DORA 2025: AI Is Ubiquitous—Success Demands Organizational Change
DORA 2025 finds AI is now ubiquitous in software development: 90% adoption, two hours of daily use, strong productivity gains, and improved code quality. Despite benefits, a trust paradox persists, and organizational outcomes hinge on cohesive practices, with AI amplifying both strengths and weaknesses. DORA introduces team archetypes and a new AI Capabilities Model to help organizations translate adoption into durable, system-wide impact.
Key Points
- AI adoption in software development has reached 90% (up 14% YoY), with developers spending a median of two hours daily using AI.
- 65% report at least moderate reliance on AI; 80%+ cite productivity gains and 59% see improved code quality.
- A “trust paradox” endures: many use AI despite limited trust (24% high trust vs. 30% little or no trust), treating it as assistive rather than a replacement.
- AI adoption is now linked to higher software delivery throughput, but assuring pre-release software quality remains challenging.
- AI acts as a “mirror and multiplier” of team dynamics; DORA introduces seven team archetypes and the DORA AI Capabilities Model to guide effective, culture-plus-technology adoption.
Sentiment
The Hacker News community is predominantly skeptical of the DORA report's positive findings. While a few commenters share genuine productivity gains, the majority questions the survey methodology, highlights Google's conflict of interest, and shares anecdotes of AI degrading code quality and developer skill. The dominant view is that self-reported productivity perceptions are unreliable and that the report functions more as vendor marketing than objective research.
In Agreement
- One developer reports a 3x increase in merged PR rate since adopting AI coding tools, suggesting the productivity gains are real and measurable for those who find the right workflow
- AI coding abilities are sometimes frighteningly good in quality, drawing parallels to the industrial revolution's displacement of craft work
- AI is genuinely useful for boilerplate, prototypes, client interfaces, and tests when treated as a sometimes competent assistant rather than a replacement
- AI assistants prove valuable when working with unfamiliar libraries or language features, even if the output isn't premium quality
- DORA's expansion beyond DevOps into platform engineering, developer experience, and AI reflects how core DevOps principles have broader impact
Opposed
- A senior engineer describes spending 45 minutes reviewing an AI-generated merge request that introduced 100 lines of changes when the actual fix was a single-line config change — developers who rely on AI don't understand what their code does
- The study measures perception rather than reality — self-reported productivity is basically useless since people are notoriously bad at self-evaluating, and multiple studies show developers feel more productive while actually taking longer
- Google has a clear conflict of interest as both an AI vendor and the survey conductor, and the report functions as marketing rather than objective research
- A prisoner's dilemma exists where professionals feel pressured to overstate AI's usefulness to avoid being seen as resistant, making honest assessment nearly impossible
- Checking AI output often takes as long as writing code manually, and AI-assisted development has led to embarrassing production incidents that negate supposed time savings
- A coding tutor warns that vibe coders who cannot explain their own code will be close to useless in real-world programming jobs, creating a generation of below-mediocre developers