Good Epoch analysis of how (in)effective agentic AI is at real jobs, not merely benchmarks, math problems, or writing code.

The answer, of course: Not very good. But, as ever, two things:

1/ This is the worst agentic AI will ever be, which is not true wrt humans.
2/ This should be a good reminder that current inference projections are wildly over-aligned with code, which is endogenous.

https://epochai.substack.com/p/how-close-is-ai-to-taking-my-job