I'm in the business of thinking.
It's popular to see LLM agents as automating thinking (a substitute for intelligence) but it's perhaps more useful for a practitioner to see them as automating making decisions (on the basis of the statistical evidence of their training corpus and the information available to them). Agents are good at making certain kinds of decisions, and not at other kinds.
Thinking this way is helpful for ensuring accountability when using LLMs and also helps people reflect: "what is the best tool for making this decision"? In my view, it's often not an LLM, it's often just talking about it with someone.
Someone might object: when LLMs are "reasoning" through a problem and finding constructive proofs to open maths problems, aren't they doing something like thinking?
It is for sure something like thinking, but arguably it is just a series of decisions being made about how to approach a problem. Some of which happen to bear fruit! If we throw hundreds of agents at a problem then some of them might decide to do what we want ;)