AI is increasing the speed of work and volume of output of the individual. We're now able to do more, across a wider set of expertise, than ever before. In many cases, the outputs are good enough to be acceptable, and in higher compliance domains the existing processes still provide the guardrails.
This works well for individual contributors, but is it organically working well for teams? Is the increase in throughput just adding more to the queue?
Do all teams benefit?
DORA provides the best evidence I've seen on this. Its 2025 State of AI-assisted Software Development report1 surveyed nearly 5,000 people. About 90% use AI at work, and more than 80% believe it has made them more productive. Throughput is up.
And so is delivery instability. Burnout and friction, the things you'd hope would fall if the work got easier, are flat. See also AI Doesn't Reduce Work - It Intensifies It2 in the Harvard Business Review.
If writing gets faster and everything after writing stays the same, more work arrives at the same reviewers, the same environments and the same approvals. The friction moves from writing the thing to verifying it, and verifying is harder to speed up, because someone still has to be sure.
A caveat, since it matters here: the survey is correlational. It shows these things move together, not that one causes the other.
What the better teams had
The report groups teams into seven profiles, and the best two, nearly 40% of the sample, deliver speed and stability together. DORA's companion AI Capabilities Model3 names seven capabilities that amplify the effect of AI. Four of them are mostly dull and mostly old:
- Small batches, so each change is easy to review and cheap to get wrong
- Version control used properly, so a bad change can be undone
- A clear focus on the user, so the team knows which changes matter
- A quality internal platform, so the safe route to production is automated
You'll notice that none of this is exotic, or even about AI. It's the same list a delivery lead would have given ten years ago. The assistants don't change what a good delivery system looks like but they do make the absence of one considerably more costly, to the team and customers as well as financially.
What leaders should do next
- Start with small changes, and small incremental updates to the delivery processes in parallel. An assistant makes it easy to produce a large change quickly, and a large change is the thing that stalls in review. The faster the keyboard, the more tempting it is to send the queue something it can't digest
- Prioritise using assistants to drive the process. The outputs are becoming commoditised and easier to produce, finding efficient ways to scale the process, governance and guardrails will be where the throughput improves next. Start with the current bottlenecks
- Manage expectations. DORA's ROI paper4 anticipates a J-curve, a temporary productivity dip and period of instability in early adoption, which it calls "the tuition cost of transformation". I'd argue the dip is necessary while the team builds the new process and builds the trust in that process
Efficiency doesn't just happen with the top-down directive to use more with AI, it comes from strong leadership guiding the whole team to realise the benefits.
References
- https://cloud.google.com/resources/content/2025-dora-ai-assisted-software-development-report | 2025 DORA State of AI-assisted Software Development report, DORA (Google Cloud)
- https://hbr.org/2026/02/ai-doesnt-reduce-work-it-intensifies-it | AI Doesn't Reduce Work - It Intensifies It, Harvard Business Review
- https://cloud.google.com/resources/content/2025-dora-ai-capabilities-model-report | Unlocking AI's full potential: 2025 DORA AI Capabilities Model report, DORA (Google Cloud)
- https://cloud.google.com/resources/content/dora-roi-of-ai-assisted-software-development | ROI of AI-assisted Software Development, DORA (Google Cloud)