Teams adopting coding assistants frequently report a growing review queue rather than faster delivery. The bottleneck moved rather than disappearing.
Production and review capacity scale differently
An assistant increases how much code an individual can write. Review capacity is unchanged, because reviewing still requires a person reading and understanding a change.
Any system where one stage speeds up and the next does not accumulates work between them. The queue is a straightforward consequence of that imbalance.
Teams often notice the symptom before the cause, because the writing improvement is felt individually while the queue grows collectively.
Generated changes tend to be larger
When writing is cheap, the natural size of a change grows. An author who would have made a minimal fix now includes the surrounding refactor because it took no additional effort.
Review effort scales worse than linearly with change size. A large diff is harder to hold in mind and more likely to hide an unrelated modification.
Larger changes also invite superficial approval, which converts a review backlog into a defect backlog further downstream.
Authors know their own code less well
Reviewers depend on the author being able to explain intent. An author who accepted a suggestion may not have a strong account of why a particular approach was taken.
The review conversation then becomes an investigation rather than a discussion. That is slower and less pleasant, and it discourages reviewers from picking up queued work.
Familiarity cues stop working
Experienced reviewers use style, structure and idiom as signals about how carefully something was written. Those cues are weakened when much of the code follows a common generated style.
Reviewers lose the ability to skim confidently and must read more evenly. Uniform effort across a large volume is exactly what a limited review budget cannot sustain.
Some teams compensate with tooling that flags high-risk regions, restoring a basis for allocating attention unevenly.
Effective responses limit input rather than add reviewers
Teams that resolved the backlog generally did so by constraining change size, requiring authors to summarize intent, and splitting work before submission.
Adding reviewers helps least, because review quality depends on context that new reviewers lack. Throughput rises while the value of the review falls.
The durable lesson is that assistants shift where the constraint sits, and a team that does not move its process with it converts a speed gain into a queue.