Grok Bot in Hong Kong: my rules for a team of bots

On 15 October I moved my recurring work off my laptop.

Not the thinking. The repeat jobs. Sorting email at nine. Checking tomorrow's meetings for a missing agenda. Reading the morning's AI news. Collecting event photos on Sunday night.

Those now belong to a small office of bots in Grok Bot. They work while my laptop is closed. None of them can send anything.

This post is the setup, the rules, and one pattern from my stage demos you can copy today.

What Grok Bot actually is

Grok Bot gives you bots with names, jobs and memory. Each one works on a persistent cloud computer with a browser, files and a terminal, so the work finishes in your real tools instead of as a chat draft. You talk to a bot by typing, dictating or voice chat, from the desktop app or your phone.

Four things make it different from a chatbot (source: docs.x.ai, Grok Bot overview, updated 7 Oct 2026):

It keeps working when your laptop is closed. Work runs on the cloud computer.
Bots talk to each other. They can share a group chat and pass a task along, so you are not the router between tools.
It learns by demonstration. Walk a bot through a multi-step job once and it can save it as a skill, then rerun it on a schedule as a routine.
All your bots share one computer. Same files, same browser logins. That makes handoffs easy. It also means anything you put on that computer is available to every bot you run.

That last point is the one most people skip. It shapes every rule below.

Can you use it in Hong Kong, and what does it cost?

I use it from Hong Kong.

Access does not come as its own subscription. As of 16 October 2026, Cursor's plan page says Grok Bot is included on the Premium, Super and Ultra tiers of The X Subscription, and on paid legacy Cursor plans. Plus and Free do not include it. You can also link an individual SuperGrok, SuperGrok Plus, SuperGrok Heavy or X Premium+ account.

Usage is a weekly pool that resets each week. When it runs out, extra usage is paid from credit if you have turned that on. The free trial is a usage credit, not a number of days, with a 7-day window. A big job can use most of it in one run.

Plans change often. Check cursor.com/help/grok-bot/plans before you buy, not this post.

One fact for any Hong Kong company: Grok Bot computers run in the United States today (docs.x.ai, Grok Bot security, updated 12 Oct 2026). If your data has to stay in Hong Kong, that decides it before anything else does.

My setup: one bot talks to me, the rest file reports

I have ADHD. Ten bots pinging my phone would be ten more things I ignore.

So only one bot talks to me. I call it Coach. It sends one action at a time, short, with A/B/C buttons. It goes quiet at night.

Every other bot is a desk. A desk does one job on a schedule and writes the result to a shared Drive folder. Coach and my main assistant read the folder. I read what reaches me.

Desk	Job	When
Email	Sort the inbox, list replies I owe, draft only	3 times a day
Calendar & chase	Check tomorrow's meetings, flag any without a purpose line	every morning
Radar	AI and competitor news for my Monday brief	weekday mornings
Workshop prep	A prep doc 1 to 7 days before each session	every morning
Content	Three drafts in my voice, never posted	twice a week
Event photos	Raw photos in, best 20 picked	Sunday night

The four rules I set before the first bot ran

1. Bots never send. Drafts only. If a bot prepares an email, it waits for me to press Send. Grok Bot supports this: when an action needs approval, you get Allow once, Always allow or Deny on screen. I never press Always allow on anything that leaves the building.

2. Bots do not touch my calendar. They can read it. Writing stays with me.

3. A bot's report is a lead, not a fact. If a bot says an invoice is unpaid, I check the bank before I chase anyone. The docs say the same: for consequential decisions, ask the bot to check the current source instead of relying on memory.

4. Anything that needs my laptop stays on my laptop. Local files, my private notes, anything that sends. In Settings, local execution has three choices: Ask every time, Always allow, Never allow. The security page itself recommends Never allow unless a bot has a specific reason to work on local files.

A pattern you can copy: the four-bot meeting

This is the part I show on stage.

One chatbot gives you one answer. Four bots in a group give you a meeting.

When I built a sales playbook live at a show on 15 September, I set up four bots in one group chat:

Sales director: owns the playbook. Conclusion first, then steps. Follows the brand's rules on what it may and may not claim.
Customer: speaks only as a customer. Asks the awkward questions. Says "too expensive" when it is.
New hire: first day on the job. After reading the playbook, repeats each step back in one sentence, and asks when something is unclear.
Reviewer: replies only "pass" or "fail, because...". Checks facts, risk, brand tone, and whether the steps contradict each other. Never rewrites. Ends with a score out of 100.

You watch. At the end you take two files away.

Copy the four profiles, swap in your product and your rules, and give the group one task.

What it cannot do: the playbook is a framework. Every real customer says something different, and judgement stays with a person. And the bots will keep talking. You have to be the one who says "enough".

Before you hand it company data: a Hong Kong checklist

Where it runs. US cloud computers. Check this against your data policy first.
Who it acts as. A bot acts as the signed-in person. It never has more access than you do, and everything it does is attributable to you.
What every bot can see. All your bots share one computer, files and logins. Don't put client files there that only one job needs.
What stops it. Keep sends and payments behind approval. Passwords and one-time codes go in yourself, never in chat.
What you clean up. Deleting a bot does not delete files or browser sessions on the computer. Sign out and remove files yourself.

If you are unsure what counts as sensitive, my traffic-light test takes 15 minutes: The Traffic Light Protocol.

Is it working?

I set this up on 15 October. For seven days the bots run beside the scripts I used before, job for job. After that I keep whichever does the job better, and I will add the result to this post.

FAQ

Is Grok Bot the same as Grok?

No. Grok answers questions. Grok Bot runs named bots that do multi-step work on a cloud computer and keep going when you close the app.

Do I need a separate subscription?

No. It is included with certain X Subscription tiers and paid Cursor plans, or through a linked SuperGrok or X Premium+ account. Check the current plan page.

Can a bot use any website?

Many, including ones without a connector. Some sites block automation or need a human step; the bot hands those to you.

What is the safest first job to give a bot?

Something that only reads and drafts: sorting your inbox into "replies I owe", or a morning summary of tomorrow's meetings. Nothing that sends.

Sam Wong (黃力桐) is a Hong Kong AI adoption trainer who has trained 11,000+ people.