Why we built @convex the way we did is embarrassingly simple. What humans find hard is also what agents find hard. Race conditions across services, large codebases, and simple architecture that holds up over time. We built for the hard parts. Agents get better at the rest.