Why not?

He just inadvertently answered his own question as to why people resist using AI.

If AI is clearly not competent enough to trust for the "really important" stuff, what makes him think it is competent for other stuff?

His AI mandate is like forcing a team to hire cheap, incompetent coders with a language barrier --- and then blaming the team for the mistakes and lack of productivity.

Still a bad coder can "learn" to get better. Can AI do this? In the meantime, "fixing" bad code can be harder than writing good code --- and a lot less fun.

Next up --- let's just mandate the use of AI to fix AI's bad code. Problem solved?

There have been a few CEOs announcing that 'X% of our code is now written by AI', where X is some totally non-credible number to anyone that has used an LLM for this purpose.

But if you attach consequences to not using AI, engineers are just going to give AI more credit for the end product, no matter how true that is.

Being a ruthless, unforgiving parent only makes your children better at hiding things.