Code review is burning out your best engineers
Every team I talk to has the same problem. Their best engineers, the ones who care most about code quality, The post Code review is burning out your best engineers appeared first on The New Stack .
The story of code review is that teams are struggling with the sheer volume and complexity of AI-generated code. Their best engineers, who deeply care about quality, are overwhelmed by review queues they simply cannot keep up with. Experienced engineers outright refuse to review AI-generated code, complaining about the challenges it presents.
A study revealed that 77% of engineers now spend less time writing code and more time reviewing AI output, shifting the job from crafting to verifying. Teams with high AI adoption see a 98% increase in merged pull requests, yet review times have skyrocketed by 91%. Engineers who embraced AI first, cared most about code quality, and built the review culture are now drowning in 15 PRs, each containing 400 lines of code.
Reviewing AI-generated code is harder because the reasoning behind the code is lost. The reviewer is left trying to reverse-engineer the intent from the diff, a fundamentally different cognitive task from reading code written by a colleague. AI-generated code passes the eye test, reading coherently and handling the happy path, but edge cases reveal misaligned assumptions.
These bugs are difficult to catch in review because they require understanding what the code was supposed to do, not just what it does. Over-engineered solutions are produced by AI models, trained on vast bodies of code that include enterprise patterns and production-hardened architectures. The models may produce 200-line abstraction layers for problems that really need 15 lines, include conventions that don't apply to the specific system, and generate confident but incorrect code such as calling non-existent APIs or using deprecated methods.
Cargo-cult patterns are also prevalent, where AI copies structures without understanding why. Engineers are spending their days reading machine-generated diffs, not crafting code. The answer is not to review harder or add an LLM reviewer. The solution lies in shifting the burden off reviewers in three ways: codifying repeated feedback, preserving the intent that produced the code, and measuring the work that actually prevents slop.
Pull your team's last 100 PR review comments and sort them into deterministic, execution-testable, or genuine judgment. Three-quarters of review feedback is codifiable. Create an AI slop registry to pull out these codifiable feedback items. Preserve the reasoning trail by capturing intent from prompts and agent sessions, and structuring it as acceptance criteria.
The reviewer should read a list of acceptance criteria and ask if the right problem is being solved with the right constraints. This high-value work for senior engineers is not reading a 400-line diff at 4 p.m. Codifying feedback, preserving intent, and measuring verification work shifts the review burden away from engineers and back towards building code.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
