AI tools and DevEx signals
- Early signals show AI coding tools raise raw output but shift pressure onto reviews and code quality processes. - One dataset reported 76% more commits and 59% longer developer sessions, yet only minor improvements in dev-experience scores. - DORA-linked commentary warns higher AI usage can correlate with lower throughput and stability, recommending focus on review burden and rework ( ).
AI coding tools are raising output faster than they are improving the day-to-day experience of building software. (dora.dev) Google Cloud’s 2025 DORA report, published September 23, 2025, drew on nearly 5,000 technology professionals and more than 100 hours of qualitative research. It found AI adoption had a positive relationship with software delivery throughput and product performance, but a negative relationship with software delivery stability. (cloud.google.com) DORA described AI as an “amplifier” rather than a fix: teams with strong testing, version control, feedback loops, and loosely coupled systems captured more gains, while weaker processes saw their existing problems exposed downstream. (research.google) That pattern lines up with newer operational data from DX, a developer intelligence vendor tracking more than 135,000 developers across 425 organizations in its Q4 2025 report. DX said AI adoption exceeded 90%, daily AI users shipped 60% more pull requests than non-users, and 22% of merged code was AI-authored. (getdx.com) DX’s own longer-running sample of 400 companies, covering November 2024 through February 2026, pointed to a smaller organization-level gain than the raw activity numbers suggest. In that analysis, AI usage rose 65% while pull-request throughput rose 9.97%, which DX said was more in line with the 8% to 12% range engineering leaders report seeing in practice. (getdx.com) The missing piece is that coding is only one slice of the job. DX said planning, alignment, scoping, code review, and handoffs still consume much of the software development lifecycle, so faster code generation does not automatically translate into much faster shipping. (getdx.com) Atlassian’s 2025 Developer Experience Report found a similar split between task-level gains and workflow friction. In a survey of 3,500 developers and managers across six countries, 99% of developers reported time savings from AI tools, but 90% also said they still lose six or more hours a week to inefficiencies such as finding information, adapting to new technology, and switching contexts between tools. (atlassian.com) Atlassian also said developers spend only 16% of their time coding, which helps explain why coding assistants can make parts of the work feel faster without materially fixing the broader developer experience. The company’s report said 50% of developers still lose more than 10 hours a week to non-coding friction. (atlassian.com) GitHub had flagged the same issue earlier, in a June 2023 survey of 500 U.S. enterprise developers. That report said developers do more than write code and noted that waiting on builds and tests remained one of the most time-consuming parts of the job besides coding itself. (github.blog) The current signal across these reports is not that AI coding tools fail to raise activity. It is that the next bottlenecks are review load, rework, testing, and the quality controls that sit between a generated diff and stable software in production. (cloud.google.com; getdx.com)
