Meet Grok Bot and Stand Up Your First Teammate

Grok Bot is not a chat window that waits for you. It is a teammate with its own computer. You give it a job, it signs into the apps you already use, and it keeps working after you close the laptop. It comes back when something needs your approval.

That is the whole product, and it is also the whole trap. If you treat it like ChatGPT with extra steps, you will get drafts you could have typed yourself. If you treat it like a teammate with one job, a login, and a hard rule about what it may not do, it starts paying for itself on day one.

This series is the practical path. Part 1 gets one bot standing. Part 2 turns that bot into an email teammate that drafts and never sends. Later parts cover routines, follow-ups, multi-bot threads, and the jobs you must never hand over.

What Grok Bot is, and what it is not

Three xAI products get mixed up. Keep them separate:

• Grok is the chat model. Fast answers, image and video, connectors. It does not keep a computer running when you leave.
• Grok Build is the terminal coding agent. You approve a plan, it edits a repo.
• Grok Bot is the teammate product. Each bot has a cloud computer, can sign into websites the way you do, can run in parallel, and can save a workflow after you show it once.

xAI's own product page puts it in one line: bots "sign in to your tools, use them just like you do, and come back with finished work." The same page says they run in parallel, 24/7, even when your laptop is closed, and that you can ask a bot to follow along once and save the steps as a routine. Those are vendor claims. They are also the features this series actually uses.

Who can use it today

Check x.ai/bot before you assume your plan includes it. As of 27 Aug 2026 that page lists Grok Bot on Cursor plans from $20/month (with weekly usage included), SuperGrok plans from $30/month, and Cursor Teams from $40/seat/month. Eligible existing Cursor, SuperGrok, and Teams plans include it. The page also shows a Get started for free path. Weekly caps and plan names still move, so treat those figures as a snapshot, not a contract.

You will want the desktop app for the first setup. iOS shares the same thread once the bot exists, which is useful for approvals on the go and a poor place to do the first Gmail login.

Create one bot, give it one job

Do not start with a "do everything" bot. A generalist with access to mail, calendar, Slack, and your bank will either stall or do something you did not mean. Create one bot. Name it after the job, not after the model.

Good first names: Inbox, Follow-ups, Research, Expenses. Bad first names: Assistant, Grok, Everything.

Then write the job in one paragraph, in the same message you would send a new hire on day one. Include all four of these:

• What done looks like. "A labelled list of what needs me today" is a job. "Help with email" is not.
• Where the work lives. Name the app. "Gmail" beats "my inbox."
• What it must bring back for approval. Drafts, yes. Sends, no.
• What it must never do. Send, delete, buy, share a password, or invent a fact it cannot point at.

Here is a first-job brief you can paste and then tighten:

You are Inbox. Your job is to read my Gmail and come back with a short brief, not to send anything.

Every morning, give me:
1. A count of unread and what can wait
2. The five threads that actually need me, each in one sentence
3. Draft replies sitting in Gmail for those five, unsent
4. Anything that looks like a bill, a school deadline, or a legal notice

Never send, never delete, never archive a thread I have not seen. If you are unsure whether something is urgent, put it in the brief and say why. If you cannot see a message, say you cannot see it. Do not guess.

That brief is already the email bot in part 2. If you want a smaller first win, swap Gmail for a single spreadsheet or a single help-desk queue and keep the same four rules.

Sign it into one app. You do the login

The product page is explicit: you log Grok Bot in once, and it uses your apps and websites the way you would, including tools that have no clean API. That login is yours. The bot can open the site. It cannot pass your 2FA, and you should not type your password into a chat window for it.

When the bot hits a sign-in screen, you take over, complete the login and the second factor, then hand the computer back. Stay on one account for the first week. A bot signed into work Gmail and a personal calendar in the same afternoon is how context leaks start.

If the site offers an app password or a less-privileged login, use that. If it does not, use the real account and treat the bot like a contractor who can see the inbox: it can draft, it cannot speak for you.

The first task should end in a review, not a send

Give the bot one closed task for the first hour. Examples that work:

• Open Gmail. List the unread threads from the last 24 hours. Do not open anything older. Do not draft yet.
• Open the help-desk queue. Screenshot the five oldest tickets and write one sentence each.
• Open the CRM. Find the three deals with no activity in 14 days. Do not email anyone.

Read what it brings back before you teach it anything else. You are checking three things: it reached the right account, it did not invent threads that are not there, and it stopped where you told it to stop.

If the brief is sloppy, the bot will be sloppy. Fix the brief. Do not add a second bot to paper over a vague first one.

What good looks like after day one

You are done with part 1 when all of this is true:

• One named bot exists, with one job paragraph saved in the thread.
• It is signed into exactly one account you meant to give it.
• It has finished one task and stopped for your approval.
• Nothing was sent, paid, deleted, or posted.

That is enough. The next page is the email bot: the same rules, applied to the inbox that actually eats your mornings.