Faros AI

Sources#

What it is#

Faros AI is an engineering-intelligence / software-delivery-analytics vendor. Its platform aggregates telemetry across the software development lifecycle — task-management systems, IDEs, static code analysis, CI/CD pipelines, version control, incident management, and HR metadata — and standardizes it per company to produce delivery and quality metrics. The platform is the data instrument behind its research: the 2026 report draws on ~two years of telemetry from 22,000 developers across 4,000 teams.

What it published#

AI Engineering Impact Report (July 2025) — coined the AI Productivity Paradox: heavy AI-tool investment, but expected delivery gains not materializing. Dataset: 10,000+ developers, 1,255 teams.
AI Engineering Report 2026: The Acceleration Whiplash — the paradox "sharpened into a crisis." See Acceleration Whiplash, AI as Primary Author, Telemetry vs. Survey Measurement. Analysis as of March 2026.
AI Engineering Report Q3 2026: The Speed Trap (blog summary 2026-09-18, full report form-gated) — the sequel to the Whiplash report on the same 22,000-developer / 4,000-team panel, over the most recent 12 months. Its thesis is that the initial shock is easing while the strain moves downstream: incidents per PR +14.5% (from +242.7%), code deletion ratio +71.6% (from +861%), deployments/week +13.8% (from −11.7%) — against QA time +300.6%, monthly incidents +125.4%, restarts +66.7%, PR size +71.8%, and an unreviewed-merge metric growing +76.3%. See Acceleration Whiplash for the full treatment.

The 2025 and 2026 reports are independent cross-sections, not a longitudinal panel — comparisons between them are directional only.

The Q3 2026 report changes the comparison in a way that must be carried with every figure. Where the Whiplash report compared each company's low-adoption quarters to its high-adoption quarters, the Speed Trap reports only period-over-period deltas between its own dataset and the Whiplash dataset, with both windows drawn from already-high-adoption teams. There is no AI-vs-non-AI cohort in it, no published definition of an "AI-assisted" change, no sampling description and no significance testing — so its percentages are changes in a rate of change, not effects of AI, and it can settle no question that needs a control cohort. One internal discrepancy is on record: its chart restates the prior report's monthly-incident figure as +57.6% where the prior report itself says +57.9%.

Method (as stated)#

Within-company comparison of lowest- vs highest-AI-adoption quarters; Spearman's rank correlation (ρ) at p<0.05; metrics reported only with ≥6 companies of data; outliers excluded. The design avoids cross-org aggregation bias by comparing each company to itself over time.

How to read it (evidence posture)#

A vendor-claim source. The measurements are genuine telemetry, but Faros sells the very platform that produces them, and the reports' prescriptions — visibility, governance, and a "context engine" for the engineering environment — map to its product. Its headline conclusions (especially "engineering maturity offers no protection" and the direct contradiction of DORA 2025) are also the conclusions most favorable to selling the instrument. Trust the direction; attribute load-bearing claims to Faros; weight against vendor-neutral sources where they conflict.

Connections#

Acceleration Whiplash — the central thesis of its 2026 report
AI as Primary Author — its framing of the 20%→60% code-acceptance shift
Telemetry vs. Survey Measurement — its methodological stance and DORA counterpoint
Compounding Data Moat — owning the cross-org SDLC telemetry stream is the data asset that makes these reports possible
Verification as the New Bottleneck — its findings supply external telemetry for Fiona Fung's qualitative thesis

Sources#

AI Engineering Report 2026: The Acceleration Whiplash
The Speed Trap: 8 takeaways from our latest AI engineering research — The Speed Trap: 8 takeaways from our latest AI engineering research, Faros Research (corporate byline), 2026-09-18, vendor-claim. Blog summary of the gated Q3 2026 report; the "Key Findings" infographic carries four deltas the body text does not state