Good Taste the Only Real Moat Left
Short summary
As AI and LLMs make competent drafts trivial, the real edge shifts to human judgment—spotting what’s generic, diagnosing flaws, and owning outcomes. True value comes from combining taste with context, constraints, and stakes, not just selecting polished AI outputs.
- AI makes generic "7/10" drafts nearly free, so the scarce skill becomes diagnosing precisely why something is generic or risky, not just picking the best AI output
- The practical drill: generate 10-20 AI versions of an artifact, write a "fails because..." line for each, then rewrite the best under hard constraints (no buzzwords, one idea per sentence, real trade-offs noted)
- Refusal—recognizing when "good enough" hides risk or mismatches real needs—matters more than selection
- Real authorship requires holding actual stakes (regulatory liability, brand damage, budget/material constraints) that AI can't bear, so curation alone leaves humans without genuine ownership
Full summary
When competent output becomes almost free, the edge shifts to good judgment. AI and large language models can spin up landing pages, memos, pitch decks—any polished first draft—in minutes. That flood of “good enough” work means standing out now depends on knowing exactly what feels generic, spotting hidden trade-offs, and explaining why something fails. It’s not enough to pick the best AI variant; you need a clear, diagnostic vocabulary—phrases like “this sounds like every other SaaS homepage” or “this glosses over regulatory risk”—to lift your critique above mere taste.
Taste here means distinction under uncertainty. You won’t get a spreadsheet telling you which headline converts or which feature pays back its engineering cost. Instead, taste shows up in what you notice, what you reject, and how precisely you diagnose flaws. AI models excel at pattern-compression: they recombine common structures into polished but forgettable results. Left unchecked, they crowd the “7 out of 10” middle, making average abundant instead of rare.
That abundance pushes value downstream. The scarce skill is refusal: knowing when “good enough” hides too much risk, erases real constraints, or mismatches user needs. A practical drill looks like this: pick one key artifact—a paragraph, a dashboard label—ask an LLM for 10–20 versions, then write a “fails because…” line for each. Rewrite the strongest candidate under hard rules—no buzzwords, one idea per sentence, a real trade-off noted—and ship it. Over time, you get faster at spotting empty specificity and borrowed tone.
Yet taste alone won’t build anything novel. If human roles shrink to selecting from AI outputs, people become curators without real stake. Authentic work arises from co-creation under constraint—arguing with budgets, materials, timelines and owning the fallout. Models can generate options; they can’t hold regulatory liability, endure brand damage, or imagine truly new ideas that look wrong at first. That combination of judgment plus real authorship remains uniquely human.