LLM Powered Autonomous Agents
Overview
Lilian Weng's seminal deep-dive into LLM-powered autonomous agents — covering planning, memory, tool use, and the architecture of systems like AutoGPT and BabyAGI.
Full Description
Lilian Weng (OpenAI) wrote what became the definitive technical reference on LLM-powered autonomous agents. The article breaks down the core components of an agent system: the LLM as the central reasoning engine, planning mechanisms (ReAct, chain-of-thought, tree of thoughts), memory systems (short-term, long-term, external), and tool use (code execution, web search, APIs). It surveys AutoGPT, BabyAGI, HuggingGPT, and other early agent systems, and explains why building reliable autonomous agents is fundamentally hard. One of the most-cited non-paper AI references on the internet.