LLM is not currently an agent (it would take a massive amount of compute that we don't have extra of at this time), but Microsoft as already wrote a paper saying we should develop agent layers to see if our models are actually general intelligences.