You could make an "agent" that is powered by an LLM, you could make an "agent" based on production rules, expert systems and all that. You could combine these things.
A critical thing is the mixed initative nature. It is not like you type
> 5 + 5
and it says
10
but rather you have a conversation that goes back and forth and then it takes some action in the world and gets notifications and does things and then at some point gets back to you to tell you what it found or ask what you think about this or that.
agent is an AI buzzword. It's a black box for a non-deterministic system, with some deterministic connections and wrappers around the whole thing to simulate "security" and "guardrails"
I think an "agent" is a role, not a technology.
You could make an "agent" that is powered by an LLM, you could make an "agent" based on production rules, expert systems and all that. You could combine these things.
A critical thing is the mixed initative nature. It is not like you type
and it says but rather you have a conversation that goes back and forth and then it takes some action in the world and gets notifications and does things and then at some point gets back to you to tell you what it found or ask what you think about this or that.agent is an AI buzzword. It's a black box for a non-deterministic system, with some deterministic connections and wrappers around the whole thing to simulate "security" and "guardrails"
Software that can screw things up on it's own.