Specs become the bottleneck once code is cheap

When agents make code cheap, engineers stop waiting on each other and start waiting on well-formed specs. The bottleneck moves from writers to the people deciding what code should exist. Focus is saying no: every vibe-coded product with 12 features is 11 away from great.

Jevons paradox applies: cheaper code means more code, not less effort. Prototypes that were not worth building three months ago spin up in an afternoon. Internal tools that solve problems nobody quite had get built and forgotten. People can only absorb features at a certain rate, and that rate stays roughly the same whether the team ships ten features or fifty. The discipline of focus, of saying no to good ideas to protect the great ones, becomes harder, not easier, with agents.

“Cheap” here means cheap to generate, not cheap to own. Bad code is the most expensive it has ever been, because a codebase that is hard to change cannot absorb what agents produce — the bounty only lands in a changeable codebase¹. The Jevons effect and the maintenance cost point the same direction: agents flood a codebase with code, and only codebases built on deep, changeable modules (see: deep modules make a codebase explorable and testable for agents) turn that flood into leverage rather than entropy.