A stateless agent makes each decision from the current input only, while a stateful agent carries useful information across steps so later decisions can depend on earlier ones.
This distinction affects how you design AI agents, chatbots, workflow automations, and any system that acts over multiple turns or tasks.
In practice, most real agents are a mix: the model call may be stateless, while the surrounding application stores state.
A stateless agent does not rely on hidden memory between calls. Given the same prompt, tools, and context, it should behave the same way. If it needs prior information, that information must be passed in again explicitly.
A stateful agent keeps state somewhere outside the model call, such as in a database, session object, task graph, conversation history, or memory store. That state might include:
The key idea is not whether the model has “memory” in a human sense. It is whether the system preserves and reuses information across steps. Many agent frameworks make this explicit by separating the model from the orchestration layer.
Stateless
If the user later says, “Make it shorter,” the system must send the document or prior summary again.
Stateful
