The AI Code Review Bottleneck: Escaping the Velocity Trap
summary
AI tools boost output but create bigger, messier PRs: median PR size +33%, lines per developer +76% and 91% longer review time while MER’s RCT shows developers using AI are 19% slower despite feeling 20% faster (a 39% perception gap). Security tests fail often — 45% overall, 72% in Java, 86% XSS, 88% log injection — and 80% of suggested dependencies have known issues. Demo shows Gemini CLI + Sonar MCP detecting AI code, enforcing stricter gates, and auto-remediating from the CLI.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:17AI Coding Adoption Is Accelerating Fast
- 0:50The Rise of AI-Assisted Pull Requests
- 1:21Google’s AI Development Ecosystem
- 2:27Bigger PRs, More Code, Longer Reviews
- 3:23The Perception Gap in AI Productivity
- 4:37Hidden Technical Debt Below the Surface
- 5:48Security Risks in AI-Generated Code
- 7:05Escaping the Velocity Trap with Assurance
- 8:12Sonar’s Role in Closing the Verification Gap
- 10:31Demo: Gemini CLI with Sonar MCP
- 16:06The Guide, Verify, Solve Framework
- 17:44Final Takeaway and Resources
related talks
43mWriting for Machines: How To Capture Your Project's "Vibe" and Survive AI Slop
44mSpeed Is Easy, Trust Is Hard: AI's Real Impact on Developer Productivity
01h 01m"Looks Good to Me": A Practical Guide to Handling AI-Generated Code
49mClean Code in the AI era
27m3 data-backed ideas to deter AI code quality problems
14mAI-Powered Code Review and Quality Assurance
26mYour pipeline is the bottleneck
34mAI and Code Quality: Building a Synergy with Human Intelligence