Code review is getting harder without getting more obvious about it. When developers wrote every line themselves, review was slower but the authorship was clear. Reviewers could ask the author what a function was supposed to do, trace the decision back to a commit message, or check the ticket it came from. That contract is breaking down as generated code enters repositories at higher velocity and volume.
The problems that come with it are concrete. Coding assistants produce output that passes syntax checks and often looks reasonable at a glance, but can introduce subtle logic errors, duplicate existing implementations, violate architectural constraints, or create security vulnerabilities that only surface at scale. The review process designed for human-paced contribution struggles to hold when the input rate increases sharply and the author is a model that doesn't explain its own choices.