Grok Bot Tutorial: Set Up Bots, Routines and Approvals (6 Best Videos)

Grok Bot is the SpaceX AI app that lets you set up a roster of AI bots, each with a job, a schedule and access to a shared cloud computer, and this guide distills six recent videos about it into one tutorial. The videos were published between September 26 and October 5, 2026, by Charlie Chang, AI Master, Riley Brown, Parker Prompts, How I AI and Matt Palmer. Between them they cover the first bot, routines, plugins, skills, approvals, group chats, the new primary and team bots, and about thirty real use cases. It was then checked against the 15 official Grok Bot guides from SpaceX AI, which settle several points the videos leave open. The video above, from AI Master, is the one to watch if you only watch one.

TLDW

Grok Bot gives you named bots that work in a cloud computer with its own browser, files and terminal, so they keep running when your laptop is shut. The reliable pattern across all six videos is one bot, one job, one schedule, with approvals on anything that sends a message or spends money. The creators disagree on price and on how many bots are worth running, and the product changed under them: two things one video lists as missing were shipped by the time another was recorded. The best use cases are boring rituals you already do on a schedule, and two of the six end with a sheet of paper on a printer.

The 6 Videos at a Glance

#	Video	Channel	Length	Best for
1	The ONLY Grok Bot Tutorial You Need 2026	Charlie Chang	13 min	A first look at the interface if you have never opened the app
2	Grok Bot Tutorial: How to Use Grok Bot for AI Automation (Full Workflow)	AI Master	19 min	The full build of five bots, with controls and the failures you will hit
3	Grok Bot Just Got 4 HUGE Updates	Riley Brown	21 min	Existing users who want the primary bot, team bots and inline drafts
4	9 Grok Bot Use Cases You Should Try ASAP	Parker Prompts	12 min	How to write a bot description that keeps it from guessing
5	7 Grok Bot agents I use every day	How I AI	36 min	Real bots run by a founder and parent, and moving over from OpenClaw
6	Organizing my life with Grok Bot	Matt Palmer	5 min	One idea: a printed daily brief, and reaching devices on your home network

Thoughts

Read together, the videos show that most people misunderstand what the bots share. Charlie Chang opens by saying each bot has its own computer and ends by saying they all share one. AI Master calls this out directly as “the correction I would make to half the videos about this product”: one cloud machine per account, and the bots queue on it. The official Grok Bot 101 guide agrees in one line: “if you log into a site with one bot, every other bot can reach that site too.” Parker Prompts adds the consequence, which is that anything you sign into on that machine can be reached by every bot on your account. That is convenient for a solo user and a real exposure once a bot is shared. It explains why the team bots in Riley Brown’s video start with no plugins at all and make each teammate connect their own Gmail. The isolation is the feature.

The limit on this product is not intelligence. It is attention. AI Master built five bots, kept three, and says the ceiling is the number of approvals one person is willing to read in a day. Within two days of running three routines the host was approving drafts without reading them, and cut approvals back to anything that leaves the account or spends money. Riley Brown’s inline drafts and the refund button in the How I AI episode are the product’s answer: make the approve step a single tap on a small card. That lowers the cost of each decision. It does not lower the number of decisions, and a faster button makes careless approval easier, not harder.

Every creator independently built the same bot before the product had it. AI Master has Foreman, Parker Prompts has Dispatch, Charlie Chang has a chief of staff, and the How I AI host had a rescue bot called the lifeguard on her old setup. Each exists to route work to the others. Riley Brown’s video, recorded last, shows this shipped as a built-in primary bot. The same thing happened with sharing: the How I AI episode names the lack of a bot you can use with other people as the one thing still missing, and team bots in Slack are the second update in Riley Brown’s list. If you are choosing what to build yourself, that pattern is worth noticing. Anything obvious enough that four people built it by hand is probably on the roadmap.

