AI Increased Our Open PRs by 36%. That Wasn't the Whole Story
Stack Builders saw a 36% increase in open pull requests after adopting AI tools, but code review velocity dropped and developers spent more time managing AI-generated code. The experience shows that raw productivity metrics alone don't capture the full impact on team workflow and code quality.
Background
- Stack Builders is a mid-size software consultancy that builds custom apps for clients. This article is a retrospective by their CTO and VP of Product on using AI coding tools (like GitHub Copilot and Cursor) inside real consulting projects.
- "Open PRs" (pull requests) are a measure of developer output — more open PRs suggests more work is being started. The 36% increase sounds good initially, but the article argues raw volume can be misleading.
- Key tension: AI lets developers produce code faster, but the same AI also creates more code that needs to be reviewed, tested, integrated, and maintained. The bottleneck shifts from writing code to reviewing and fixing it — "busy work" grows even when velocity seems up.
- Context for tech followers: there's a fierce debate right now between "AI accelerates everything" and "AI creates a hidden tax of lower quality." This piece lands firmly in the middle, offering practical lessons rather than hype or doom.