I run a two-product SaaS alone, around $14k MRR. Six months ago I was the constraint on output. Now I'm the constraint on judgment. Last sprint I let an agent loop write a billing refactor over a weekend. It produced 2,400 lines across 31 files, all tests green. I spent four days reading it before merging. Found three subtle bugs: a webhook idempotency key that worked but reused a field that customers can edit, a retry path that double-charged in one race, and a migration that locked a table I forgot was hot. None of these would fail a test suite I could realistically write. They failed against a model of the system I carry in my head. The productivity story I see other indies post is the generation half. Nobody posts the review half because it's boring and slow and the number isn't impressive. My honest throughput is maybe 1.8x what it was, not 10x. The cap isn't typing. It's how fast I can hold the diff in my head without lying to myself that I have.