AI Coding Surge Shifts Bottleneck from Writing Code to Reviewing It

A study involving over 100 tech leads, architects, and senior engineers found that the rise of AI coding agents has significantly increased pull request sizes, overwhelming human reviewers with cognitive load rather than just time constraints. Interviews conducted over four months revealed that reviewers consistently prioritize high-risk code — such as data models, APIs, and authentication logic — before skimming lower-stakes boilerplate. Standard code review interfaces like GitHub were flagged for presenting all files with equal weight in alphabetical order, forcing reviewers to reconstruct reasoning file by file without context. To address this, the team built a 'change map' that assigns criticality tiers to each changed file, customized per reviewer rather than per repository. The findings underscore that human attention remains a fixed resource, and tooling must adapt to how engineers actually prioritize reviews rather than how code happens to be structured.