The AI Engineering Report 2026 dropped some uncomfortable numbers, and if you're running an engineering org that has gone all-in on AI coding tools, you need to read this carefully. Faros tracked two years of telemetry across 22,000 developers in more than 4,000 teams, measuring what actually happens when you flood a system built for human-paced development with machine-generated output. What they found has a name now: the Acceleration Whiplash. Throughput is up. So are bugs, incidents, and the hidden costs accumulating at every stage downstream.

The Productivity Numbers Look Great. They Aren't the Whole Story.

Let's be clear about what's real here—epics completed per developer are up 66%, task throughput climbed 33.7%, and PR merge rates increased 16.2%. If you're a VP of Engineering presenting to the board, these numbers look fantastic. But dig into the code churn metric and the picture gets ugly fast: under high AI adoption, the ratio of lines deleted to lines added has surged 861%—nearly ten times the prior rate. More code is being written, shipped, and then ripped out. That asterisk belongs on every headline about developer productivity gains. Throughput measures what was shipped, not what survived.

The Incident Rate Tripled. Nobody Seems Alarmed Enough.

Here's where this stops being an abstract metrics conversation: the incidents-to-PR ratio is up 242.7% as teams move from low to high AI adoption. For every pull request merged under heavy AI tooling, production failures are occurring at more than three times the rate relative to the baseline. Monthly incidents across finance, healthcare, infrastructure—you name it—are up 57.9%. The bugs-per-developer figure jumped from a 9% increase in last year's report to 54% now. This isn't flattening as organizations mature their AI programs; it's steepening.

Your Senior Engineers Are Drowning.

The most underappreciated finding is what Faros calls the senior engineer tax. AI-generated code is superficially convincing—it looks idiomatic, well-named, stylistically consistent. The structural and logical failures hide beneath the surface. Catching them requires slow, expensive cognitive work: reading carefully, reasoning about intent, reconstructing the problem being solved. Median time to first PR review is up 156.6%. Average time in code review has climbed 199.6%. Median time a PR spends in review status? Up 441.5%. The engineers with the deepest system knowledge are spending their most valuable hours untangling plausible-looking garbage that should never have reached them.

Strong Foundations Don't Protect You—Two Years of Data Proves It.

The DORA 2025 report concluded, based on survey data, that strong engineering foundations amplify AI's benefits and offer protection against its downsides. Two years of telemetry across thousands of teams tells a different story. High-performing organizations with mature DevOps practices and disciplined delivery processes are experiencing the same downstream deterioration as everyone else. Surveys capture how developers feel—right now they feel productive because, individually, they are. What surveys can't show is review queues backing up, incidents clustering, bugs reaching customers that never should have passed inspection.

More Code Is Reaching Production With Zero Review.

Pull requests merged without any review—human or agentic—are up 31.3%. This isn't likely a deliberate bypass of oversight; the more plausible explanation is that reviewers simply can't keep pace with AI-generated volume. The result is production systems receiving code with no scrutiny at meaningfully higher rates than before high adoption. Combined with the incident data, this defines the core risk.

Key Takeaways

  • 80% of teams now exceed 50% weekly active user threshold for AI tools; acceptance rate hit 60%, up from 20%
  • Code churn surged 861%—AI ships fast and wrong at scale
  • Incidents per PR more than tripled (up 242.7%); monthly incidents up 57.9%
  • Bugs per developer jumped to +54% this year, nearly six times last year's rate
  • Median time in code review increased 441.5%; senior engineers are burning out
  • Zero-review PRs up 31.3%, meaning more broken code reaching production unchecked

The Bottom Line

Every org cutting engineering headcount based on AI output gains should have this report tattooed to their forehead—the work required to ensure that output is safe, correct, and maintainable hasn't decreased; it's exploded. And the engineers being considered for cuts are often exactly the ones absorbing the quality gap AI is creating.