State of DevOps 2025: Review of the DORA Report on AI ... - Splunk

### The "trust paradox" Here lies one of the report's most intriguing findings. While over 80% of users report significant productivity gains from AI, a substantial **30% still have little to no trust in the code it generates**. This isn't a sign of failure, but rather one of sophisticated maturity. Developers aren't blindly accepting AI's output. They are treating it as a brilliant but fallible junior partner, applying the same healthy "trust but verify" skepticism they've honed for years with solutions from Stack Overflow or open-source libraries. This nuanced relationship is key to harnessing AI's power safely. ... The 2025 data provides a critical update to last year's findings. AI adoption now clearly and positively correlates with **software delivery throughput**. Teams are successfully leveraging AI to write and ship code faster. However, this acceleration comes at a cost. The report also finds that AI adoption continues to correlate with **higher instability**, leading to more change failures, increased rework, and longer cycle times to resolve issues. As one analysis aptly puts it, we may be faster, but are we any better? AI is exposing the downstream bottlenecks in testing, code review, and quality assurance that are not equipped to handle this new, accelerated pace.