“Agentic” shit is dangerous for two reasons:

• It gets a lot of things wrong.
• Local computers usually don’t have nearly enough oomph to run a decently-capable LLM locally, so your “agent” lives in the cloud. And in this age of surveillance capitalism and wanton corporate data exploitation, it basically acts as a trojan to exfiltrate your data and serve it on a platter to the monopolistic Big Data sumbitches.

As long as those two major issues aren’t addressed, AI automation - or indeed AI-anything - can fuck right off.