Alex Komoroske: Why gardeners beat builders in the AI era

AI lowers the cost of producing mediocre software and insight, but outputs are probabilistic, brittle, and non-deterministic. Design products and workflows that assume occasional failure, use AI to augment rather than replace judgment, and build around its cost structure instead of advertising-driven models that can’t clear inference costs.

Instead of over-planning and trying to force outcomes, plant many small, cheap seeds (features, experiments, ecosystems) that could compound if they catch. Then watch where real energy and pull emerge, water those 'acorns,' and let winners grow, accepting that you can’t know in advance which ones will become oak trees.

Your real options are the small, plausible moves right in front of you that will almost certainly work; stringing these together with a clear but low‑resolution North Star lets you reach radical outcomes with safe, incremental steps. Over-planning with faux precision in uncertain domains is mostly a comforting illusion.

In big companies, optimism, status reporting, and the rule 'don’t make your boss look dumb' create a widely shared fiction that compounds up the org chart. Recognizing this enables you to surface disconfirming evidence in ways that don’t shatter the system, choose cheaper bets, and avoid zombie strategies everyone privately knows won’t work.

As AI makes content and code easy to generate, what stands out is a distinctive, human point of view that’s clearly not 'what the LLM would have written.' You build that by following your curiosity, exploring weird edge cases (e.g., WebSim), and noticing which of your ideas genuinely resonate across very different people.