New data: AI’s impact on engineering velocity is more modest than expected
DX analysis of 400+ companies finds AI adoption up 65%, but PR throughput up just 8% at the median—well below the 3x to 10x gains industry leaders have been promised.
As AI coding tools become standard across engineering organizations, a gap has opened between what vendors promise and what the data shows. Claims of 3x–10x productivity gains have set expectations high, but most leaders are seeing far more modest results and wondering if they’re falling behind. Today, we released our AI and Engineering Velocity report, a longitudinal analysis of engineering velocity across 400+ companies that shares the data behind how AI is impacting developer productivity and what leaders can do about it.
The report combines telemetry data from DX’s platform with qualitative interviews from developers. By tracking TrueThroughput (DX’s proprietary measure of PR throughput) alongside AI tool adoption from November 2024 to February 2026, it provides a clearer, grounded picture of AI’s impact on engineering output.
Key findings from the report
AI adoption is up 65%. Throughput moved less than 8%. Across the sample, median PR throughput rose 7.76% despite a 65% increase in AI tool usage. The mean was higher at 13.1%, driven by a small number of outliers. Even at the 90th percentile, gains reached only ~44%. Most organizations fall in the 5–15% range. For a 500-engineer organization, a 10% gain is the equivalent output of 50 additional engineers without the headcount cost.
The bottleneck isn’t code generation. Coding represents only ~16% of a developer’s day. Developers described a “net-zero” dynamic: time saved writing code is consumed by increased review, validation, and remediation of AI-generated output. The other 84% of the SDLC is where the higher-impact opportunities are.
Developers report saving time, but it’s not showing up in output. Self-reported time savings are real, but they aren’t translating proportionally into throughput. Where that time is going is still an open question. The report surfaces contributing factors: skill gaps that steepen the learning curve, AI tools that can’t access institutional context, and team-level friction that prevents shared workflows from forming.
More output doesn’t mean more value. Higher velocity introduces measurable risks:
- AI-generated defects reaching production (Amazon’s early 2026 outages resulted in ~120,000 lost orders in a single incident)
- Cognitive debt — shared understanding of systems erodes as AI-generated code outpaces what teams can absorb
- False velocity — more PRs merged without a corresponding increase in business outcomes
The full report is available now. To learn more, schedule a demo or reach out to a DX representative.