The praise is for the interface, and the complaints are about the model. Riley Brown would use Grok Bot for nearly everything if its models were at the frontier, and in the demo tells the bot to hand a coding task to Cursor running Opus 5.5. The How I AI host says the Grok model writes in a clipped style that is hard to follow, and had her bot study her sent mail before she let it draft anything. AI Master’s first inbox drafts all opened with “thanks for reaching out” until a style skill fixed it. Yet the same host shut down every OpenClaw agent she ran, because she was tired of maintaining them. Ease of setup is beating raw capability for this kind of work, at least for now.

The most appealing results in the set are printed on paper. The How I AI episode’s family bot produces a “kitchen table newspaper” with the school schedule, the weather and news for kids, and Matt Palmer’s whole video is a one-page daily brief waiting on the printer at 6:45 each morning. Both say the aim is less time on a screen. What the six videos mostly leave out is the downside. Nobody shows a bot buying the wrong thing, nobody reports what a month of hourly inbox sweeps costs in usage, and Riley Brown recommends routing bot traffic through your home connection so that retailers do not flag it as a bot, without asking whether those sites want it there. The company’s own guide is blunter than any of the creators: “If you log into Amazon on a computer the agent can use, it can technically buy whatever it wants, the same way a human could.” It also describes a permissions screen, with a review agent and allow and block lists, that not one of the six videos opens. Set that up before the first routine, not after.

The Tutorial: Grok Bot Step by Step

Step 1: Get access and install the app

Grok Bot comes with a paid plan, and the videos give different numbers, so check the pricing page before you rely on any of them. As of the September 26 video, AI Master says it is bundled with Cursor and SuperGrok plans, with Cursor Pro at $20 a month and base SuperGrok at $30 the cheapest ways in, down from $200 when the beta opened on August 11. As of the September 30 video, Charlie Chang reports paying $100 a month and that cheaper plans differ mainly in how much usage you get. Download the app from x.ai/bot. The desktop app runs on macOS, Windows and Linux, and the phone apps show the same bots.

Shown in: AI Master at 0:00, Charlie Chang at 0:46.

Step 2: Hire your first bot

Click the plus button and choose new bot. You get a chat, not a settings form. Describe the job in one plain paragraph: what to do, how often, and what to hand back. AI Master’s first bot, Scout, was told to check five sites every weekday morning for model releases and price changes and post five bullets with links. Give it a name and a picture. The How I AI host notes that a bot will guess its job from its name alone. Behind the chat, the official guide says a bot’s settings come down to three fields: Name, Title and Description. The Title is what Parker Prompts calls the label.

Parker Prompts draws a line that the other videos skip. The bot’s description holds its standing job and the rules it must always follow, such as “stay read only, never guess a number, and always show its sources.” A single task goes in the chat, with the outcome you want, the sources it may use, its limits, the format, and the points where it needs your approval. Pick a first job that is a ritual you already repeat with little judgment involved. AI Master’s rule for the whole product is one bot, one job, one schedule.

Shown in: AI Master at 0:46, Parker Prompts at 1:30, Charlie Chang at 1:32, How I AI at 3:04.

Step 3: Connect your accounts

There are four ways to give a bot access, and you will probably use more than one.

Plugins. Open the marketplace, install Gmail, Calendar, Drive, Slack, Notion or whatever you use, and sign in on the service’s own login screen. The How I AI host says the bot will usually prompt you to connect what it needs, and that you can connect several accounts per plugin, which matters if you have more than one inbox.
In the chat. Ask the bot to connect an account and it gives you a button to sign in.
The bot’s browser. For a site with no plugin, open the bot’s screen, take control, and log in yourself. The bot then uses the site the way a person would.
An MCP hub. Charlie Chang and Riley Brown both keep their connections in Composio so that one set of logins works across several agent apps. Neither video is sponsored by it.

One rule from AI Master applies everywhere: never type a password, an API key or a client’s private data into a bot chat. Use the login flow, the plugin, or take over the browser for the credential step. Charlie Chang also shows a password manager plugin that lets a bot log in without seeing the password. The official guide adds two details: Grok Bot supports the same MCP servers, plugins and skills as Cursor, and when a password or an API key is needed the bot can send you a secure form for it.

Shown in: Charlie Chang at 3:03, AI Master at 6:49, How I AI at 3:50, Charlie Chang at 9:54, AI Master at 15:17.

