AI coding agents reason over a context window and a pile of statistical priors. They thrive on conventional, explicit, locally-readable code and fail on clever indirection, metaprogramming, and hidden magic.
Cleverness is context you forgot to write down. Every implicit behavior the agent can't see in the window is a fact it has to guess — and guessing is where it goes wrong.
The codebase is the context. Boring, predictable, grep-able architecture isn't just kind to the next human — it directly ...