It's like the rest of Agentic AI - a lot of talk, but when it comes to the crunch, very few people are actually willing to hand over full autonomy to anything where significant value / liability is in play.

The truth is, a lot of the value of agentic AI in general is negated by the sheer power of agentic coding itself. If I can prompt an AI to write an actual deterministic process to solve a problem in a couple of minutes (still potentially AI assisted, just in a deterministic manner), why would I delegate it to a non-deterministic AI? You have to come up with a category of tasks where the actual process itself cannot be anticipated. Intersect that with high value processes and you whittled down to almost nothing. Not actually nothing but a far smaller category than people make out.