AI-Written Code Takes 5x Longer to Review, New Data Shows

A new report from code review platform Codacy finds that pull requests generated by AI agents take 5.3 times longer to clear review than those written by humans — and engineers say the delay is not a sign of thoroughness. Reviewers report that AI-authored PRs routinely lack the contextual explanation that makes human code readable, require more rounds of back-and-forth to validate, and bundle changes in ways that are difficult to parse at a glance.

The data punctures a narrative the AI developer tools industry has been careful to maintain: that AI coding assistants make engineering teams faster. They may write code faster. But if the bottleneck shifts from writing to reviewing, the net gain can turn negative for teams already stretched thin.

Part of the problem is structural. AI agents are optimized to produce syntactically correct code, not to communicate intent. Human reviewers are trained to evaluate changes from a fellow engineer — with explanations, commit messages, and context that gives meaning to the diff. That social layer of software development has no AI equivalent yet.

Engineering teams that have adopted AI coding tools at scale are quietly building new review frameworks to cope. The irony is that some are now evaluating AI-powered review tools to handle the AI-generated code they created to save time in the first place.

Read the full story at Codacy