Scheduled and Background Agents
Agents that work while you sleep: scheduled runs, triggers, notifications, and keeping unattended agents safe.
Every agent you've used so far waits for you: you type, it works, you check. The next step is agents that work without being asked. A briefing on your desk at 7 a.m. every Monday. A competitor's pricing page checked daily, with a ping only when it changes. Every new support form summarized and routed before anyone's awake.
Unattended agents are where the time savings compound, and also where mistakes compound, because nobody is watching when something goes wrong. This lesson covers the ways to run agents on a schedule or in response to events, how to write instructions for a run nobody supervises, how to design notifications you'll actually read, and the guardrails that keep a background agent from becoming a background liability. You'll finish by setting up one scheduled agent of your own.
What changes when nobody is watching
When you use an agent interactively, you're the safety net. You notice when it misunderstands, you stop it before it does something odd, and you answer its questions. Take yourself out of the loop and three things change:
- The agent can't ask you anything. Every ambiguity has to be resolved in advance, in the instructions.
- Mistakes repeat. A bad rule in an interactive session costs you one bad result. A bad rule on a daily schedule costs you one every day until you notice.
- Silence is ambiguous. If you hear nothing, did the agent find nothing, or did it crash three weeks ago?
Good background agents are designed around these three facts. The rest of this lesson is how.