The more I think about it, the more I think terms like "agent" are no longer useful.
Historically, before the industry redefined it in 2024, an "agent" was any long running stateful process. In safety research, we used to think of "agents" as problems that needed to be solved, because it usually meant that an LLM was doing something unexpected or unpredictable.
Today, agents are utilities. We use the term to describe smart functions, workflows, actors, chatbots with function calling capabilities, and yes... any long running stateful process involving an LLM, most of which are positive and useful things we're trying to do.
Oh, how far we've come.
The problem is, that "agent" is no longer a word with enough specificity to be helpful. In fact, we're over-using it the same way we overused "ai" back in 2019.
This makes it hard to hone in and communicate what you're doing without going into detail, and defeating the purpose of the shorthand to begin with.
I think we need to break it down into half a dozen different and totally new terms, for the simple reason that it would be easier to understand what people are talking about, when they're talking about their exciting AI apps here on LinkedIn.
They might get overused later...
but we're probably good til at least Q2.
Call me selfish.