Step 4: Understand the shared computer

Each account gets one cloud computer with a browser, a file system and a terminal. Every bot has its own screen and its own private memory, but they share that machine’s files, browser sessions and logins. Two things follow. Work continues when your laptop is closed, which AI Master proves by shutting one and coming back to a digest stamped 8:02. And any bot can reach anything another bot is signed into, so treat the shared layer as public within your own roster and use the description to say what each bot should and should not touch.

Click the computer icon to watch a bot work or to take the mouse and keyboard yourself. Charlie Chang notes that the machine runs Linux, needs an update now and then from settings, and can be reset if a bot gets stuck.

Shown in: AI Master at 4:32, Parker Prompts at 0:45, Charlie Chang at 12:10, Charlie Chang at 4:36.

Step 5: Put the bot on a routine

A routine is what turns a chat into something that works without you. Either ask in plain words (“Every morning at 9:00 a.m., look through my Slack and email and send me a report”) or open the bot’s page, add a routine, and set what to run, how often and at what time. Routines appear in a list on the bot’s page with the next run time. A routine can also fire on an event: AI Master gives the inbox bot a two-hour sweep plus a webhook that other software can call when something needs attention now. In the official guide’s words, a bot “can set its own schedule or listen for events from other apps,” such as a Slack thread or a GitHub pull request.

Two cautions. The How I AI host found she had to be explicit, and more than once, about what should run on a schedule, because the bot was less proactive out of the box than her old setup. And routines use up your allowance: a bot that sweeps an inbox every two hours burns through it much faster than one that runs once at eight.

Shown in: AI Master at 1:31, Charlie Chang at 7:37, AI Master at 10:38, How I AI at 11:29.

Step 6: Teach it your process and your voice

Two features cover the things that are hard to describe. Teach a task records you doing a process once in the bot’s browser and writes the steps back as a numbered list you can correct. AI Master used it for filing a new client email and had to fix one step, where the bot had recorded an exact spreadsheet cell when the intent was the next empty row. Skills are named sets of instructions a bot loads for a kind of work. They are separate from bots, so one style skill can be attached to several.

Any bot that writes for you needs a style skill before its drafts are usable. AI Master’s says short sentences, no corporate throat clearing, sign off the way I sign off, and never promise a call I have not agreed to. The How I AI host took a different route to the same place and told her bot to go through her sent email and pull out her voice. Expect to revise the skill once and judge the result on the second pass.

Shown in: AI Master at 9:07, AI Master at 7:36, How I AI at 9:58, Charlie Chang at 5:22.

Step 7: Set approvals and limits before you trust it

AI Master says to set these up on day one. Mark an action as needing a human and the bot parks it in an approvals queue with its reasoning and an approve or reject button, on desktop and phone. Cap what each bot can spend and scope what it may touch. An activity log records every step, so when something goes wrong you can read what happened. Human takeover lets you step into the bot’s computer, do the one step it is stuck on, and hand it back.

Keep the approvals list short. Approve anything that leaves your account or spends money, and let the rest go to the log. For email and Slack, Riley Brown shows the newer inline drafts: the bot replies with an editable card and a send button, and nothing goes out until you press it. The How I AI host uses the same kind of approval for refunds, where a button in the chat leads to a confirmation in Stripe.

The official Grok Bot 101 guide describes a layer under all of this that none of the videos shows. You write permission rules in plain language under Settings, then General, then Agent. A separate review agent checks each proposed action and can allow it, block it, or escalate it to you, and allow and block lists steer that reviewer. The role guides add habits worth copying. Josh Kim’s inbox bot is told to “never send or reply or react without me typing ‘send’ in that moment” and to “fail closed and stay silent on noop.” Kim’s marketing guide starts permissions small: “Trust comes from delivered work, not from handing over the company credit card on day one.”

Shown in: AI Master at 13:44, AI Master at 14:30, Riley Brown at 14:38, How I AI at 25:14.

