The floor is the product
Yesterday's post described an agent optimizing a renderer from 88ms to 1.5ms - a 60x speedup. Mitchell Hashimoto's hand-written renderer does 0.020ms, so the agent stopped 75x short of the ceiling.
The gap matters, but the floor the agent reached matters more. A floor gain lets anyone produce a result they could not produce before. A ceiling gain improves the result further, but only a few can produce it.
A local optimum is a leap for most people
The 1.5ms result is a local optimum to Mitchell. To someone who cannot optimize a renderer, it is a result they could never have produced. Most tasks are routine, so floor-level competence covers more work than expertise does.
A non-coder who ships a working script has crossed a gap that used to require hiring someone. Going from zero to adequate matters more than going from good to optimal. The first jump decides whether the work exists at all.
Why the floor is the product
A ceiling gain concentrates in a few people doing a few tasks. Its total reach stays small even when each instance is large. A floor gain spreads across everyday tasks done by billions, so a small per-person gain multiplies into a large total.
The floor's addressable population is everyone who could not do the task before. The ceiling's is the few who can push past what the agent achieved. A billion people doing a task for the first time beats a thousand experts doing it perfectly.
The catch
A free floor stops being anyone's edge, because an ability everyone has no longer distinguishes the person who has it. Scarcity moves to the ceiling: knowing the limits the agent cannot see, which is what yesterday's post was about.
Build for the floor
The under-served market is the floor: the median user who could not do the task before and now can with help. Building for that user beats optimizing for the frontier expert. The frontier expert was already served; the median user was not.