Many software developers use AI, but not all trust it, Google’s 2025 DORA State of AI-assisted Software Development report found.
The DORA research team publishes reports on high-performing, technology-driven teams, establishing reference points to measure performance. While AI was mentioned in last year’s report, AI takes the spotlight this year because AI adoption has increased.
“The idea here is really to help teams get the most out of AI,” said Nathen Harvey, DORA lead and developer advocate at Google Cloud, in an interview with TechRepublic.
The DORA researchers surveyed more than 5,000 technology professionals and analyzed more than 100 hours of qualitative data. The survey participants included product managers, site reliability engineers, quality assurance professionals, security professionals, team leaders, and others.
Productivity slightly increased with AI for 41% of developers
AI adoption has grown to 90% among software development professionals, the study found. Compared to last year, the Google DORA research program saw a 14% increase in AI adoption.
“I suspect that by next year we probably don’t have to ask this question anymore,” said Harvey. “Everyone will be using AI.”
Of all of the professionals Google DORA researchers surveyed, 65% say they use AI “heavily.” The median user turns to AI for approximately two hours per day.
Developers’ perceptions of AI’s impact on their own productivity vary:
- 13% said productivity increased extremely.
- 31% said it moderately increased.
- 41% said it slightly increased.
- 9% said it had no impact.
- 3% said their productivity slightly decreased.
- 1% said their productivity moderately decreased.
- Under 1% said it increased extremely.
AI use is becoming more reflexive, with 39% of the surveyed software developers turning to it when encountering a problem or task, the DORA report found. When asked how often they rely on AI at work:
- 37% said a moderate amount.
- 30% said a little.
- 20% said a lot.
- 8% said a great deal.
- 5% said not at all.
Where AI helps — and where it doesn’t
How are people using AI? Most (55%) use chatbots like ChatGPT. Others (41%) use generative AI built into their development environment, in external web interfaces (31%), or internal web interfaces (22%). Relatively small percentages use AI as part of an automated tool chain (18%) or in other dev tools and platforms (18%).
Using a chatbot was relatively popular, with 25% of respondents using AI-assisted conversations for coding tasks a few times a day. Less popular was agentic AI, with 61% of respondents indicating they “never” used an autonomous agent mode.
This spectrum, in terms of the type of AI being used, probably comes down to what is available on the market, Harvey said. “Agentic capabilities are the last thing that’s been brought to market. A lot of people are not using it. That’s one big takeaway. I think you can see that progression of how these capabilities have been brought to market.”
He framed the disparity between agentic AI — the least popular type — and chatbots — the most popular type — as a snapshot of June and July, when the data was gathered.
“I think that’s going to change over the next year,” he said.
30% trust AI “a little” or “not at all”
However, not all software developers trust AI; 30% trust it “a little” or “not at all.”
Looking at perceived code quality could offer a clue as to why. Of the respondents, 31% said AI slightly improved their code, while 30% said it had no impact.
Overall, 49% of software developers said they trust AI-generated output “somewhat.”
AI adoption tips
The DORA research team recommends leaders do the following to smooth out AI adoption:
- Treat your AI adoption as an organizational transformation.
- Shift the conversation from adoption to effective use.
- Diagnose team health with more than just software delivery performance metrics.
- Prioritize and fund your platform engineering initiatives.
- Turn localized productivity gains into significant organizational advantages.
DORA AI Capabilities Model can assess where teams can improve
The report introduces the DORA AI Capabilities Model, which is a blueprint of seven essential capabilities for amplifying AI’s impact. Those qualities are:
- Clear and communicated AI stances.
- Health data ecosystems.
- AI-accessible internal data.
- Strong version control practices.
- Working in small batches.
- User-centric focus.
- Quality internal platforms.
“A team can sit down, whether that’s the leader or the team itself, and say, what is the thing we want to improve the most? Where’s our focus?” Harvey said.
The DORA AI capabilities model, available in the full report, shows which technical and cultural factors are connected to which element of team performance.
For example, Harvey said, “Maybe, on a particular team, your focus is product performance. Well, you can use this capabilities model to say we want to focus on product performance. Let’s follow the arrows backwards. What conditions drive product performance? In this case, it’s accessible internal data. It’s working in small batches, and it’s that clear and communicated stance on how to use AI. So now, as a team, we can say, how are we doing against those three things? Maybe we have a weakness or an opportunity to lean in even more.”
Google integrated a bundle of Gemini AI tools directly into Chrome, adding the capacity to ask questions across multiple tabs and more.
