How AI Code is Forcing a Reckoning with Traditional Code Review

The surge in AI-assisted coding has fundamentally shifted how teams manage code quality. With 90% of developers now using AI tools at work, the volume of pull requests has exploded—developers are merging 98% more PRs than in the pre-AI era. But this explosion has come at a cost: bugs per developer have jumped 54%, and incidents per pull request have surged 243%.

The root problem is stark: traditional code review—built on the assumption that humans could meaningfully audit changes—is breaking under the weight of AI-generated code. A single agent-created pull request can span tens of thousands of lines, making it nearly impossible for a human reviewer to trace the reasoning behind the changes or understand all the downstream effects.

The Case Against Human-Centric Review#

John Bristowe, Principal Developer Advocate at Octopus Deploy, argues that code review has become more theatrical performance than genuine quality check. No reviewer can reasonably comprehend a 40,000-line change authored by an AI system they didn't oversee. The mismatch between the speed of AI code generation and the capability of humans to review it has effectively turned review into a bottleneck that doesn't actually catch problems.

This dynamic compounds an existing issue: teams have rushed to adopt AI coding tools while lagging in deploying automated review mechanisms, simply pushing the human bottleneck further down the pipeline without solving it.

An Alternative: Policy-Driven Pipelines#

Rather than expecting humans to review everything, Bristowe proposes shifting quality gates into the deployment pipeline itself. The strategy centers on encoding organizational policies directly into infrastructure—establishing rules that flag code violating defined standards, rather than attempting to audit every change. Humans then focus only on handling exceptions, freeing them from the impossible task of comprehensive review.

This approach doesn't necessarily require more AI. Instead, it substitutes judgment-based review with systematic, policy-based verification that applies consistently regardless of who or what wrote the code.

The Debate Continues#

Not everyone agrees. Platform engineering expert Viktor Farcic will challenge this framing at a live event scheduled for Tuesday, September 29, questioning whether policy-as-code can truly absorb the kind of nuanced judgment that code review demands, and whether it simply shifts accountability rather than improving quality.