DORA 2025, Part 5. 7 DORA AI Capabilities: How Technical and Cultural Practices Strengthen AI Impact in Software Engineering

DORA has always aimed not just to describe the state of software delivery, but also to help organizations make data-driven decisions, especially in a rapidly changing environment of tools, practices, and technologies. AI is already radically changing development.

Rapid progress opens up new opportunities, but it also raises questions about how engineering practice should evolve to match the new conditions. This year, we went beyond simply analyzing who uses AI and how.

We studied the conditions under which teams using AI achieve better results.

The result is the first version of the DORA AI capabilities model.

It includes seven organizational capabilities that increase the benefits of AI adoption.

These capabilities cover both technical and cultural aspects.

Our data show that systematic investment in these areas helps unlock the real potential of AI tools.

As with the DORA Core Model, work on the AI model will continue.

We will keep reviewing, refining, and extending it with new research, and share updates with the DORA community.

To build the model, the DORA team proposed a broad set of organizational capabilities that can amplify the impact of AI adoption.

The foundation was 78 in-depth interviews, expert assessments, and previous DORA research.

After discussion and prioritization, 15 candidate capabilities were included in the survey.

Seven of them showed a sustained and statistically significant interaction with AI use.

When teams combined these capabilities with active AI use, AI's effect on key outcomes increased substantially.

These seven capabilities became the core of the new model: 1.

A clear and open company stance on AI use 2.

Internal data available for AI 4.

is employees' clear understanding of how and under what conditions they may use AI tools at work.

This capability is built on four elements.

They reflect how employees perceive:

How expected and supported AI use is in their workplace.

How much the organization encourages AI experimentation.

How clear it is which AI tools are allowed.

How much the company's AI policy applies to their work specifically.

A company with a clear AI stance creates an environment where developers understand that using AI is normal and an expected part of the job; AI experiments are supported; AI usage rules are transparent; and employees know which tools they can use and how company policy applies to their tasks.

This approach reduces uncertainty and helps use AI safely and effectively.

Our analysis shows that the positive effects of AI adoption increase only in organizations with a clear and transparent stance on AI use.

When such a position exists, the following is observed with high confidence:

Individual productivity gains are stronger.

Overall organizational performance gains are stronger.

AI's neutral effect on friction becomes positive, reducing obstacles in day-to-day work. With less, but still notable, confidence, we see that AI's positive impact on software delivery speed becomes more pronounced.

Based on the interviews, developers regularly noted that they lacked clarity and understanding of the company's position on AI use.

This leads to two opposing effects:

Some employees act too cautiously, using AI less than they could for fear of breaking the rules.

Some use it too freely, applying AI where it goes beyond acceptable limits.

Both situations create risks for the organization.

That is why we previously emphasized that a clear and open company stance on AI helps build developer trust, reduces unfounded concerns about data privacy, and accelerates the scaling of AI tools across the organization.

New data confirm that when a company clearly states what is allowed, what is expected, and where the boundaries are, results improve. Importantly, this capability describes not the content of the policy itself, but its clarity and transparency for employees.

Each team and organization can define its own position based on role, industry, and data infrastructure.

But if that position is clearly articulated and communicated to developers, the organization gains more benefit from introducing AI into the development process. "Why didn't I start using AI earlier? Maybe because I did not understand how colleagues and management would react to it.

So there was no sense that they could be punished for it.

At the same time, I did not understand how much it was actually encouraged or whether they wanted us to keep using it. I did not want to do it in secret.

We do have an AI policy, but it mostly covers what data can be shared from a client confidentiality perspective. Something like that. I think if we had more explicit support, I would use AI more often, at least for routine tasks"

A "healthy data ecosystem" refers to the overall quality of an organization's internal data. In the study, this measure is assessed as a single factor across three criteria reflecting employee perceptions: the quality of internal data sources; the accessibility of that data; and the degree of data fragmentation across teams and systems.

