Grok Bot and sub agents never start their first routines at the beginning of the day. So, I realized I need to start the day by saying “wake up the sub agents” to the main agent, and then the main agent and all sub agents complete their routines.
I have 5 sub agents, so I thought maybe the issue was having the routines all starting at the same time. After staggering their routines by 10 minutes, the issue persists.
How can I get Grok Bot to execute routines every day without needing any human interaction to wake it up?
Steps to Reproduce
Have a Grok Bot and Sub agents with a routine to read different sectors of the news every morning at 6:00am.
Expected Behavior
I expect the Grok Bot to autonomously send me a news summary every morning without me needing to turn on the app or interface with it in any way.
I also don’t understand why my routines don’t finish on time.
For example, this routine is just a handoff of notes to another agent. It can’t possibly be computationally heavy. So why is it still thinking about it nearly 30 minutes later?
Hey @Ryan_Daley, your routines aren’t asleep or skipped, they’re firing. The issue is a queue: scheduled runs across Grok Bot wait in line before they start. Your 1:15 PM handoff didn’t actually begin until about 1:37, so at 1:42 it had only just started, which is why it looks like it’s been “thinking” for 30 minutes. It’s the wait, not the task being heavy. A direct message (“wake up the sub agents”) skips that queue, which is why that always works.
The wait is worst in the early-morning US window (roughly 5 to 9 AM Pacific), where runs land 20 to 45 minutes late, but it happens at other busy times too (your 1:15 PM run was about 20 minutes).
This is on our side and an issue we’re tracking. Two things that help now:
For time-sensitive morning jobs, schedule them earlier. Before about 4:30 AM Pacific the delay is usually under 10 to 15 minutes (routines also carry up to ~10 min of built-in randomization, so leave a little slack).
Since your prompts tell scouts not to catch up when late, they stand down silently. Have them send a one-line “skipped, ran too late” note instead, so you get a signal rather than silence.
Staggering by 10 minutes won’t help here. I’ll follow up when there’s an update.
@mohitjain , it sounds like the wait is related to how much traffic you have on your servers. I imagined that I had my own dedicated virtual machine, and that it would only be impacted by the traffic that I generate. I think that you are saying that I am being impacted by the overall traffic from all users who are engaging with Cursor and Grok. This makes me realize that Grok Bot can not reliably trigger jobs at an exact time, no matter how I set up my agents and their routines.
Thank you for the advice on how to best use the existing system. I will try to implement your suggested workflow.
As you say, I’m sure this is something that the team is actively working on, and I look forward to seeing improvements. Do you think this problem would be addressed by software updates? Or is this a fundamental problem of constrained compute?