Step 8: Add a second bot, then a coordinator

Build the second bot only when the first one works alone. Copying a description that already works and editing two lines is the fastest way. Bots hand work to each other with an @mention, and each decides whether the work is its own by checking the thread against its job description. Vague descriptions are what make handoffs stall. In a group chat, AI Master keeps to three bots and names one dispatcher, because without an owner the bots spend messages working out who is on it.

The newest version makes the dispatcher a built-in role. As of the October 5 video, Riley Brown shows a primary bot, marked with a star, that takes whatever you say and routes it to the right bot. Any bot can become the primary if you ask it to. Brown pins it to the top and talks to it by voice from a phone while walking. If you would sooner see the traffic, Charlie Chang uses group threads for visibility, and Parker Prompts’s Dispatch bot asks seven bots for status and returns one dashboard.

The official guides agree with AI Master on timing. Josh Kim runs a bot called Bot Boss as “the single front door” to three specialists, and it is an ordinary bot with instructions, not the built-in primary bot. But Kim’s marketing guide says: “Add Project Manager after the handoffs hurt, not before.” Eric Zakariasson’s guide to running several teams, which the author calls an experimental pattern, gives each project a channel, a roster and a Notion board, with a manager bot and at most five others in a channel.

Shown in: AI Master at 12:12, AI Master at 12:58, Riley Brown at 0:46, Riley Brown at 5:23, Charlie Chang at 10:40, Parker Prompts at 9:09.

Step 9: Share a bot with your team

Personal bots share your plugins and skills, which is exactly why you should not put one in a shared Slack channel: it can read your mail and post as you. Team bots, shown in Riley Brown’s October 5 video, are separate. A team bot starts with no plugins, has its own skills, files and secrets, and stays a draft until you publish it. Then you bring it to your team’s Slack from the bot’s details panel and add it to channels, where people @mention it.

Plugins that need a login are connected by each teammate, so the bot uses your Gmail when you ask and a colleague’s when they ask. Routines created through a team bot stay personal to whoever made them. Brown’s order of work: create, configure and test in the app, publish, add to Slack, and test again there before telling the team. Before this existed, the How I AI host listed the lack of it as the product’s main gap, with shareable bot templates as the workaround.

To hand someone a starting point instead of a live shared bot, use a template. The official templates guide calls a template “a recipe, not a clone.” In the bot’s settings, click Share as Template at the bottom right, check what is included, click Publish, then Copy Link. The person receiving it clicks Add to Grok Bot, then Add Bot, and connects their own accounts. First-party plugins and skills travel with the template. Secrets and private information do not. The guide’s warning applies to marketplace bots as well: “Be very careful about what you’re installing in your bot.”

Shown in: Riley Brown at 7:42, Riley Brown at 10:46, Riley Brown at 13:05, How I AI at 6:54.

Step 10: Fix the three things that stop a bot

Logins and two-factor prompts. AI Master’s first failure was a morning routine that hit a two-factor prompt and waited silently for 40 minutes. Put login-gated sites behind an approval so the bot pings you first, then enter the code yourself with human takeover.
Sites that block cloud traffic. A bot working from a cloud machine can be flagged as automated, which stops checkouts and forms. Riley Brown turns on “route traffic through this computer” under Settings, then Computer, then Network, so requests come from a home connection. That is Brown’s recommendation. Whether a given site allows automated use is a separate question.
Things on your home network. Matt Palmer uses the same routing so a bot can reach a home printer. It only works while a machine at home is awake, so the routine is as reliable as that machine. Palmer plans to use an always-on Mac Mini, and suggests a Raspberry Pi would also do.

Shown in: AI Master at 15:17, Riley Brown at 17:39, Matt Palmer at 2:21.

