Briefs · II · Ideatives Inc. · Updated
Stability drops as AI use rises.
According to Google's DORA 2024 report, every 25% increase in AI adoption is associated with an estimated 7.2% decrease in software delivery stability, and DORA 2025 (nearly 5,000 respondents) found the negative relationship persisted. This is a modeled correlation from survey data; it does not measure cause.
DORA 2024, Google Cloud · survey, modeled estimate
Does AI adoption hurt delivery stability?
In Google's DORA 2024 model, each 25% rise in AI adoption goes with an estimated 7.2% drop in delivery stability and a 1.5% drop in throughput, while documentation quality rises 7.5%. These are survey-based correlations.
Incidents and review
2025
“AI adoption does continue to have a negative relationship with software delivery stability.”
DORA 2025, Google Cloud · nearly 5,000 respondents, Sept 2025
3.4×
The chance of a production incident per merged change. Incidents per PR up 242.7%.
Faros AI, April 2026 · vendor telemetry, 22,000 developers
+31.3%
Pull requests merged with no review, human or agentic, from low to high AI adoption.
Faros AI, April 2026 · vendor telemetry
Output, review and incidents
More output
Epics per developer
+66% from each organization's lowest-AI to highest-AI period.
Faros AI, 2026
Bigger changes
PR size
+51% over the same comparison.
Faros AI, 2026
Less review
No-review merges
+31.3%. Cortex: “pressure to maintain velocity often leads to rushed reviews.”
Faros AI, 2026 · Cortex
More failures
Incidents per PR
More than tripled, up 242.7%.
Faros AI, 2026
Less stability
Delivery stability
−7.2% per 25% rise in AI adoption, modeled.
DORA 2024
Each step is correlated with the next; no study here shows that one causes another. Our view: shipping faster without review builds up debt, and the debt is paid in incidents.
What to do
- Gate AI-written PRs on size and require a review. Track no-review merges as a metric.
- Report change failure rate and incidents per PR alongside throughput.
- Spend the time AI saves on tests and smaller batches before adding scope.
Questions
- Does AI adoption reduce delivery stability?
- DORA's 2024 model estimates a 7.2% drop in delivery stability for each 25% rise in AI adoption. DORA 2025 (nearly 5,000 respondents) found the negative relationship persisted.
- Does AI increase incidents?
- Faros AI's 2026 telemetry shows incidents per PR up 242.7% (3.4×) in teams' highest- versus lowest-AI-use periods. It is vendor data and shows correlation only.
- Does AI improve delivery throughput?
- DORA 2024 estimated a 1.5% drop in throughput per 25% rise in AI adoption. DORA 2025 found throughput rose.
Sources
- DORA 2024, Google Cloud blog, October 2024. Survey; funded by Google, which sells AI tools. Sample size not confirmed.
- DORA 2025, Google Cloud blog, September 2025. Survey, nearly 5,000 respondents; funded by Google.
- Faros AI, Engineering Report 2026 “Acceleration Whiplash”, April 2026, and research page. Vendor telemetry, 22,000 developers, 4,000+ teams, two years. Full report gated. Faros says its data contradicts DORA 2025 on strong foundations. A newer Faros report (Q3 2026) was not reviewed.
- Cortex, 2026 Benchmark Report, November 2025. Vendor; 50 surveyed leaders plus metrics from an unstated number of organizations. Year over year, no control group: PRs per author +20%, incidents per PR +23.5%, change failure rate +30% (rollbacks only).
- METR, July 2025: 16 developers, 246 issues, 19% slower with AI. February 2026 update, 57 developers: −18% and −4%, intervals crossing zero; the authors call it “only very weak evidence”.