The unreasonable effectiveness of an LLM agent loop with tool use

I've also been defining agents as "LLM call in a while loop with tools" to my co-workers as well — I'd add that if you provide it something like a slack tool, you can enable the LLM to ask for help (human in the (while) loop).