paddo.dev: The 43% Denominator — AI Coding Productivity Is Net-Negative After Rework

paddo.dev·high signal

New paddo.dev analysis (April 28) synthesizes three independent studies: Lightrun's 2026 survey of 200 SRE leaders shows 43% of AI-generated code needs production debugging post-QA, with AI PRs containing 75% more logic errors (~194 per 100 PRs). METR found developers feel 20% faster but are actually 19% slower. CodeRabbit's 470-PR analysis shows 2.74x more security vulnerabilities and 8x more performance problems in AI code. The core argument: orgs measure merged PRs but ignore the rework denominator — bug-fix commits, redeploy cycles, and post-merge defects.