A long-running agent is an autonomous software worker that owns a single goal across days or weeks, taking many actions over time — calls, texts, emails, and system updates — until the goal is met or it hands off to a person. Unlike a chatbot, it is not tied to one conversation; unlike a script, it reacts to what happens between its actions.
How it differs from a chatbot
A chatbot answers while you are on the page and forgets you when you leave. A long-running agent persists. If a borrower stops replying on Tuesday, the agent remembers the open thread, waits the right interval, and follows up on Thursday through the channel most likely to land. The unit of work is the job, not the message.
How it differs from RPA
Robotic process automation replays fixed clicks against a stable screen. It breaks when a portal changes or a call goes to hold. A long-running agent decides what to do next from the current state, so it can wait on hold, retry a different number, or escalate when the situation falls outside policy.
Where long-running agents earn their keep
The best fit is work that is multi-step, spans elapsed time, crosses channels, and has a definite finish line: an insurance renewal, a denied-claim appeal, a dormant account reactivation. These are jobs where the cost is not the individual action but the follow-through nobody has time to do.
Related reading and workflows: