Better to be
roughly right
than precisely wrong.

A vibe-coded demo looks like proof the idea has value and proof it can be built. Usually it's neither. Drug trials solved a version of this problem long ago: decide in advance what evidence each stage has to produce.

We know how to fix broken software, and we know how to supervise fallible people. AI takes the first playbook away and quietly breaks the second, because every control we've built for human error is a bet on where mistakes cluster.

A chart from a16z puts an agent at $80 an hour — roughly what an engineer costs, one choice away. But there are two other numbers on the same chart that nobody quotes: managed well, four dollars; managed badly, seven thousand. The four-dollar number requires a human to accept the work. The metric was built by dividing by you.

An LLM can now generate the work each role used to make by hand — the PRD, the mockup, the code — which makes it fair to ask whether the role is still needed. But producing the artifact and exercising the judgment behind it used to be one act. Pull them apart, and you find out which half was the job.

Usage-based AI billing put a per-person price tag on knowledge work. But it shows only one side of the ledger — and the reflex to cap the side you can see costs more than the spend.

Chat isn't AI's interface — it's AI's bottleneck. The best AI interfaces are defined by their output shape, not their input method. Five patterns that move past the text box.