The DORA Report Was Right: AI Is an Amplifier, Not a Fix
The 2025 DORA research, retitled the State of AI-Assisted Software Development, drew on nearly 5,000 technology professionals and landed on a thesis worth sitting with: AI is not a solution in a box. It is an amplifier.
For teams with solid foundations, it accelerates work that was already flowing. For teams carrying technical debt, unclear process, and weak testing, it magnifies exactly those problems — faster.
The productivity paradox
Telemetry from separate research showed the shape of the problem clearly: individual developers completed substantially more tasks and merged far more pull requests with AI assistance, while organizational delivery metrics stayed flat. Work moved faster into the pipeline and then queued somewhere else.
That is a systems problem, not a tooling problem. If review capacity, environment availability, or release cadence was the constraint before, generating code faster does not touch it.
What changed in the metrics
The framework itself evolved. Rework rate — the share of unplanned deployments made to fix user-visible issues — joined the classic four, and it is the metric to watch when AI-assisted output increases. Reliability remains a quasi-metric anchored on service level objectives. And the old elite-to-low performance bands gave way to a set of team archetypes that combine technical and human factors, which is a more honest model of how organizations actually differ.
Platform engineering is the multiplier
The platform capability most correlated with a good developer experience in the 2025 data was unglamorous: giving clear feedback on the outcome of a task. Not a service catalogue, not a portal — knowing whether the thing you just did worked, quickly and legibly.
- Get the foundation in order before scaling AI use. Version control discipline, automated testing, observability. AI on a shaky base widens the cracks.
- Define trust boundaries. Engineers should know which AI output ships and which gets reviewed, written down rather than assumed.
- Invest in rollback. The faster you ship, the better your brakes need to be.
- Measure the whole pipeline. Individual throughput gains that never reach production are treadmill miles.
AI does not create high-performing organizations. It reveals which ones already were.
How we work
On client engagements this shapes sequencing more than tooling. We get delivery fundamentals working — environments, tests, deployment path, feedback loops — before layering AI-assisted development on top, because the alternative is measurably shipping defects faster. It is a less exciting order of operations and it holds up better six months in.