Six Bots, Seven Days, One Person Runs the Company
The free 1-hour workshop is just the entry point. What actually gets a one-person company running is six Bots, a playbook, and three rules.
X user Angelic reposted a message: xAI has released a free 1-hour workshop in which three engineers start from scratch and build a complete product live. The three are Matt Palmer, Lauren Tan, and Roshan Sadanani. The video is below. If you want to jump straight to the key sections, use these timestamps: 04:12 build the first bot; 28:40 engineers hand real work to an agent; 44:25 run a product without a traditional PM; 52:25 build the founder tech stack.
Angelic put it plainly: people would pay 700 dollars for this kind of hands-on agentic engineering workshop. Right now it is free.
But what was actually being repeatedly shared in the comments was something else. broke boy's article 'How To Run a One-Person Company on Grok Bot: The 7-Day Setup' — roughly, in Chinese: how to run a one-person company with Grok Bot, a 7-day setup. If the workshop's highlight is three engineers building a product from scratch, this article is about how one person gets a company running.
The article opens with a judgment: most people do not lack time; their energy is drained by switching. Inbox, CRM, customer support, content, invoices, research: six jobs, one person, forty tabs. Grok Bot is designed to eliminate this part.
First, some background. Grok Bot is a named, long-lived AI teammate. You assign work to it the way you would message a colleague. Each Bot has a role and runs on a cloud computer with a browser, terminal, and file system; it keeps running after you close your laptop. It can run routine tasks on a schedule, use written skills, and connect to Gmail, Calendar, Drive, Notion, and GitHub. Two to six Bots can be pulled into a group to collaborate. It is included in X Premium's 30-dollar tier, and usage comes from a weekly pool shared with Grok and Cursor. The official exact numbers are not published, and since it is beta, details will change; check the docs before you start.
Three rules, set them first.
Rule one: separate accounts, least privilege. Multiple Bots share one computer, and xAI's own docs say not to treat a Bot as a security boundary. Create a dedicated email for Bots; use restricted seats for CRM and customer support systems, default to read-only, and enable write access only when the work requires it. Bank logins, card numbers, and payment management never go in.
Rule two: irreversible actions must be confirmed by you personally. Sending, publishing, refunding, deleting, signing, and paying: Bots can only prepare; the button stays in your hands. Reading and drafting can be opened up. Any operation that sends something out, spends money, or deletes data should stop and wait for you. The Require Approval setting overrides Always Allow.
Rule three: business knowledge lives in files you own. Grok Bot's memory currently cannot be inspected item by item or exported, so put company knowledge in Drive, Notion, or GitHub, and then connect the Bots to it.
Then comes the playbook. broke boy says this is the highest-ROI 30 minutes. One folder, six files:
company.md: what you sell, to whom, and for how much
voice.md: how to write, with 5 real examples
never.md: which phrases and topics to avoid
faq.md: every customer question + standard answer
policies.md: refunds, response times, escalation
customers.md: what the ideal customer looks like
Two habits. Instructions must be specific. No exclamation marks; answer in the first sentence works; Be more professional does not. If you correct a Bot twice on the same issue, write the correction into the playbook. Corrections that stay only in the conversation will repeat forever.
Below is the six-person team. The original article is very detailed; here I relay the setup directly as written.
Brief, chief of staff. At 07:00 on workdays, read the other Bots' output and the calendar, and send a briefing under 200 words: first what needs a decision, one-line meeting prep, key numbers, and anomalies. Report only, do not act.
Scout, research plus sales leads. On Monday, check competitor pricing and changelogs, scan public discussion, and write a weekly report; every conclusion includes a source, and anything unverified is marked separately. Every day, write a two-line summary and fit score for new leads.
Quill, content. On Friday, turn one long article into a newsletter, a thread, three standalone posts, and one short-video script, all saved as drafts. Write in your voice, but it does not have your experiences, so the opinions still have to come from you.
Helm, customer support. Draft replies according to faq.md and policies.md. For refunds, legal issues, angry customers, or questions not covered by the FAQ, write an escalation note; do not invent policy.
Penny, finance. On the 1st of each month, read exported data (read-only), categorize expenses, reconcile payments and invoices, and list items due for renewal in the next 30 days. All totals must be calculated with code and checked against source files; no estimates.
Kit, review. The role most people skip. Before Quill, Helm, and Scout drafts reach you, Kit checks them against never.md, voice.md, and policies.md and attaches a one-line conclusion: clean, or which rule was violated. You no longer read raw drafts, only the versions that have already been checked.
Set it up over 7 days:
Day 1, lock down permissions. Subscribe to Premium, create the Bot email and restricted seats, and set up approvals according to rule two.
Day 2, write the playbook. Six files; rough is fine, and real examples are more useful than adjectives.
Day 3, hire Brief. Run the morning briefing manually once, then schedule it.
Day 4, hire Helm. Customer support is the safest first step: repeated questions, ready answers, drafting only, and record the time saved.
Day 5, hire Kit. Have it check Helm's drafts, and tune it until it catches the mistakes you would catch.
Day 6, hire Quill and Scout. One routine task each; do not add more yet.
Day 7, hire Penny and do a retrospective. Manually reconcile the month's accounts completely against Penny's results. Then open the Usage page and see which routine tasks burn the most usage.
Add group chat last. Multi-Bot collaboration burns usage quickly, so reserve it for truly cross-role work, such as a competitor changing prices, triggering a Scout report, a Quill draft, and a Helm FAQ update in one run.
Once it is running, a week looks like this:
Workdays 07:00, Brief sends the morning queue.
Workdays 08:00, Scout processes new leads.
Workdays 09:00, Helm handles overnight tickets, and Kit reviews.
Monday 06:00, Scout competitor report.
Friday 10:00, Quill content drafts.
Friday 16:00, Brief weekly review.
The 1st of each month, Penny finance.
Your work is down to three things: approve the queue, do the things only you can do, and read the Friday summary.
The article does not portray Grok Bot as a universal tool. It lists several real limitations:
In beta, processes need to tolerate surprises.
A shared computer is not a security boundary, so permission design must account for that.
Usage comes from a weekly pool, and bloated routine tasks can exhaust it midway.
Browser automation gets blocked by CAPTCHAs and logins; prefer official APIs or plugins.
Skills generated from screen recordings are only first drafts; test them before use.
These limitations matter because they determine what this system is and is not good for.
The last paragraph is the most valuable point in the whole article. Bots make labor cheap: research, drafting, categorizing, summarizing, preparing. But they do not make judgment cheap. What to do, for whom, how to say it, and when to say no remain your job, and they will become most of your job.
With structure, one person operates like a small company. Without structure, one person just becomes a very busy approval officer.
Write the playbook tonight. Hire Brief tomorrow.
The workshop tweet and the original article: How To Run a One-Person Company on Grok Bot: The 7-Day Setup
发布时间: 2026-11-20 09:11