A study conducted over 2 years, with 22,000 software engineers and more than 4,000 development teams, collecting telemetry data. Based on facts and data.
Productivity up 860%, production incidents up 245%, and developer-introduced defects up 55%.
Faros has just released one of the largest studies ever conducted on AI use in software engineering. The central concept presented in the report is the “Acceleration Whiplash”, where Artificial Intelligence has altered the software development flow, generating more code than humans can sustain at an acceptable quality pace. While raw delivery capacity has increased, errors, incidents, and costs in subsequent stages of the DevOps pipeline have also surged.
Kent Beck (creator of XP and TDD) recently cited this study, highlighting the increases AI is driving across the software industry
Key Findings
- 860% increase in Engineering Productivity
- 245% increase in production incidents (Fail Rate)
- 55% increase in bugs introduced into production (Defect Density)
- AI is the primary author of code: 80% of the teams analyzed use AI on a weekly basis, and the acceptance rate of AI-suggested code has risen from 20% to 60%.
- Deliveries are happening, and faster: 66% more epics delivered and 16.2% more integrated PRs. More features shipped and more lines of code in repositories.
- But “throughput” remains an unknown: 861% more code churn, meaning code is being removed at a rate 10 times higher than before.
- Production incidents have nearly tripled: 242.7% more incidents per PR, and monthly global incidents rose 57.9%.
- Defects are accelerating, not stabilizing: In 2025, defects rose just 9%; in 2026, they jumped 54%. As companies mature in their AI usage, the relationship between generated code and defect rate is becoming increasingly pronounced.
- Easy to start, hard to finish: 67.4% more daily PRs, and tasks sent back increased 13.8%. Additionally, tasks started but stalled for more than a week rose 26%.
- Senior engineers are more overloaded: The average time spent on code review increased 199.6%, while the median time jumped 441.5%.
- More code shipping without review: The rate of PRs merged without any review rose 31.3%. Engineers are unable to keep up with the volume being generated by AI.
- Strong foundations don’t protect organizations: This study reveals that even companies with mature DevOps practices and strong DORA metrics suffer the same workflow deterioration. Developers feel individually more productive, but the system as a whole accumulates bottlenecks in the stages that follow code generation and integration.
- Companies downsizing teams based on AI gains should read this report: While the volume of code production has increased, the effort required to ensure that code is safe, correct, and sustainable has grown substantially.
Who will fix your AI-generated code down the line? Probably your best Senior Engineer — if they’re still around.
Sources
