DORA Reports Reconfigure AI as the centre of software development • Registers

Google Cloud's 2025 DevOps Research and Assessment (DORA) report has been published, with 90% of respondents using AI for software development, so the question is not whether they should adopt it, but how to achieve its value.

This study is based on approximately 5,000 survey responses from IT professionals interpreted by Dora Team. The DORA Research Project has been measuring its annual report for many years since its acquisition by Google in December 2014.

The biggest use of AI is to create new code (71% of developers), and the most common interaction with AI is followed by IDES (Integrated Development Environment) via chatbots (not all users, not all users). The use of agent mode where AI autonomously makes changes is less common, with 61% saying it doesn't do this, and only 17% do so more than once a day.

Although 80% of those surveyed believe that AI productivity is improving, there is a drawback. Thirty percent do not trust AI-generated code, which increases delivery instability, and while the entire AI acts as an amplifier, increasing the strength of high-performance organizations, it exacerbates dysfunction among people struggling, the report says.

The research paper includes a Dora AI functional model with seven technical and cultural best practices for AI adoption. These consist of clear communications of AI usage policies, high quality internal data, AI access to that data, powerful version control, small batches of work, user-centered focus, and a high quality internal platform. This end is vaguely defined, but refers to the software and systems in which developers build applications and services.

Co-founder of the Dora Project and famous for books including effective DevOps. Phoenix Projectcontributed to the report. He explains how he embraced the vibe and wrote a book on the subject, along with redundant and influential developer blogger Steve Yegge (Ex Amazon and Google). They saw how it didn't work, he wrote, “deleted tests, stops, and even deleted code repositories.”

Nevertheless, it has changed their lives, allowing faster, more ambitious projects. As a result, Kim writes, “Our control system – we need to speed up too,” the learning environment, taking into account the fast feedback loop, independent testing and deployment systems, and “the specificity of AI and its rapid progress rate.”

DORA Research divides DevOps teams into seven team archetypes, roughly counting in order of what is desirable, and is called the lowest name basic challenge and the highest achievement. Teams are defined based on six metrics. These are product performance, software delivery throughput, software delivery instability, burnout, friction, and valuable tasks.

According to the report, 10% of respondents are stuck at the lowest level, 20% are at the highest and 45% are at the highest level or above.

The complete report has 440 pages of details.

The 2025 report illustrates the dramatic shift in the DORA project's focus on AI. The 2024 report also featured AI, but its conclusions were mixed and measured. “AI doesn't look like a panacea,” reported at the time, and there was evidence in support of adopting AI, but “there are many obstacles, growth pain, and ways in which AI can have harmful effects.” One such effect was the reduction in the stability and throughput of software delivery.

In 2024 and the past few years, DORA research assessed the performance of software delivery based on four keys. Failures in lead time, deployment frequency, failure rate changes, and deployment recovery time between code commits and deployment. In 2025, these are only referenced in footnotes as one of many ways to measure software development.

What caused the Dora Team to shift their attitude to AI? In a 2025 paper, Kim said, “We started calling last year's report and its discovery, “The Dora 2024 Anomaly.” In his view, what needs to be changed is how DevOps is practiced in the age of AI.

As long as AI adoption does not drop (and note that Google is invested in something that isn't happening), promoting best practices about how it is used might be a practical option, even among skeptics. ®

Bootnote

The DORA Report is now called the “AI-Assisted Software Development State” and was previously called the “Accelerating State of DevOps.”