Some links on this page are affiliate links. If you buy through them, we may earn a commission, at no extra cost to you. That does not change the verdict. Read the full disclosure

Advanced20 minGrok BotProductivityUpdated 2026-09-24

Grok Bot: first scheduled routine (2026)

If you finished grok-bot-primer-teammate, you already have a Bot whose handoffs deliver a stable format. This tutorial is your first scheduled routine: a prompt that runs on its own in the Grok Bot app even when you are away — via cron or event listeners. Grok Bot means xAI/Cursor teammates plus Agent Computer (cloud VM); it is not Grok web chat. Official onboarding: https://cursor.com/bot/onboarding. Do not copy a routine from another Bot or schedule at 3 a.m. «because it is quiet».

The problem it solves

Without a routine, you repeat the same handoff every Monday. With bad intent or a bad schedule, the Bot runs overnight, opens the wrong machine state or executes another profile's job. The symptom is «nothing happened» or «something I did not ask for happened».

What you will learn

  • Pick a job that already succeeded twice the same way by hand
  • Write routine intent: goal, format and limits — not a frozen click list
  • Set sensible weekday daytime cron (Europe/Madrid mindset)
  • Save the routine on the correct Bot and confirm timezone
  • Let it fire once and find the run in history
  • Verify the deliverable and tune intent if needed
  • Avoid typical mistakes: overnight cron, copied Bot, grok.com confusion

Before you start

  • Completed grok-bot-primer-teammate (at least one successful handoff with stable format)
  • SuperGrok Plus/Heavy or Cursor Pro+, Ultra or Teams
  • Grok Bot app on macOS, Windows or iOS (not grok.com)
  • A repeatable job with a reviewable deliverable (bullets, draft, table)
  • 20 minutes to schedule, wait for one fire and fix
  1. Steps 1

    Pick a job that already repeats the same way

    Open the Grok Bot app (not Grok web chat) and identify which handoff you ran twice with the same outcome: internal PDF summary, read-only inbox triage, metrics snapshot in a dashboard. A routine does not fix an unstable flow — it only automates what is already predictable. If you have two Bots, pick the profile whose job matches that work; do not schedule on the other job's Bot.

    What you should see: A note with the Bot name, deliverable (e.g. «five bullets») and dates of two successful manual runs.

  2. Steps 2

    Write routine intent (goal, not recipe)

    In that Bot's routine editor, describe the goal like you would to a teammate: which sources to check, output format, what not to touch (production, sends, deletes) and where to stop («leave a draft and wait for my OK»). Avoid pasting a rigid tool sequence («open Chrome, click X, click Y») — Agent Computer picks steps when the goal is clear. Add quality bars: «if there is no new data, reply «no changes» in three lines».

    What you should see: An 8–15 line intent block with goal, format, limits and stop point. No click list copied from a single run.

  3. Steps 3

    Schedule weekday daytime hours (Europe/Madrid)

    Pick cron or recurrence on a weekday at a time you usually review results — for example Tuesday and Thursday at 09:30, thinking in Europe/Madrid (UTC+1 winter, UTC+2 summer). Avoid 02:00–05:00: nobody reviews, failures pile up and the Bot may leave the VM in a odd state until your next session. If the app shows timezone, confirm it; if you travel, adjust before the next fire. For the first test, add a one-off run in the next 30–60 minutes in addition to the weekly cron.

    What you should see: Calendar or picker with Mon–Fri, time around 09:00–11:00, Europe/Madrid zone (or explicit equivalent) and a nearby test fire.

  4. Steps 4

    Save the routine on the correct Bot

    Before saving, confirm the sidebar profile is the job you automate — not the second teammate or an empty test Bot. Name the routine for the work («Monday-CRM-brief»), not «Routine 1». Check notifications: email or push if your plan offers them, so you know it fired while you were away. Save and confirm it appears only in that Bot's routine list.

    What you should see: The new routine under the chosen Bot, active state, next run with readable date/time. Other Bots without a duplicate copy.

  5. Steps 5

    Let it fire once (no micromanaging)

    Close the app or move on — the point is an autonomous run. Do not open Agent Computer unless the routine needs a login you already prepared in a prior run; if new credentials are required, reschedule after a manual handoff where you log in. When the time hits, the Bot starts on the cloud VM. If you added a test fire, wait for that window before trusting the weekly cron.

    What you should see: Notification or run history entry with start time, correct Bot and status «running» or «completed».

  6. Steps 6

    Verify the deliverable like a PR

    Open the run result: chat, attachment or summary in the app. Check format (bullet count, tone), no sends or production changes, and dates/figures match the source. If the Bot left a draft for approval, do not send until you read it. Agent Computer may keep tabs open — close them if the next run is another job or Bot.

    What you should see: Deliverable aligned with your two manual runs. Explicit «waiting for OK» if you asked for it in intent.

  7. Steps 7

    Fix typical mistakes and tune

    If it did not fire: check timezone, paused routine and eligible plan. If the wrong job ran: delete the copy on the other Bot and rewrite intent on the correct profile. If it ran at night unchecked: move cron to business hours. If you tried to «schedule Grok» on the web: Grok web chat has no routines or Agent Computer — return to https://cursor.com/bot/onboarding. Tune intent (tighter limits, «no changes») and schedule another test fire before trusting the weekly cron.

    What you should see: A short list of applied fixes (time, Bot, intent text) and a second test run scheduled or completed successfully.

Real use cases

Monday brief without opening the app

Every Monday at 09:00 the Bot summarizes tickets and CRM in five bullets. You read the push on the commute and approve drafts later in Agent Computer.

Biweekly competitor radar

Biweekly office-hours routine: read PDFs in a shared folder and return a comparison table. No external sends.

Dashboard health check

Wednesday and Friday at 10:30: capture metrics on a dashboard with login already set, attach screenshot and three alerts if something crosses threshold.

Common mistakes

  • 3 a.m. cron «to stay out of the way»

    Nobody reviews; failures stack and the VM stays dirty. Use weekday mornings in your zone (e.g. Europe/Madrid).

  • Routine on the wrong Bot

    Fires the other teammate's job on the shared machine. Create the routine only under the profile for that work.

  • Click recipe instead of goal

    A UI change breaks the run. Write goal, format and limits; let the Bot choose tools.

  • Automating before two manual wins

    The routine amplifies instability. Run the same handoff twice by hand before scheduling.

  • Confusing Grok Bot with Grok web chat

    grok.com does not schedule teammates or a VM. The right product is the Grok Bot app; onboarding at cursor.com/bot/onboarding.

Close

Your first scheduled routine turns a stable handoff into work that happens without you — but only if you pick the right Bot, write clear intent, schedule human hours and review the first fire like a PR. Grok Bot is not Grok web chat; it is teammate plus Agent Computer. Tune schedule and intent after the first run before you trust recurring cron.

Next steps

  1. Let the routine run two weeks on the same schedule and note whether the deliverable stays stable.
  2. If you add a second job, keep routines per Bot following grok-bot-segundo-teammate.
  3. When work shifts back to code in your repo, use Cursor in parallel — it does not replace Bot routines.

Free guide

Want the 15-tools guide?

We send it free. We will also ping you when the next guide in this path goes out.

Unsubscribe whenever you want. We do not sell the list.

Tools in this guide

Read next

You may also like