AI Agents

An agent combines a model with actions and observations. In ReAct, researchers interleaved reasoning traces with task-specific actions, using external sources and interactive environments to gather information and update plans (Yao et al., ReAct). For a builder, that loop creates concrete design choices: which tools to expose, what results to return, how to preserve state and when to stop. Start with a bounded task whose outcome you can inspect before broadening the agent's scope.

episode 2 with Weaviate, Unstructured and CrewAI compares agentic workflows and architecture. João Moura in episode 29 discusses the production stack around agents, including provisioning and measurement. What makes an AI agent? establishes the boundary with a single model response; AI agent autonomy levels helps define operating scope. Continue to how AI agents use tools for the interface between a model's decision and an application's action.