What is a long-running artificial intelligence agent?
Definition
A long-running artificial intelligence agent operates beyond a short request-response interaction. It may maintain plans, task state, memory, credentials, and scheduled work while interacting with tools or other agents. Its lifecycle can span many model calls and external events.
Persistence enables complex projects, but it increases the need for checkpoints, budgets, permission expiry, audit trails, and recovery controls. Operators must be able to determine whether the original goal is still valid, whether the environment changed, and when the agent should stop or request human review.
Acronyms and aliases
long-running AI agent variantpersistent AI agent variant
General terms
Related terms
Frequently asked questions
How does a long-running agent retain continuity?
It stores durable task state, decisions, observations, and pending actions outside any single model call, then reloads that state as work continues.
What is a major risk of long-running artificial intelligence agents?
Outdated goals or accumulated errors can persist over time, so agents need bounded permissions, checkpoints, monitoring, and reliable stop conditions.