An organization with a healthy data ecosystem is one where data is high-quality, accessible, and integrated. With a high degree of confidence, we can say that the positive effect of AI adoption increases in companies where the data ecosystem is mature. In such conditions, AI's impact on work outcomes is significantly stronger.

It is commonly said that the quality of AI models is determined by the quality of the data they are trained on. In this case, the rule also applies at the organizational level: the better the internal data is organized, the more tangible the impact of AI.

If a company invests in building a high-quality, accessible, and unified data ecosystem, it gains a much greater boost in efficiency than from AI adoption alone.

This refers to how well AI tools are connected to the company's internal data and systems.

The measure includes four aspects: employees see that AI tools have access to internal information; AI responses are grounded in context from internal data; developers regularly provide internal data to AI in their prompts; developers use AI to retrieve internal information.

A company with internal data accessible to AI is an environment where employees understand that AI can see the needed data and can use and process it.

The study shows that the positive effect of AI adoption strengthens when AI has access to internal data. Then AI has a greater impact on individual productivity, and AI improves code quality more strongly. Tools trained on general knowledge already help, but real value emerges when AI can access data that reflects the company's context.

This suggests that organizations that connect AI to their internal systems gain more value from AI than those that rely only on the capabilities of foundation models.

To achieve a noticeable increase in efficiency and quality, buying licenses alone is not enough. AI needs access to enterprise data.

In essence, that makes sense: if AI cannot see the company's data, how useful can it really be? "Many of my clients are not yet ready to use AI effectively. Their data is not even organized: it is scattered across the company, and there is no common storage standard. And for AI to actually work, a significant amount of data engineering is needed to bring everything into a form the model can consume.

Most companies have not even reached this stage yet.

shows how much the team takes into account the experience and needs of the product's end users.

This measure is built on three judgments: the team aims to create value for users; prioritizes their experience; and understands that business success depends on user satisfaction.

Teams with a strong user focus get more value from AI use. In such teams, AI's positive impact on efficiency and outcomes increases noticeably. Where this orientation is absent, AI's effect becomes negative - the team performs worse.

The research emphasizes that a deep understanding of users, their goals, and feedback increases AI's impact, creates a unified product strategy, and helps prevent risks.

Without this approach, AI does not help the team - it can harm it. "That is why I have been here for five years already - I feel like I am doing something truly meaningful and helping ordinary people.

Even if it were for one person, that would already mean a lot to me. Here, I work for millions.

When difficult days or a bad mood come along, the thought that my work matters and that even a small feature will help hundreds of thousands of people this year is a strong source of support and gives me the energy to keep going." "The real insight was that clinicians do not need new tools - they need less noise.

We thought value would come from giving doctors more advanced capabilities.

The opposite turned out to be true: the technology's simplicity and invisibility became the main innovations.

A hybrid AI + human model delivers higher accuracy, more empathy, and greater trust than a fully automated approach" - IKS Health

Internal platforms refer to a set of shared capabilities available to different products and services within the company.

Their job is to make these capabilities consistent and available across the organization.

We assess internal platform quality using a single index: respondents indicate how many of 12 key characteristics their company's platform has. With a high degree of confidence, the data show that AI's impact on outcomes depends on internal platform quality. In companies with strong internal platforms, AI's positive effect on organizational effectiveness is amplified.

There is also a downside: AI's neutral effect on friction can turn negative - people feel more obstacles when working in organizations with strong platforms.

However, the overall picture remains positive.

Strong platforms increase individual productivity through a consistent standard of capabilities, while also setting boundaries - for example, limiting tool use through internal APIs with higher security requirements.

These platforms simultaneously expand access to the right tools and restrict access to undesirable ones.

Until the industry develops a common set of best practices for AI tools, we assume that strong platforms most often play a protective role: preventing improper AI use.

This may increase subjective friction for active AI users, but it does not necessarily harm the organization. With that in mind, we believe that designing and maintaining high-quality internal platforms is a key capability for companies that want to develop software successfully in an environment where AI is integrated into workflows.