Making Sense of the DORA AI Capabilities Model: Where Testing Teams Should Start

Google’s DevOps Research and Assessment (DORA) team has spent over a decade studying what makes software teams perform at elite levels. Their latest evolution — the DORA AI Capabilities Model — expands the framework to account for artificial intelligence. But as many in the testing community are asking: if AI amplifies both the good and the bad in your delivery pipeline, where should your team actually start?

What Is the DORA AI Capabilities Model?

The DORA Capabilities Model has long been the gold standard for measuring software delivery performance. It identifies specific technical and cultural practices that correlate with better outcomes — faster deployments, fewer failures, happier teams. The model groups capabilities into categories like continuous delivery, architecture, process, and cultural enablers.

With the 2024–2025 update, DORA introduced a new layer: AI-specific capabilities. These aren’t standalone features you bolt onto an existing pipeline. Instead, they describe the conditions under which AI tools actually deliver on their promises — rather than creating new bottlenecks, security risks, or maintenance nightmares. The model now acknowledges what testers have been saying for years: garbage in, garbage out applies to AI just as much as it does to manual processes.

The New AI-Specific Capabilities

DORA identifies four capabilities that specifically enable or constrain AI adoption in software delivery:

• AI-accessible internal data: Your AI tools need paths into internal documentation, codebases, and test repositories. If your Confluence is a graveyard of outdated pages and your test cases live in someone’s head, AI-assisted testing tools won’t magically fix that.
• Clear and communicated AI stance: Teams need explicit policies on when and how AI can be used. Without clarity, some engineers will paste proprietary code into public chatbots while others refuse to touch any AI tool at all — both extremes create risk.
• Healthy data ecosystems: AI benefits are significantly amplified by high-quality, accessible, and unified internal data. If your bug tracker, test management tool, and monitoring dashboard don’t talk to each other, your AI won’t either.
• Platform engineering for AI: A platform provides the automated, secure pathways that allow AI’s benefits to scale across the organization. This means building shared toolchains that everyone can use, rather than letting each team cobble together their own AI stack.

Why Core Capabilities Are the Foundation

Here’s the critical insight from the model: the AI-specific capabilities only work when the foundational capabilities are already in place. You can’t have AI-accessible data if your documentation quality is poor. You can’t build a platform for AI-assisted testing if you haven’t automated your deployments. AI amplifies what already exists — it doesn’t create capability from nothing.

For testing teams, this means the fundamentals still matter enormously. Test automation, continuous integration, test data management, and a learning culture are prerequisites for any meaningful AI adoption. Teams that skip straight to “let’s use an AI test generator” without solid automation foundations invariably discover that their AI-generated tests are unreliable, unmaintainable, or simply wrong.

The model also reinforces something experienced testers already know: culture eats tools for breakfast. Capabilities like generative organizational culture, team experimentation, and transformational leadership remain just as important in an AI-augmented world. A team that doesn’t feel safe admitting mistakes won’t suddenly embrace transparency just because an AI dashboard exists.

Where Should Your Team Start?

The DORA AI Capabilities Model doesn’t prescribe a one-size-fits-all roadmap. But its structure suggests a clear starting point: assess your foundations before chasing AI features. Here’s a practical sequence:

1. Audit your data health. Are your test cases, requirements, and bug reports structured and searchable? If you asked an AI tool “what are our most frequently failing test scenarios?” would it be able to answer from the available data?
2. Define your AI stance. Draft a one-page policy covering which AI tools are approved, what data can be shared, and how AI-assisted work should be reviewed. Make it simple enough that every team member can explain it.
3. Strengthen one core capability. Pick the weakest link in your delivery pipeline — maybe it’s test data management, or deployment automation — and invest in making it genuinely solid. The AI benefits will follow naturally.
4. Experiment deliberately. Once the foundations are steady, pick a single AI-assisted workflow — like generating test ideas from user stories — and run a controlled experiment. Measure whether it actually improves outcomes before scaling.

The Ministry of Testing community frames this well: “AI amplifies the good and the bad, so where should your team start?” The answer, supported by the DORA model, is clear. Start by making sure there’s more good than bad for AI to amplify.

The Bottom Line

The DORA AI Capabilities Model is not a call to throw out everything you know about software delivery. It’s a recognition that AI shifts the landscape — and that teams who invest in data quality, clear policies, shared platforms, and strong fundamentals will be the ones who benefit most from AI-assisted development and testing. For testing teams, the message is reassuring: the skills you’ve been building all along — automation, critical thinking, collaboration — are exactly what makes AI work well in practice.

Source: “Making sense of the DORA AI Capabilities Model” — Ministry of Testing. Learn more about the DORA capabilities at dora.dev/capabilities.

Photo: Daniil Komov / Pexels