I don't think that's entirely correct. You can do TDD style development with AI and it leads to better results.

It's just that messing with a company's core manufacturing is something they don't do lightly. They work with multiple shifts of staff that are supposed to work in these environments. People generally don't have a lot of computer skills, so things need to be simple, repeatable, and easy to explain. Any issues with production means cost increases, delays happen, and money is lost.

That being said, these companies are always looking for better ways to do stuff, to eliminate work that is not needed, etc. That's your way in. If there's a demonstrable ROI, most companies get a lot less risk averse.

That used to involve bespoke software integrations. Those are developed at great cost and with some non trivial risk by expensive software agencies. Some of these projects fail and failure is expensive. AI potentially reduces cost and risk here. E.g. a generic SAP integration isn't rocket science to vibe code. We're talking well documented and widely used APIs here. You'd want some oversight and testing here obviously. But it's the type of low level plumbing that traditionally gets outsourced to low wages countries. Using AI here is probably already happening at a large scale.