Chain-of-thought loses value for steering agentic AI, Ethan Mollick says

Ethan Mollick says on X that chain-of-thought, even in summarized form, has lost much of its value for steering and observing agentic AI. He says people now rely on asking agents for interim outputs and tests, then using AI to troubleshoot the problems those outputs reveal.

The post is the only report in this story so far, and it gives no data or examples.