AI is great for prototyping, but that is far different to AI in production-grade software, including with the hidden cost of maintenance. You have to know what you are doing.

Why even risk using AI directly in mission critical high risk software powering cars, planes and financial transactions or control systems with no human oversight?

If a disaster happened and an investigation was launched and the inquiry found that the software was "vibe coded" and no-one understood the code, would that look great towards the software vendor's reputation?