Stateless (AI Agent)

Stateless refers to a system that carries no information forward between discrete requests or sessions. An AI agent is stateless across sessions by default: each new session begins empty, with no memory of prior interactions, because the model has no way to retain information between separate API calls or separate harness invocations. Between individual model provider requests within a session, the entire context window is resent — the model has no persistent internal state.

This statelessness is a fundamental property of how transformer-based language models operate. Each request is processed independently given its input context. The model cannot learn from a conversation in real time or update its weights based on a user's feedback without a separate training or fine-tuning step.

The implication is that any information a user wants the agent to retain must be explicitly reintroduced in each session — either by pasting relevant context at the start, by writing it to a memory file like AGENTS.md that the harness loads at session startup, or by relying on external systems like vector databases for retrieval. Without such mechanisms, the agent resets completely between sessions.

Open questions include how to give agents genuinely stateful memory without requiring manual context management, and whether context management files can be reliably kept in sync with the evolving state of a codebase or project without becoming a maintenance burden themselves.