Faros AI study: AI coding tools tripled production incidents
What happened
Faros AI on June 26, 2026 published its annual AI Engineering Report 2026 — Acceleration Whiplash. The study analyzes 2 years of telemetry from 22,000 developers across 4,000 teams, showing AI coding tools (Claude Code, Cursor, Copilot) increase output but at the cost of dramatically higher error rates.
Context and impact
The report lands as enterprise teams ramp up agentic coding spend (Uber's cap is $1,500/employee/month). The findings undermine ROI narratives — more code equals more problems, especially without governance.
Details
- Bugs per developer +54% (vs +9% in 2025)
- Production incidents per PR tripled
- Average PR size +51.3%, files touched per developer +149.9%
- Time in progress +225.2%, lead time commit-to-prod +480.4%
- Recommendations: tighter PR size control, earlier quality gates
Open original source
VentureBeat