Google Cloud's DORA research team — the group behind the industry's most-cited software delivery benchmarks — published a follow-up this year to its 2025 State of AI-Assisted Software Development report, and the core finding hasn't really changed: AI acts as an amplifier, not a fix. Teams with strong underlying practices get measurably faster with AI tools. Teams without them tend to get faster in isolated pockets while the overall system gets messier.
DORA lead Nathen Harvey put it plainly in the report's framing: the biggest returns come not from the AI tools themselves but from the quality of the internal platform and the clarity of team workflows underneath them. Without that foundation, the report argues, AI creates pockets of individual productivity that get lost the moment the code has to integrate with everything else.
That tracks with what a lot of engineering leaders have already noticed anecdotally — a developer moving faster with an AI assistant doesn't mean the pull request review queue gets shorter, or that the technical debt older code already carries gets any easier to work around. If anything, faster individual output puts more pressure on the review and integration process, which usually wasn't the bottleneck to begin with.
The practical read for a smaller team without a dedicated platform engineering group: before investing heavily in AI coding tools, it's worth an honest look at whether the actual bottleneck is developer typing speed, or something upstream of that — unclear requirements, a slow review process, deployment friction. AI tools are very good at making a developer faster at the part of the job that usually wasn't the slowest part.