The DORA 2025 report landed and it confirmed something I’d been watching at client after client.
AI is making your teams faster. It’s also making your worst teams break things more.
Both are true at once. That’s the part nobody wants on a slide.
What the data actually says
DORA surveyed around 5,000 people, plus a hundred-plus hours of interviews. The headlines:
- 90% of teams have adopted AI in some form, up 14 points in a year. It’s not coming. It’s here.
- 80%+ say it boosts their productivity.
- And 30% have little or no trust in the code AI produces. They use it anyway.
Now the reversal. In 2024, AI correlated negatively with throughput. It was slowing teams down. In 2025 that flipped. AI now correlates positively with throughput and with product performance.
But one thing didn’t flip. AI still correlates negatively with delivery stability.
Translation: AI helps you ship more, faster, and it makes more of what you ship break.
”AI is an amplifier”
That’s DORA’s own framing, and it’s the most useful sentence in the report.
AI doesn’t make a good team good or a bad team bad. It amplifies what’s already there.
Give a team with strong testing, real version control, fast feedback loops, and good observability an AI assistant, and they ship more good code. The guardrails catch the AI’s mistakes the same way they catch human ones.
Give the same assistant to a team with flaky tests, manual QA, slow CI, and a green-dashboard observability gap, and you get more code flowing through a pipeline that can’t tell good from bad. More volume, same broken filter. The instability isn’t an AI problem. AI just turned up the dial on a problem you already had.
DORA’s line: AI accelerates development, and acceleration exposes weaknesses downstream.
Where this shows up
I have the same conversation now almost monthly.
A VP of Eng tells me velocity is up and they’re thrilled. Then change failure rate crept up too, and MTTR with it, and now their best engineers spend more time cleaning up after fast-shipped changes than they did before. Net productivity is flat or down, and nobody can quite explain why because the commit graph looks amazing.
The commit graph was never the problem. The thing that catches bad changes before customers do, that’s the problem. AI just made it matter more.
What the report says to actually invest in
The fix is not more AI. It’s the unglamorous capabilities that turn throughput into stable throughput.
Fast, trustworthy feedback. If your tests are flaky or slow, AI-generated volume overwhelms them. Tests you don’t trust are worse than no tests when the code is coming twice as fast.
Observability that catches business-logic failures, not just 500s. More change volume means more silent breakage, the 200 OK that didn’t actually work. If your dashboards stay green while checkout breaks, AI just gave you more ways to break checkout quietly.
A quality internal platform. DORA found 90% of orgs have adopted at least one platform, and that a good internal platform is the key multiplier for getting ROI out of AI. The platform is what makes the safe path the easy path at the new speed.
The part nobody wants to hear
If you’re rolling out AI coding tools and your delivery stability is sliding, the answer your vendor wants you to hear is “you need more AI.” The data says the opposite.
You need the boring infrastructure that catches mistakes, tests, feedback loops, real observability, strong enough to keep up with how fast you’re now moving. AI raised the ceiling on your throughput. Your safety net decides whether that’s a good thing or a very expensive one.
Fix the net first. Then go fast.
The cost of letting bad changes reach production, and what actually reduces it, is in my reliability costs breakdown.
— Youn