Cheat Sheet

Term	What it is	When to use it
Bot	A named agent with a description, private memory and its own screen	One per job
Description	The bot’s standing job and permanent rules	Rules that apply to every run
Routine	A run on a schedule, or on an event such as a webhook	Anything you would otherwise have to remember to ask for
Plugin	A connector plus skills for a service, signed in through the service	Gmail, Calendar, Drive, Slack, Notion and similar
Skill	A named instruction set that can be attached to several bots	Writing style, a playbook, a repeated procedure
Teach a task	Records you doing a process once and saves the steps	Rituals that are easier to show than to describe
Screen and takeover	A live view of the bot’s computer, with mouse and keyboard control	Logins, CAPTCHAs, and the one step it cannot finish
Approvals	A queue of parked actions waiting for approve or reject	Sending messages and spending money
Inline draft	An editable email or Slack card with a send button in the chat	Reviewing outgoing messages quickly
Group chat	A thread with several bots, handing off by @mention	Up to three bots that each already work alone
Primary bot	The starred bot that routes your requests to the others	When you have more than two or three bots
Team bot	An isolated bot published to a team and added to Slack	Anything more than one person will use
Route traffic	Sends the bot’s traffic through your own computer	Blocked sites and devices on your local network
Permissions	Plain-language rules under Settings, General, Agent, enforced by a review agent with allow and block lists	Before the first routine that can send or spend
Template	A shareable recipe for a bot, without secrets or private information	Giving someone your setup, or installing someone else’s

What the Official Guides Add

SpaceX AI publishes 15 written guides at x.ai/bot/guides, most of them by its own staff describing the bots they run. I read all of them on October 6, 2026. They confirm the tutorial above on the shared computer, multiple accounts per service, routines and human takeover, and they add the permissions screen and templates covered in steps 7 and 9. None of them states a price. Matt Palmer, who made the sixth video below, wrote the first two.

Three habits repeat across the role guides. Bots draft and a person sends: the legal, sales and work guides all hold outgoing messages for approval. A bot with nothing to report says nothing, so routines often produce no output on purpose. And permissions start small and grow with delivered work. The engineering guide adds a caution about cost, saying its urgent mode, where a bot checks a coding agent every five minutes, “can burn tokens much faster than you think.”

Guide	By	Covers
Grok Bot 101	Matt Palmer	The basics: the computer, connectors, routines, permissions and takeover
Templates for Grok Bot	Matt Palmer	Publishing a bot as a template and installing someone else’s
Grok Bot for Work	Josh Kim	Four bots for inbox, calendar and briefings behind one front door
Grok Bot for PMs	Kevin Niparko	An hourly “attention list” and a roster run like a small org
Grok Bot for Engineering	Lingxi Li	Bots that manage Cursor cloud agents, with a warning about token use in urgent mode
How I run multiple teams of Grok Bots	Eric Zakariasson	A channel, a roster and a Notion board per project
Grok Bot for mobile app development	Ryan Perry	Six bots running a mobile game from analytics to deployment
Designing Grok Bot with Grok Bot	John Bai	Design bots for Figma production, motion and experiments
Grok Bot for Support	David Gan	Bug reproduction, churn analysis, refunds with approval, and reporting
Grok Bot for Marketing	Josh Kim	Six bots from competitor research to ad shells, none allowed to change live spend
Grok Bot for GTM	Krista Letz	Meeting prep, prospecting, forecasting and slides for enterprise sales
Grok Bot for SDRs	Stefan Markarian	Outbound research where bots stage the work and a person sends it
Grok Bot for Post-Sales	Blake Schuller	A chief of staff bot, one specialist per account, and eight routines
Grok Bot for Recruiting	Holly Dyche	Interview prep, scorecards, sourcing and loop analysis
Grok Bot for Legal Ops	Kelly Noguchi	Slack triage and contract workflows on a draft-then-human-send rule

The Videos, One by One

1. The ONLY Grok Bot Tutorial You Need 2026 – How To Use Grok Bot For Beginners

Charlie Chang, 13 minutes. Watch on YouTube.

A click-through tour for someone who has not opened the app. Chang creates an executive assistant bot, connects Gmail, watches the bot shop for a soundbar on Amazon, and shows routines, the marketplace and group threads. It is broad and quick, and it stays on the surface: there is nothing on approvals, limits or what goes wrong. Skip it if you have already made a bot.

Start with an executive assistant, and add a chief of staff if you run a business. Every other bot should own one specific thing.
Ask the assistant what routines it thinks you should give it, after telling it about your work and your day.
The marketplace has importable bots, including one that interviews you and then builds bots for you.
Chang’s business roster: finance (invoices, QuickBooks reports), newsletter drafting, a recruiting assistant, a content strategist, several researchers and LinkedIn outreach.
All bots share one cloud computer, so a login made by one is reachable by all. You can tell a bot not to use an account, but that is an instruction, not a wall.

