The context
The software industry is experiencing an unprecedented transformation as AI tools become standard equipment in every developer’s toolkit. Corporate investment in AI hit $252.3 billion in 2024—a 26% increase from the previous year. Job postings mentioning generative AI skills grew by 323% in just one year. This gold rush is driven by a powerful assumption: that AI will transform not just individual productivity, but overall team and organizational performance.
However, early signals have been mixed. While developers report feeling more productive with AI, last year’s DORA report revealed an unexpected finding—teams using AI actually experienced lower throughput and more instability in their software delivery. This created a puzzle: how do teams apply AI in a way that gains in one area are not offset by losses in other areas? Now with a fresh batch of data, this year’s DORA has discovered new findings.
The research
Google’s DORA team surveyed nearly 5,000 technology professionals globally and conducted over 100 hours of qualitative interviews to measure AI’s impact on ten key outcomes, from individual effectiveness to organizational performance. The study employed Bayesian statistical modeling with explicit causal assumptions to estimate AI adoption’s effects while controlling for confounding factors like individual experience, available resources, and organizational stability.
Key Findings:
- AI now improves throughput but instability persists: AI adoption shows a positive relationship with software delivery throughput (reversing last year’s negative finding), but continues to increase delivery instability. This suggests teams have adapted to generate code faster with AI, but underlying systems haven’t evolved to safely manage AI-accelerated development.
- Individual effectiveness increases while friction and burnout remain unchanged: Higher AI adoption correlates with increased individual effectiveness, more time spent on valuable work, and improved code quality. However, AI shows no relationship with reducing friction or burnout, indicating these challenges stem from organizational systems rather than individual productivity.
- Organizational and product performance improve: The research found positive associations between AI adoption and both organizational performance and product performance, suggesting benefits can scale beyond individuals when conditions are right.
- Stubborn problems reveal systemic constraints: Despite productivity gains, friction remained unaffected and instability increased. In the authors' view,“friction doesn’t vanish so much as move: It shifts from manual grind to deciding and verifying, possibly in the form of prompt iteration, result vetting, and assessing code that looks remarkably similar to correct code.”
- Sociotechnical impacts show pride increases: counterintuitively, increases in AI adoption has shown statistically significant improvements in authentic pride. Researchers hypothesize that this is because higher levels of AI adoption lead to a higher percentage of time spent on more valuable work, which increases feelings of pride.
The application
The research reveals a nuanced picture: AI is delivering measurable benefits for individual effectiveness and organizational performance, but its value is constrained by organizational systems that haven’t adapted to the new paradigm. The increase in delivery instability despite productivity gains suggests that simply adding AI tools without evolving processes, platforms, and quality controls creates new bottlenecks. Engineering leaders must recognize that AI’s impact depends less on the technology itself and more on the environment in which it operates.
Actionable steps for engineering leaders:
- Evolve your delivery pipeline to handle AI-accelerated output: The increase in throughput coupled with persistent instability signals a mismatch between code generation speed and quality controls. Invest in automated testing, enhanced code review processes, and platform capabilities that can safely handle higher volumes of changes. Don’t simply optimize for velocity—ensure your systems can maintain stability at increased speed.
- Address friction through system-level improvements, not just AI adoption: Since friction and burnout remain unchanged despite productivity gains, focus on the organizational factors that create friction: inefficient processes, unclear priorities, inadequate documentation, and coordination overhead. AI can accelerate coding, but it won’t fix broken workflows or reduce administrative burden without deliberate process redesign.
- Create feedback loops to measure AI’s actual impact on outcomes that matter: Track team-level data on throughput, stability, product quality, and team well-being. Use tools to identify where AI is creating value versus where it’s being absorbed by downstream bottlenecks.
As shared best by the authors, "without intentional changes to workflows, roles, governance, and cultural expectations, AI tools are likely to remain isolated boosts in an otherwise unchanged system—a missed opportunity.”
.webp)
