The DORA AI Capabilities Model is a framework from Google's 2025 State of AI-assisted Software Development report that names seven foundational organizational capabilities which determine whether AI adoption produces positive outcomes or compounds dysfunction. Derived from a survey of nearly 5,000 technology professionals, the model reframes the question from "which AI tool should we buy" to "which organizational conditions must exist for any AI tool to pay off." It is the most empirically grounded answer so far to why AI investment lands so unevenly across teams.
The seven capabilities
- A clear and communicated AI stance — explicit policy on what AI may be used for, with which data, and by whom; ambiguity suppresses adoption and pushes usage underground
- Healthy data ecosystems — data that is findable, trustworthy, and governed; AI amplifies whatever signal its inputs carry, including garbage
- AI-accessible internal data — organizational knowledge exposed to AI systems in usable form (not just present somewhere in a wiki)
- Quality internal platforms — paved-road developer platforms that make the correct action the easy action; without them AI-generated code bypasses guardrails
- Strong version control practices — small, reviewable commits, trunk-based development, clean history; AI-generated diffs magnify the cost of weak review practice
- Working in small batches — tight feedback loops so mistakes surface quickly; the "walking speed to 50 mph" control-theory analogy requires the sensing loop to speed up with the generation loop
- A user-centric focus — teams that stay anchored to user outcomes route AI productivity into value, not toward building more of the wrong thing faster
Amplification, not addition
The model's organizing claim is that AI multiplies existing organizational capability rather than adding a fixed productivity delta. When the seven capabilities are present, AI benefits compound across individual effectiveness, team performance, and software delivery throughput. When they are absent, AI creates "localized pockets of productivity that are often lost to downstream chaos." This is why 2025 is the first year DORA finds AI improving throughput — earlier results were dominated by teams adopting AI into weak systems.
The instability finding is the flip side: AI still increases delivery instability on average, and the gap between high- and low-capability teams is widening. See ai-coding-assistants for how this shows up at the line-of-code level and organizational-ai-enablement for how enablement programs should target these capabilities rather than tool rollouts.
Practical use
Leaders can use the seven capabilities as a diagnostic before scaling AI investment:
- Score each capability 1–5 at the team or organization level
- Prioritize weakest capabilities before purchasing additional AI seats
- Treat the model as a checklist for platform and data investments, not as a training syllabus
- Re-score after each major AI rollout — improvement in capability scores should precede improvement in delivery metrics