Most organizations have already decided to adopt AI. Nearly 90% of technology professionals use it in their work. The harder question — the one most leaders still haven’t answered — is how to make that investment pay off.
Google’s DORA team has published a new framework that provides a clear, evidence-based answer. The DORA AI Capabilities Model, based on more than 100 hours of qualitative data and survey responses from nearly 5,000 technology professionals worldwide, identifies seven foundational capabilities that determine whether AI helps or hurts an organization’s performance.
The headline finding is blunt: AI is an amplifier. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones.
That means buying AI tools without fixing the underlying problems is like putting a bigger engine in a car with bad brakes. You’ll go faster — but not in a good way.
Seven Capabilities That Actually Matter
The DORA AI Capabilities Model identifies seven capabilities spanning technical and cultural domains. Each has been shown to amplify the positive impact of AI on performance. They are:
- Clear and communicated AI stance — Ambiguity around AI tools stifles adoption and creates risk. Teams need to know what’s permitted, what’s expected, and where the boundaries are.
- Healthy data ecosystems — AI tools are only as good as the data they can access. High-quality, accessible, unified data amplifies organizational performance.
- AI-accessible internal data — Connecting AI to internal documentation, codebases, and style guides transforms it from a generic assistant into a specialized expert. DORA calls this shift moving from “prompt engineering” to “context engineering.”
- Strong version control practices — As AI increases the volume and velocity of code changes, version control becomes the critical safety net. Teams that commit frequently and use rollback capabilities see better performance.
- Working in small batches — AI can generate large amounts of code quickly. But large changes are hard to review, test, and integrate. Small-batch discipline channels AI speed into stable product performance.
- User-centric focus — This one carries a warning. Teams without a strong user focus that adopt AI actually see a decline in team performance. If you’re moving fast in the wrong direction, AI makes it worse.
- Quality internal platforms — A solid internal platform provides the automated, secure pathways that allow AI’s benefits to scale across the organization. Without it, individual productivity gains get absorbed by downstream bottlenecks.
The User-Centric Warning
The user-centric finding deserves extra attention. DORA’s data show that when teams with a strong user focus adopt AI, their effectiveness increases. But teams with poor user focus that adopt AI see their team performance decline.
This makes intuitive sense. AI helps you move faster. If your product priorities aren’t aligned to actual user needs, you’ll just build the wrong things more efficiently.
From Prompt Engineering to Context Engineering
One of the report’s most practical contributions is its framework for making internal data AI-accessible. DORA describes a three-phase approach: start with manual context engineering (assembling briefing packets for AI), move to automated pilots using retrieval-augmented generation (RAG) or model context protocol (MCP) servers, then scale across the organization.
The key insight is that feeding an AI tool your company’s codebase, style guides, API docs, and security policies transforms the quality of its output. Generic AI assistance becomes context-aware expertise.
“DORA’s AI Capabilities Model reinforces a point many organizations are learning the hard way: AI amplifies the system it enters. If your delivery practices, data discipline, and platform foundations are strong, AI accelerates throughput and insight. If they are fragmented or misaligned, AI accelerates confusion and rework,” according to Mitch Ashley, VP and practice lead, software lifecycle engineering, The Futurum Group.
“The shift from prompt engineering to context engineering signals where we are heading. Context-aware AI, grounded in internal codebases, policies, and documentation, changes output quality and reliability. That requires intentional platform engineering and governed data access. Organizations that treat AI adoption as an operating model evolution rather than a tooling decision will see durable gains across the software lifecycle.”
Where to Start
DORA’s report doesn’t just describe the problem — it provides a practical assessment framework. The research identified seven distinct team archetypes, ranging from “harmonious high-achievers” to teams facing “foundational challenges.” Each archetype maps to specific improvement priorities.
The report also includes guidance on value stream mapping to identify real bottlenecks, as well as a 90-minute team workshop format for prioritizing which capabilities to tackle first.
The bottom line: organizations that treat AI adoption as a tool procurement exercise will be disappointed. Those that treat it as an organizational transformation — investing in data quality, platform engineering, version control discipline, and user-centric culture alongside their AI tools — will see compounding returns.
AI doesn’t fix what’s already broken. But for teams that have built strong foundations, it makes good work significantly better. That’s the amplifier effect. And it’s the most important thing to understand about AI in software development right now.
