Good taste: the last defensible edge in the AI era

It has been reported that in a recent blog post Raj Nandan argues that as AI and large language models make competent output cheap, "taste" has emerged as the primary competitive moat for builders and teams. A landing page, a slide deck, a product memo — all can now be assembled in minutes. So what separates the forgettable from the memorable? Judgment. Plain and simple.

What changed

LLMs are excellent at stitching together patterns. They compress and recombine language, layouts, and tropes faster than any junior hire. The result: a crowded middle. Think polished-but-blurry product copy, interchangeable onboarding flows, modern-but-forgettable visuals. Nandan calls it a “7 out of 10” world — average has become abundant, and that abundance erodes the separation that used to reward basic competence.

Why taste — and more — matters

Taste, the post argues, isn’t fashion or status. It’s the ability to notice what matters, to reject what’s merely plausible, and to explain precisely why something misses the mark. But taste alone is not the full answer. If teams stop at selection — picking the best of ten AI-generated drafts — they risk becoming curators, not builders. The rarer, higher-value skills are contextual judgment, the discipline to impose constraints, and the courage to refuse complacent, safe outputs.

AI also functions as a mirror: it exposes how clear—or fuzzy—your own judgment really is. So the practical takeaway is simple: don’t train people to generate. Train them to diagnose, to dissent, and to build the awkward, specific things that AI tends not to produce. In a landscape where first drafts are free, the premium goes to those willing to do the hard, unglamorous work of making something distinct. Who’s ready to pay that price?

Sources: rajnandan.com, Hacker News