When I was taking an MIT AI course (in ancient pre-LLM times), an autonomous agent was defined as a system that perceives its environment and acts on it (we were focusing on reward-expectation-maximizing agents, but it's not that important). Peter Norvig has said something like, technically, anything can be described as an agent (a rock maximizes the "follow physical laws" objective), but naturally, it doesn't make much sense to model a rock as an agent. With AI agents, the situation is significantly less controversial: they do perceive, deliberate, and act.

All of these terms were picked by individuals, years ago, while reaching for metaphors that made sense to them personally.

None of these "agent" / "thinking" / "reasoning" terms were dreamed up in boardrooms to intentionally mislead people. They are useful but faulty metaphors; there is no conspiracy.