AI writes code faster. Your delivery system decides what that costs

DORA’s 2025 research reports that 90% of technology professionals now use AI at work and more than 80% believe it has increased their productivity. It also reports, in the same dataset, that higher AI adoption correlates with higher software delivery throughput and higher software delivery instability at the same time. Both are true, and the tension between them is the finding.

Where the time actually goes

One of the more useful details in the research is that time saved during generation is often reallocated to verification, with about 30% of respondents reporting little or no confidence in AI-generated code. That is not an argument against the tools. It is an argument for noticing that the bottleneck has moved: if generating a change is now cheap and reviewing it is not, review is the constraint, and adding more generation makes the queue longer rather than shorter.

What has to be in place for the speed to be worth having

• —Automated tests that a reviewer trusts enough to rely on, covering the paths that actually break.
• —Small changes, so a failure is attributable and a rollback is boring.
• —Fast feedback: a pipeline that reports in minutes, because a slow one gets bypassed under delivery pressure.
• —Version control discipline and a documented rollback path for every deployable change.
• —A review standard that applies equally regardless of what generated the code.

Measure the pair, not the headline

Throughput on its own will flatter almost any adoption programme. Read it alongside change failure rate and time to restore service, and the picture becomes honest: rising deployment frequency with a rising proportion of changes causing failures is not an improvement, it is a transfer of work from development into operations and support. Teams that track both make the trade visible early enough to correct it.

The organisations getting a genuine return are not the ones that adopted earliest. They are the ones whose delivery practices were sound enough that going faster was safe.