The Agentic Loop: Three loops in a trench coat
Summary
Robert Ross’s The Agentic Loop argues that autonomous agents are built from three intertwined loops: an inference loop for handling chat completions, a tool loop to execute external actions, and a human loop for oversight and approval. The post provides pseudo-code examples and discusses the production challenges of building agentic systems, emphasizing that the real value lies in the tool interactions rather than the LLM alone, and highlighting considerations for durability and safety.