More is being produced. Less is being trusted.
Throughput is up and everyone can feel it. So are the incidents. Reviews take longer because there is more to review and less confidence in it. Somebody has started saying the quiet thing: we are shipping faster and it is getting worse.
Speed did not cause this. It revealed it.
If your tests were thin, your observability partial, and your definition of done aspirational, that was already costing you, quietly, at the old pace. Raise the volume and the same weaknesses produce failures faster than you can absorb them.
AI did not introduce the fragility. It removed the slack that was hiding it.
The instinct to slow down is the wrong one
Throttling output puts you back where you were, minus the investment. The fix is the discipline that lets speed be safe: tests that actually gate, telemetry that tells you what happened, small changes, fast recovery.
None of that is new. It is the engineering excellence work that was worth doing before AI and is now non-negotiable.
Where to start
Usually with an honest look at what your delivery system does when something goes wrong. How fast you find out, how fast you recover, and what your tests genuinely prove.
We do that work with your teams, on your estate. Sometimes the answer is that you need engineering discipline before you need any more AI, and we will say so.