Survey: 93% of large North-American IT shops have hit an AI-coding incident

The number

A survey published today on DevOps.com reports that 93% of 406 IT decision makers at North American organizations with more than 250 employees have experienced at least one infrastructure incident caused by reliance on AI coding tools. The sample is North American only, and the size cut is 250-plus employees. Startups and mid-market shops sit outside the frame. Read at face value, the AI-coding incident is now a baseline event at large IT.

That changes the conversation CI/CD owners have been carrying on their own. Until now the case for harder review on agent-generated changes was thin: anecdote against vendor demo. 93 out of 100 large IT shops is a number that fits on a slide in front of a steering committee.

What it doesn't tell you

The respondents are IT decision makers, not the on-call engineers paged at 3am, so the operational shape of these incidents (rollback time, blast radius, whether the offending change came from an autonomous agent or an autocompleted line) sits outside the headline. "Caused by reliance on AI coding tools" is also a wide net. A merge of a hallucinated import is one thing. A refactor pushed through a stale test suite is another. They call for different controls.

So the practical read is narrower than the headline. The 93% says the argument has shifted from whether AI-assisted code can break production at large IT shops, per DevOps.com, to which gate stops it and how much of the existing CI suite is enough. That second half is what teams will be answering with their own incident postmortems in the meantime.