According to a Google Dora research, AI coding tools are now ubiquitous among software developers, affecting stability, amplifying organizational problems and amplifying organizational problems without delaying software delivery.
An annual Google Research Group study measures software delivery performance in two key categories known as instability: release speed and efficiency, throughput, and release quality and reliability. It also measures individual software developer outcomes such as code quality, friction, and burnout.
Last year, a DORA survey of 3,000 respondents found that software delivery throughput decreased by 1.5% and delivery stability decreased by 7.2% with each organization's 25% increase in AI adoption. These results were measured differently this year out of 5,000 survey respondents and over 100 hours of survey interviews, but these results were significantly different from last year's results, according to Nathen Harvey, Dora Lead and developer advocate at Google Cloud.
“This year we are using standardized effects. [respondents are] Harvey uses more AI to measure the change in standard deviation from the mean. “Essentially what we're saying is that these numbers are relative, but they show improvements. It's not a major improvement, but it's obviously an improvement.”
This week's Dora's AI-supported software development report raises some hypotheses about what explains this change.
Nathen Harvey
“If AI handles some of the grantwork underlying the coding process (scaffolds, boilerplates, daily conversions), developers can spend more time focusing on deploying their code, leading to increased software delivery throughput, and ultimately improved product performance,” the report reads. “We can also observe organizational systems that adapt to the more fruitful environments of AI.”
These results resonated with one software engineering leader.
“Webops Company Pantheon has a great timeline of 1999,” said David Strauss, Chief Architect and Co-Founder of Webops Company Pantheon. “Expectations are more reasonable and the quality of the model has improved.”
Harvey said the negative impact of AI on software release stability is expected as the technology matures.
“To be honest, it's not surprising that throughput is beginning to start to be initiating inches first before instability decreases,” he said. “There's always pressure to move faster, move faster, move faster and get second in the type of stability.”
AI Trust issues rise
Other changes revealed in this year's survey include an increase in AI developer recruitment, with AI adoption rising from 76% in 2024 to 90% in 2025. Over 80% of respondents reported increased productivity, while 59% reported improved code quality.
However, while AI adoption increased significantly, respondents' confidence in their technology did not improve proportionally. Harvey said: Thirty percent of respondents said they trusted AI as “a little” or “nothing” from 39.2% last year, while 70% said they trusted AI's generated output as “somewhat” and “A~”.
Harvey interpreted these results as a sound adjustment to what AI coding tools can do.
I think people are caught up in awe of our feelings of respect due to the frustration of AI's current capabilities and the inability to truly understand the world. The probabilistic nature of AI makes it difficult for people to fully understand and trust it.
Torsten VolkAnalyst, Omdia
“The reality is, you shouldn't trust 100%. 100% trust in AI would be wrong,” he said. “As the instability in software delivery continues to increase, we need to make sure that the checks are in place to verify what comes up.”
An industry analyst said his own research shows a similar disconnect between the fact that IT decision makers not only claim to pay more for software with high quality AI capabilities and trust AI-based decisions, but also in fact frequently overriding AI decisions.
“I think people are caught between adoration at AI's current capabilities and frustration over the lack of real understanding of the world,” said Torsten Volk, an analyst at Omdia, division of Informa TechTarget. “The stochastic nature of AI makes it difficult for people to fully understand and trust it. AI can provide a response that looks human, but at other times the response is illogical.”
AI best practices emerge
Overall, the DORA report found that AI use is beginning to amplify, better and worse, reflecting existing tissue characteristics.
“We see complex outcomes with the people we see with AI. At the same time, we see 90% of people who use AI,” says Harvey. “So it's clear not whether you're using the AI that's driving this, but how you're using the AI that's driving that impact.”
Based on this year's survey results, Dora has identified seven best practices that are common to organizations that benefit AI.
A clear and conveyed stance of AI.
A healthy data ecosystem.
AI-accessible internal data.
Powerful version control practices.
Work in small batches.
User-centric focus.
High quality internal platform.
Dora also identified how applications of these practices lead to specific AI outcomes. For example, teams who want to improve product performance while using AI should focus on having accessible internal data, working in small batches, and clarifying their AI stance.
“It's about having your home tidy with organizations that are trying to use AI,” said IDC analyst Matthew Flag. “AI needs to be robust in workflow, processes and security attitudes all need to be solid as AI finds armor gaps.”
Beth Pariseau, senior news writer at Informa TechTarget, is an award-winning veteran of IT journalism covering Devops. Any hints? Please email her Or reach out @pariseutt.
Nathen Harvey