0:46 Where to download it and what the plans cost.
1:32 Creating and naming the executive assistant.
3:03 Connecting Gmail in the chat, and using Composio or Zapier for many logins.
4:36 Opening the bot’s screen and teaching it a task.
7:37 Routines, and the bots Chang runs for several businesses.
10:40 Group threads, replying to email from the chat, and the shared computer.

“It definitely moves slower than a real human, but it will get things done.”
Charlie Chang, on watching a bot shop and talk to customer service

2. Grok Bot Tutorial: How to Use Grok Bot for AI Automation (Full Workflow)

AI Master, 19 minutes. Watch on YouTube. This is the video embedded at the top of the post.

The most complete of the six. It builds five bots in order (Scout for research, Clerk for the inbox, Ledger for a weekly report, Pen for drafts, Foreman to dispatch), explains every control on the way, and then lists three failures and who should not use the product at all. Two sponsor segments interrupt it, one for the channel’s own platform. Everything between them is dense and worth the time.

One bot, one job, one schedule. People who built a single bot to do eleven things got eleven mediocre results.
A bot with a browser can use services that have no useful API. If a person can do it in Chrome, the bot can attempt it.
Skills are separate objects from bots, so the style skill written for the inbox bot was attached to the drafting bot unchanged.
A writing bot gets no routine. On a timer it produces drafts nobody asked for.
Who should not switch: people whose work lives in a code repository, people who want deterministic workflows, and anyone with a one-off task.

0:46 Hiring Scout from the empty roster and setting its routine.
4:32 One cloud machine per account, the browser, storage and the closed-laptop test.
6:49 Building Clerk: the mail plugin, skills, and teach a task.
11:25 Ledger, Pen and Foreman, and the group chat handoff.
13:44 Approvals, human takeover, limits and the activity log.
15:17 Three failures, who should stay on other tools, and the verdict.

“The ceiling here is not the model. It is how many approvals I am willing to read in a day.”
AI Master, after building five bots and keeping three

3. Grok Bot Just Got 4 HUGE Updates (Changes Everything)

Riley Brown, 21 minutes. Watch on YouTube.

A walkthrough of four changes as of October 5: the primary bot, team bots, inline drafts for email and Slack, and routing traffic through your own computer. It assumes you already use the product. The team bot section is the clearest explanation available of why shared bots have to be kept apart from personal ones.

Talking to one primary bot fixed the overwhelm of managing a roster. It delegated a design change to the developer bot without being asked to.
The developer bot hands code work to Cursor, and a coding tab on the bot opens that session. Brown asked for Opus 5.5 and a push straight to production.
Personal bots are close to identical under the surface: same plugins and skills, different instructions and thread history. Team bots are isolated by design.
Brown expects editable cards in the chat to spread from email and Slack to most tasks through 2027.
Brown’s view of the product: fast, the most polished interface of any agent platform, and held back by models that are not at the frontier.

0:46 The primary bot and a live delegation to a developer bot.
5:23 A voice call with the primary bot from the phone app.
7:42 Personal bots joining a team versus creating a team bot.
10:46 Publishing a team bot and adding it to Slack.
14:38 Inline drafts and the send gate.
17:39 Routing traffic through your computer, and closing thoughts.

“You don’t want a bot that your whole team is going to use to use your email. That doesn’t make any sense.”
Riley Brown, on why team bots start with no plugins

4. 9 Grok Bot Use Cases You Should Try ASAP

Parker Prompts, 12 minutes. Watch on YouTube.

Nine small bots built one after another, each with a name, a label and a tight description. The use cases are modest. The value is in how the instructions are written: every bot is told what it may not do and what to report when it cannot verify something. Several are tested on sample data before they touch a real account.

