DORA Metrics in the AI Era

Narrative

Analyst and VC coverage of DORA-era metrics splits cleanly along a hype-to-sobriety timeline. Sequoia's 2023 "Developer Tools 2.0" piece framed Copilot as a category-defining "generational shift," noting Copilot's rapid growth to millions of users and speculating it "could grow to generate $1 billion or more in annual revenue among GitHub's 100 million users." That framing, alongside Sequoia's 2022 generative-AI market map predicting AI would "largely replace or comprehensively complement human labor by 2030," represents the accelerator thesis at its most bullish, and it predates the harder delivery-metric data that followed in 2024-2025.

The correction arrived through DORA itself rather than VC commentary. The 2024 Accelerate State of DevOps report found that a 25% increase in AI adoption is associated with an estimated 1.5% decrease in delivery throughput and a 7.2% decrease in delivery stability, alongside a 2.6% drop in time spent on valuable work, even as individual-level productivity, flow and job satisfaction rose. DORA's own explanation, reported via The New Stack, is that AI increases batch size (larger changesets), and "DORA's data has consistently shown that larger changesets introduce risk." The 2024 report also introduced rework rate as a stability-side metric and began separating throughput (deployment frequency, lead time) from stability (change failure rate, failed deployment recovery time) rather than assuming they move in lockstep, a shift RedMonk called one of several "holy shit moments" in that year's 120-page report. Platform engineering showed a similar bifurcation: internal platforms lifted productivity by roughly 8% and organisational performance by roughly 6%, while reducing throughput by about 8% and stability by about 14%, per the same report.

The 2025 DORA State of AI-assisted Software Development report, drawing on survey responses from nearly 5,000 technology professionals and over 100 hours of qualitative interviews, reframed the story around an "amplifier" thesis: AI adoption reached 90% (up 14 points year on year), over 80% of respondents said AI increased their productivity, but 30% still reported little or no trust in AI-generated code. Google Cloud's own framing is explicit that "AI doesn't fix a team; it amplifies what's already there," with struggling teams seeing existing problems "highlighted and intensified." DORA operationalised this with the new AI Capabilities Model, a companion report identifying seven capabilities empirically linked to positive AI impact: a clear and communicated AI stance, healthy data ecosystems, AI-accessible internal data, strong version control practices, working in small batches, user-centric focus, and quality internal platforms. DORA selected these seven from an initial candidate list of fifteen after finding they showed "a sustained and statistically significant interaction" with outcomes in the 2025 survey wave.

Independent and consultancy research complicates the acceleration narrative further. METR's July 2025 randomised controlled trial of 16 experienced open-source developers across 246 real tasks (averaging two hours each, using Cursor Pro with Claude 3.5/3.7 Sonnet) found that developers "take 19% longer to complete issues" with AI allowed, despite forecasting a 24% speedup beforehand and still estimating a 20% speedup after the fact. GitClear's telemetry across 211 million changed lines of code (2020-2024) found code churn rising toward roughly 7.9% of new code revised within two weeks by 2024, "moved" (refactored) code falling from about 25% of changed lines in 2021 to under 10% in 2024, and an eightfold rise in duplicated code blocks. Bain's Technology Report 2025 found that while two-thirds of software firms have rolled out generative AI tools, teams using AI assistants see productivity boosts of only "10 to 15 percent," arguing that coding is just 25-35% of the development lifecycle and gains evaporate without lifecycle-wide process redesign. Forrester's 2025 predictions similarly warned that "at least one organization will try to replace 50% of its developers with AI and fail," citing survey data that developers spend only about 24% of their time actually coding. McKinsey's earlier generative-AI lab work (2023) found developers using AI tools were more than twice as likely to report happiness and flow, but its 2025-2026 commentary (via a Jellyfish-partnered interview) shifted toward organisation-wide adoption thresholds, citing more than 60% of tracked organisations seeing at least 25% productivity improvement and claiming 80-100% developer adoption correlated with gains "of more than 110%," a vendor-sourced figure that has not been independently replicated. Gartner's September 2025 Magic Quadrant for AI Code Assistants evaluated 14 vendors (GitHub, AWS, Google Cloud, GitLab, JetBrains, Cognition/Windsurf, Anysphere/Cursor among them) and separately forecasts a 30% enterprise productivity gain "through 2028" from multi-tool AI adoption, an aspirational projection rather than a measured outcome. CB Insights' market maps track the code-generation and code-review vendor landscape, noting Cursor-maker Anysphere, Replit and Lovable crossed $100 million ARR "in record time" and that Microsoft's Satya Nadella has said "as much as 30% of Microsoft's code is now written by AI."

Taken together, the analyst and VC lane shows three distinct claim types operating simultaneously and rarely reconciled within a single report: AI as an individual-level accelerator (self-reported and telemetry-confirmed, per DORA 2025 and Faros's July 2025 telemetry of 10,000+ developers), AI as a source of new instability and rework (DORA 2024's throughput/stability declines, GitClear's churn data, METR's RCT), and AI as grounds to add new metrics entirely (DORA's rework rate, AI Capabilities Model, and value-stream-mapping tools). Vendor-published telemetry (Faros, Jellyfish) consistently reports larger and more favourable productivity numbers than academic or DORA survey data, and should be treated as a claim requiring independent corroboration rather than a settled fact.