction, practice, and joy in a craft.
- It shows up in choices about architecture, UX, readability, and long‑term maintainability, not just visuals.
- Others worry “taste” is vague, subjective, and easily used as cope or gatekeeping; similar ideas appear in aesthetics, philosophy, and “quality” discussions.
LLMs, coding, and convergence toward the average
- Some fear LLMs will normalize an “average taste” and reduce diversity in styles, languages, and architectures.
- Others argue most software already reflects average taste; people with strong taste will either avoid LLMs or aggressively steer them.
- There’s a recurring distinction between “vibe coding” (accepting output without scrutiny) and “AI‑assisted engineering” (using models under human judgment).
Quality, “AI slop,” and maintainability
- Multiple comments report AI‑generated code and docs as verbose, generic, and hard to reason about at scale, especially over months of evolution.
- Concern that short‑term “it runs” hides long‑term costs: tech debt, security holes, performance cliffs discovered only after real load.
- Some argue LLMs are already used effectively for tests, refactoring, and glue code, with humans supplying taste and review.
Market and economic dynamics
- Debate over whether taste is actually rewarded: many users and businesses accept buggy, mediocre software if it’s cheap and familiar.
- View that commoditization of “making” shifts scarcity from creation to adoption and discernment; others counter that capital, distribution, and social context matter more than taste.
- Question whether “taste” can be a moat if competitors can quickly copy features, UX patterns, and even overall style.
Simplicity, friction, and craft
- Several defend “simple” as the hardest thing to achieve: simple to reason about and maintain, not just simple to implement.
- Friction and toil are seen as the “curriculum” that develops judgment; concern that skipping hard work via LLMs will atrophy these muscles for most people.
Meta: AI authorship and detection
- Large subthread argues over whether the article itself is LLM‑written; detectors and stylistic cues are invoked on both sides.
- Some find the prose obviously “AI‑slop”; others insist it has human rhythm and dismiss detectors as unreliable.
- This dispute becomes a proxy for broader anxiety about distinguishing human from machine writing.