Skip to main content
Module 3: Run unattended safely

The 3 a.m. problem: approvals expire

An approval for unattended work expires after about 10 minutes. A routine that needs you overnight doesn't wait — it stalls. Decide its ending in advance.

This is the failure mode that catches almost everyone once. From Grok Bot's approvals documentation: an approval for work that started without you — a routine, a trigger, or a message from another Bot — expires after about 10 minutes. The card then shows Expired, and the action does not run.

Play it out over one night. You're asleep for eight hours. A routine that needs your approval to finish fires, waits about ten minutes for a tap that never comes, and expires. On a five-minute schedule it does this again and again — and you wake up to a column of grey "Expired" cards. Nothing broke. Nothing got done either. The routine spent the night asking a question no one was awake to answer.

The key realization: there is no "wait until I'm back" option. It looks like there should be. There isn't. An unattended approval is a short-fused thing, by design — so that a Bot can't sit indefinitely holding a half-finished sensitive action. That's good for safety and fatal for a routine you expected to just work overnight.

So every routine that runs while you're away needs its ending decided in advance. There are exactly two good endings:

  1. Allow it to finish on its own. If the action is safe, reversible, and something you've already approved several times, pre-authorize it with an Allow automatically rule (we cover how in the next lesson). Then it completes overnight without asking. Helpfully, an Expired card offers "Always allow this in the future," which adds that rule for you — though note it adds the rule without running the action that just expired, so you'd ask the Bot to try the task again.
  1. Stop at a draft. If the action is not safe to run unattended — sending, publishing, purchasing, deleting, anything touching production — then design the routine so it prepares the work and leaves a draft or recommendation in the conversation, and you approve it in the morning. This is the docs' core principle restated: automate the preparation, not the irreversible action.

Which ending to pick isn't a judgment call you make fresh at 3 a.m. — it's a property of the action. Reversible and routine → allow automatically. Risky or one-way → draft. Decide it when you build the routine, not when you're staring at grey cards over breakfast.

Try it

For a routine you'd run overnight, list every action it takes. Sort each into "safe to allow automatically" or "must stop at a draft." If any action is risky and you can't make it a draft, that's a sign the routine shouldn't run unattended yet.

Stay in the loop

Enjoying the free lessons? Get an email when we publish new courses and updates — no spam, unsubscribe anytime.

Discussion (0)

Ask a question or share what worked for you. Comments are reviewed before they appear.

Log in to join the discussion and ask questions about this lesson.

No comments yet. Be the first to start the discussion!