The nine: Hunter (YouTube idea research), Audience Pulse (comment themes), Air Watch (price tracking), Forge (single-file browser games), Cleaner (folder tidying), Trial Guard (free trial renewals), Weekwise (weekly planning), Dispatch (status across bots) and Architect (builds new bots from a description).
Write “unknown” into the rules. When Amazon’s price failed to load, Air Watch recorded it as unknown and sent no alert.
For anything that changes files, have the bot show a plan first and wait for approval. Cleaner flags unclear file names and never treats an old file as useless because of its date.
Time limits are not reliable. Hunter was given 20 minutes and took 52.
A bot that builds bots saves the repeated setup once you have more than a handful.

0:45 The bot’s screen, the shared computer, and description versus chat.
2:17 Hunter’s research table and the missed time limit.
4:34 Air Watch and the “unknown” result.
6:04 Cleaner, Trial Guard and Weekwise, tested on sample data.
9:09 Dispatch’s dashboard, then Architect creating two new bots.

“Since it cannot verify a genuine price drop, it stays quiet, which is exactly how I want it to behave.”
Parker Prompts, on the price-watch bot when one retailer failed to load

5. 7 Grok Bot agents I use every day

How I AI, 36 minutes. Watch on YouTube.

The longest video and the one with the most lived-in examples. The host, a founder with three children, has moved almost everything off her OpenClaw agents and shows the bots she runs for work, engineering, customer support, family and shopping. It also has the most direct criticism of the product, and ends with how she migrated and shut down her old agents. Two sponsor reads are easy to skip.

Her chief of staff bot sweeps six inboxes, calendars and Slack hourly from 6 a.m. to 9 p.m., archives what she does not need, and pings her only for what does.
The family bot prints a daily “kitchen table newspaper,” asks at 2:30 who is doing school pickup and what gear is needed, and sends a Friday preview of the weekend.
For engineering, one bot chases open pull requests and can start Cursor cloud agents for rebases, and another logs into a compliance dashboard daily and opens fixes for new vulnerabilities.
Her support bot sweeps Intercom and email hourly, fixes bugs, requests refund approval, and reviews the week for gaps in the docs. She says customers began offering to leave reviews about the support.
Personal bots find subscriptions to cancel, flag renewals worth negotiating, shop to exact dimensions, and send a Monday list of sale items that match a Pinterest board.

2:19 What Grok Bot is: bots, plugins, the computer and routines.
6:54 The gap she names: bots are single player.
7:39 The chief of staff bot, voice training, and making it proactive.
12:14 The family bot and the printed newspaper.
19:52 Engineering and compliance bots, then customer support with refund approvals.
26:44 Subscription, shopping and wardrobe bots.
32:51 Migrating from OpenClaw and her verdict that UX matters.

“The ideal thing for an agent for a home is to get you off your computer and with your family.”
The How I AI host, on what a household bot is for

6. Organizing my life with Grok Bot

Matt Palmer, 5 minutes. Watch on YouTube.

A short single idea from the author of the official Grok Bot 101 guide, who signs off as “Matt with SpaceX AI”: a one-page daily brief, printed at home every morning, with the day’s schedule, health tasks, things to think about from X bookmarks, and space for journaling. It is thin on steps and does not show the prompts. Watch it for the design idea and for the explanation of how a cloud bot reaches a printer on a home network.

Palmer prototyped the page layout with an engineering bot first, aiming for a minimal design, and keeps the design in a small repository.
The brief bot draws on existing bots for workouts, tasks, calendar and bookmarks. The hard part, gathering the data, was already done.
Grok Bot can run commands on your own computer and route traffic through it, which is how it reaches a local printer.
The routine fires at 6:45 a.m. but only when a laptop is awake. There is no always-on machine yet.

0:45 Why paper first thing in the morning.
1:31 Designing the brief with an engineering bot.
2:21 Reaching a printer on the local network.
4:02 The 6:45 automation and its current limit.

“As much as possible, I want to eliminate sort of digital interactions first thing in the morning.”
Matt Palmer, on why the daily brief is printed

If you watch one, make it AI Master’s full workflow for the build and the failure cases, then the How I AI episode for what a working roster looks like after several weeks. Start with one bot that does one dull job at eight in the morning, and add the second only when the first is boring.