The promise of AI-assisted coding has always been seductive: ship faster, fewer errors, more productivity. But new data from Faros AI and Google’s DORA research is telling a different story. As teams rush to adopt AI-powered development pipelines, the incident-to-PR ratio has surged 243% and bugs per developer are up 54%.
The finding comes from a broad study of enterprise AI adoption, and it’s raising hard questions about whether the industry has been building software factories or just high-speed bug factories.
The Productivity Paradox
The numbers paint a confusing picture. Per-developer task throughput is up 33.7% and PR merge rates have jumped 16.2%. On the surface, AI is delivering on its promise of speed. But dig deeper and the cracks show.
“The barrier to writing functional code has effectively collapsed,” writes industry analyst Luca Rossi in his widely-read analysis of the software factory trend. “But that changes the bottleneck from ‘Can someone write code?’ to ‘Should this be written?’”
The issue isn’t that AI writes bad code—it’s that it writes code at industrial scale. A single engineer can now generate more code in a day than a team could produce a decade ago. When that code contains bugs, the sheer volume multiplies the blast radius.
Codebase Mutation
The problem extends beyond raw bug counts. Researchers observing enterprise codebases have noticed a new phenomenon: rapid “mutation” of coding styles within months, a process that previously took years.
“Codebases developed five to six different styles within months,” noted one fractional data head who requested anonymity. “Between multiple engineers trying to move quickly and a lack of standards, these projects became unruly.”
The pattern echoes what happened a decade ago with self-service tooling—early productivity gains that masked downstream complexity. AI doesn’t just accelerate good code; it accelerates every decision, including the bad ones.
The Platform Gap
The solution isn’t to slow down, according to experts. Instead, organizations need to think platform-first: defining how work moves through the system and how code is generated, reviewed, tested, traced, deployed, and improved when something goes wrong.
Otherwise, they’re just putting another one-off machine into an empty room and calling it a factory.
The industry is taking notice. Several major enterprises are now requiring AI-generated code to pass through additional review stages before merge—a practice that would have been considered anti-productivity a year ago.
Whether this marks a course correction or a slowdown in AI adoption remains to be seen. But one thing is clear: speed without discipline is just fast technical debt.