harness + llm = agent
Nowadays most “LLM” endpoints include some sort of server side harness as well, and I’d bet more than one model involved, so it’s really just agents all the way down