Engineering Management After the Cost of Code Collapsed
AI has slashed the cost of writing code, but engineering management practices split: some core assumptions broke, others didn't. Here's how to tell which.
AI coding tools have sharply cut the cost of producing code, but the broader claim that this makes engineering orgs dramatically faster remains unproven, and the idea that code review, documentation, or onboarding are now obsolete is simply wrong. The real test for any management practice is what assumption it rests on: if it depends on the cost of writing code, it deserves review; if it depends on human coordination, trust, or judgment, it stays valid no matter how dated it looks.
Verification now splits into two layers. Mechanical checks, tests, types, lint rules, contracts, are collapsing in cost because agents can run and fix them faster than any human reviewer. Semantic verification, whether code implements the policy the business actually needs, remains a human bottleneck, since sign-off is risk acceptance rather than information processing.
Metrics like velocity and pull-request counts were always weak proxies, tolerated because code was scarce. Now that code is cheap, they become actively misleading since volume is trivial to inflate. Meanwhile the junior engineer pipeline, historically built by doing the small tasks AI now absorbs, faces an unsolved problem whose effects won't be visible for years.