DORA
Confidence 0.75 · 1 source · last confirmed 2026-08-30
DevOps Research and Assessment, now within Google Cloud — the longest-running research programme on software delivery performance, and the source of the throughput/stability metric vocabulary the industry uses.
Its 2025 report, refocused entirely on AI, is the corpus’s largest-sample source on AI in software development: nearly 5,000 technology professionals plus 100+ hours of qualitative data. Two contributions matter here.
The amplifier thesis: “AI doesn’t fix a team; it amplifies what’s already there.” Strong teams get more efficient; struggling teams find AI surfaces problems they already had. This is DORA’s explanation for why AI-adoption studies produce such scattered effect sizes — the moderator is the delivery system, not the tool. The seven team profiles derived by cluster analysis are its operational form.
The throughput/stability split: in 2025 AI adoption’s relationship with delivery throughput turned positive, while “AI adoption does continue to have a negative relationship with software delivery stability.” The mechanism they name is that acceleration exposes downstream weakness: without strong automated testing, mature version control and fast feedback, higher change volume becomes instability. That single finding is the survey-scale statement of what nearly every other source in this ingest describes locally.
The DORA AI Capabilities Model names seven amplifying capabilities, including fortify your safety nets and invest in your internal platform; DORA’s position is that internal platform quality correlates directly with the ability to unlock AI value.
Caveat worth carrying: DORA’s adoption, trust and perceived productivity figures are self-report, and METR’s RCT shows developer perception of AI productivity to be wrong by roughly 39 points for experienced developers. DORA’s delivery metrics are behavioural and survive this.
The 2025 announcement was authored by Nathen Harvey (DORA Lead) and Derek DeBellis (Researcher).