DX's latest AI Impact Report finds AI adoption among developers nearing 100%, with AI-authored code now over half of merged code. But gains are uneven: average PR size grew from 42 to 72 lines of code, raising review and quality concerns, while code maintainability is improving even as change confidence declines. Change failure rates are becoming more volatile across organizations, and DX's Developer Experience Index has dropped about 2% over two quarters. AI spend has grown roughly 28x year-over-year among large companies, but the innovation ratio rose only about one percentage point, suggesting time saved by AI is often absorbed by existing organizational friction like meetings and review delays.
Questions this post answers
How much has average pull request size increased since developers started using AI coding tools heavily?
Average pull request size has grown from 42 lines of code to 72 lines of code, according to DX's AI Impact Report. This growth raises concerns because larger PRs are harder to review and revert, more likely to be rubber-stamped by reviewers, and can move more slowly through the development pipeline despite being generated faster. daily.dev helps engineering teams track how AI-driven code growth is reshaping review practices.
Is AI spending in software engineering actually producing more innovation?
Not proportionally. Among the largest companies studied by DX, AI spend has risen to roughly 28 times year-ago levels, while the innovation ratio increased by only about one percentage point. Developer ramp-up time did improve, but time saved by AI is often absorbed by existing organizational friction like meetings, build waits, and review delays rather than converting into shipped customer value. engineering leaders weighing AI investment against real output gains can follow this data on daily.dev.
Why is code maintainability improving while developer confidence in their changes is declining?
AI tools are making code easier to understand and modify, which improves measured maintainability, but developers report less confidence that their changes won't break something in production. DX's research found these two measures, which historically moved together, are now diverging, suggesting AI may be changing what maintainable code actually means as developers rely more on agents to understand and modify it. developers navigating AI-assisted code changes can track shifts like this on daily.dev.