Can I ask what kind of work, and what models? Because I don't, I'm a senior embedded engineer working with frontier models, and I haven't seen a major mistake in at least few months now, since earlier in the year.

It will sometimes get things wrong because it made non-viable assumptions, but that's on me and prompt accuracy.

What I see is: it makes tests against desired features or bugs or mistakes, and iterates until it gets it correct. I steer as it goes.

(I hate this, by the way, zero job enjoyment anymore, but I don't know what else to do.)