AI Assistants Shift From Suggestions to Shipped Code
Review, not generation, has become the bottleneck on engineering teams using them heavily.
Engineering teams report that AI-drafted changes now make up a meaningful share of code reaching production, a step beyond the autocomplete role these tools played until recently.
The constraint has moved. Generating a plausible change is no longer the hard part; reviewing it is. Teams describe spending more time on code review than before, because a reviewer must reconstruct intent that was never in a human's head. Several have responded by requiring smaller changes and stricter test coverage rather than by limiting the tools.
Measured productivity gains are uneven, concentrated in routine work — migrations, test scaffolding, boilerplate — rather than in novel design.