AI is an amplifier; strategic focus on the organizational system brings the greatest returns, Nathen Harvey explained in his talk Scale High Performance, Not Dysfunction at QCon London. DORA’s 2025 research on AI in software development provides team profiles and success capabilities that can be used to put the research into practice.
DORA researches how teams get better at doing the technology, Harvey explained. It comprises software delivery, but not just the technology; also how we work as teams, as people, covering how we drive our businesses forward and really serve our customers. DORA runs a survey that is augmented with qualitative research. In 2025, they published the State of AI-Assisted Software Development at DORA reports.
AI’s primary role in software development is that of an amplifier, Harvey mentioned:
The greatest returns on your investments in AI are going to come from a strategic focus on the underlying organizational system. Without that foundation, AI tends to create localized pockets of productivity that are often lost to downstream chaos.
The researchers wanted to understand how teams were doing across different factors, like team performance, product performance, effectiveness, friction, burnout, etc. Harvey mentioned that they identified seven unique team profiles in the data:
-
Foundational challenges
-
The legacy bottleneck
-
Constrained by process
-
High impact, low cadence
-
Stable and methodical
-
Pragmatic performers
-
Harmonious high-achievers As an example, he showed the spider chart for the "constrained by process" team profile:
For this team profile, burnout is high, friction is high, individual effectiveness is low, and the time spent doing valuable work is low, Harvey said. Software delivery instability is pretty good; it’s lower than everyone else’s, but throughput is a little bit lower. It’s not the perfect team to be on, but we’re not really delivering a whole lot of value, Harvey mentioned. Harvey mentioned that they wanted to know if there are specific things that help some teams get more value out of AI than others. They identified seven capabilities that, when combined with AI adoption, drive better outcomes:
- Clear and communicated AI stance
- Healthy data ecosystems
- AI-accessible internal data
- Strong version control practices
- Working in small batches
- User-centric focus
- Quality internal platforms
Every piece of software that we create is made for humans, Harvey argued:
Our job is to solve problems. Our job is to understand who our users are and build the things that they need, the things that are going to improve their lives.
AI can do things a lot faster; it unlocks optionality, Harvey said. We can prototype five things in less time than it took to prototype one thing. As we’re really bad at understanding exactly what our users want, Harvey suggested building many things, trying them out, and debating them as a team.
Last December, DORA published the AI Capabilities Model report. For putting research into practice, Harvey suggested assessing your current state, setting priorities, and developing a culture of continuous improvement. When you work in an organization, you have your own context and customers; you’ve got to contextualize what’s in the DORA reports, Harvey said. Take the findings and use them as your hypothesis, he concluded.
InfoQ interviewed Nathen Harvey after his talk. InfoQ: What are the main conclusions from the DORA report when it comes to the perceived impact of AI on individual productivity and code quality?
Nathen Harvey: As AI adoption increased, the perceived impact on both code quality and individual effectiveness increased. In fact, self-reported individual effectiveness saw the highest increase relative to every other aspect we looked at. Perceptions of code quality also improved.Interestingly, while individual effectiveness and code quality improved, software delivery instability also increased. This means that we’re more likely to see deployments being rolled back or hot fixes required to correct an error introduced by a new deployment.
This contradiction- increased code quality and software delivery instability- may be a signal of misaligned incentives or lack of visibility into the impact of changes introduced. This can happen when a software engineer is primarily concerned with building new features, not worrying about how well those features land in production..
InfoQ: How do the AI capabilities help teams to get more value out of AI?
Harvey: The capabilities help teams understand how to get the most out of their investments in adopting AI. For example, DORA’s data has shown for a long time that working in smaller batches leads to better software delivery performance. AI puts pressure on this idea because the models can suggest large change sets or pull requests.Larger changes are more difficult to review and assess to see if they’re fit for purpose. Beyond that, large changes tend to have more risks.
Teams can use these insights to collaborate with their AI assistants to break large changes down into smaller, independently deployable units.
InfoQ: What role can platforms play in AI adoption and value generation?
Harvey: Quality internal platforms play a significant role in helping organizations get value from AI. First, a platform is an optimal way to make a particular capability accessible to the entire organization. Making models, workflows, and other capabilities widely accessible helps drive adoption and value realization.But it goes beyond simply making a capability accessible. Platforms also allow organizations to shift complexity down into the platform, hiding it from users of the platform.
For example, if your organization has a specific security policy that it must follow, the platform can enforce that policy for any new solutions that are built. In doing so, it improves compliance with the policy while also abstracting away the complexities of having to think about the policy.