It's going to take decades and massive amounts of capitol investment to fully implement AI into mission critical, fault intolerant systems.
A classic example is lazy and stupid lawyers using AI to write briefs and memoranda. It seems invariably to create fake cases and get those lawyers into terrible trouble. It no doubt has similar deficiencies in other applications. Thus it is basically an immensely powerful idiot child with an absolutely literal way of understanding instructions, a way that can get ignorant, lazy people into terrible trouble.