The DORA 2025 "State of AI-assisted Software Development" report is out, and its conclusions confirm what I've been observing in the field for two years: AI is not just helping developers, it is redefining the entire software value chain. But not in the way most organizations imagine.
The numbers that change everything
Three statistics from the report deserve serious attention.
90% of development professionals now use AI, a 14-point increase from 2024. In one year. This is no longer a trend, it is a standard. If your organization is not integrating AI into its development workflows, you are already behind — not on innovation, on the baseline.
+80% perceived productivity, +59% perceived code quality. Developers themselves report a qualitative and quantitative leap. The word "perceived" matters — I'll come back to that — but the direction is undeniable.
+7.2% instability. And this is where the report gets truly interesting. AI amplifies everything: strengths and weaknesses alike. A team with strong practices (testing, code review, CI/CD) sees its gains multiplied. A team with fragile practices sees its problems accelerated.
This is the most important conclusion in the report, and it explains a parallel MIT study finding: only 5% of companies truly succeed at structurally integrating generative AI.
Why 95% of companies fail
On the ground, I see three recurring failure patterns.
Adoption without integration: developers use ChatGPT or Copilot alongside their existing tools. It is a local optimization, not a transformation. Generated code piles up without architectural coherence. Individual productivity gains translate into collective complexity.
The productivity bias: the +80% perceived productivity is real at the individual level. But at the system level, the equation is different. More code produced means more code to maintain, more tests to write, more reviews to conduct. Without process adjustment, individual gains become collective bottlenecks.
The absence of measurement: most organizations don't measure the real impact of AI on their delivery. They know developers use AI. They don't know whether it actually improves lead time, defect rate, or user satisfaction.
Value Stream Management as a force multiplier
DORA strongly recommends Value Stream Management (VSM) as the integration approach. And this is exactly what I have been advocating in my engagements since 2023.
VSM acts as a force multiplier for AI. Instead of optimizing isolated tasks, you map the entire value stream — from idea to production deployment — and identify where AI creates the most systemic impact.
Teams that map their value streams see their AI benefits amplified exponentially. This is not hyperbole: when you optimize the right bottleneck, the leverage is massive. When you optimize a task that isn't on the critical path, the impact is zero.
This approach is also at the heart of Scaled Agile's SAFe AI-Native training, which integrates VSM as the framework for deploying AI at scale.
The four VSM principles for succeeding with AI
1. Map collectively: bring together developers, ops, product, and business stakeholders to identify the real bottlenecks. Not the ones management imagines, the ones teams experience daily. This step often reveals that the bottleneck isn't development, but code review, approvals, or deployment.
2. Optimize flow, not resources: target system constraints. If your CI pipeline takes 45 minutes to run, accelerating code generation changes nothing. Goldratt's Theory of Constraints applies perfectly here.
3. Data-driven continuous improvement: experiment and adapt. Measure classic DORA metrics (lead time, deployment frequency, MTTR, change failure rate) before and after AI integration. No decisions based on gut feeling.
4. Technical excellence as foundation: solid platforms before adding AI. If your tests are brittle, your CI unstable, and your code spaghetti, AI will amplify the chaos. Clean up first, augment second.
My field experience: what actually works
I have been using this VSM + AI approach in my engagements since 2023. Three concrete results:
Use case volume multiplied tenfold vs. traditional brainstorming. A VSM mapping workshop with AI as co-explorer generates 5 to 10 times more integration opportunities than a traditional brainstorm. AI suggests optimizations that domain experts hadn't considered.
Precise before/after impact measurement. VSM provides the baselines. AI provides the levers. The combination allows you to quantify each integration: "AI on test generation reduced lead time by 23% on this value stream." Not "AI helps us."
Co-construction support for continuous evolution. The VSM mapping becomes a living artifact. Each sprint, you measure, adjust, add new AI integration points. Improvement is continuous, not one-off.
The real test: are you in the 5%?
The DORA 2025 report poses an uncomfortable question to every organization: is AI structurally integrated into your value streams, or is it being used opportunistically by individuals?
Warning signs:
- You don't measure AI's impact on your DORA metrics
- AI usage varies by developer, with no team standard
- You haven't mapped your value streams to identify optimal integration points
- Your CI/CD pipeline hasn't been adapted for AI (quality gates, generated tests, assisted code review)
- There is no governance over the quality of AI-generated code
If you check three of these five, you're in the 95%. The good news: the path to the 5% is well-marked. VSM, DORA metrics, technical excellence, governance. It's not a revolution. It's a methodical integration.
I help organizations structurally integrate AI into their development workflows. If you want to assess where you stand and identify your next steps, let's talk.