A chatbot responds to messages; a fixed workflow follows steps you coded in advance. An agent uses an LLM to decide its own next steps: it plans, calls tools, observes results, and repeats until the goal is met.
The core pattern is a loop (such as ReAct: reason, act, observe) with memory, tools, and stop conditions.
Prefer a simple workflow when the steps are known; use an agent only when the path genuinely varies, since agents cost more, are slower and harder to test.
Loading…