21 plays for Grok Bot: what to actually build in your first month

The people getting the most out of Grok Bot are not the ones with the cleverest prompts. They are the ones who arrived with a list. A Bot is a role, and a role needs a job to walk into on day one. Anyone who opens the app without that list ends up chatting with an agent instead of delegating to one, which is a slower version of what a normal chat window already does. So this is the list. Twenty-one plays grouped by how much setup they need, followed by the craft tricks that separate a Bot producing usable output from a Bot producing a paragraph nobody reads. The plays get more valuable as they go down the page, and the first six can all be running before lunch.

The shift that makes everything else work A standard AI assistant answers inside a window and stops when the window closes. Grok Bot runs on its own cloud machine with a real browser, a file system, and a terminal, and it keeps going after the laptop shuts. That single difference reframes the whole question. The right task is no longer "what can I ask" but "what would I hand to a capable new hire who works while I sleep and reports back in the morning." Roman, a product team member at xAI, put the value in one line: "There is a huge difference between 90% done and 100% done." The claim is that the work lands where a human would put it, inside the actual tool, rather than coming back as a suggestion someone still has to execute. Every play below is written to end in a finished artifact for that reason.

Tier one: six plays that need one demonstration each

1. Inbox triage with a labeled output. The Bot reads the mail, sorts by category, and writes a single morning file with what needs a reply, what needs a decision, and what can wait. The output living in one file is what makes this checkable in ninety seconds.
2. Support reply drafts. Lenny Rachitsky, who got early access, put auto-replying to support emails on his list of four favorite uses and said it saves him hours. The safe version drafts and queues rather than sends, which keeps a human on the last click while removing the typing.
3. Subscription audit. Another of Rachitsky's four: scanning credit card statements and finding recurring subscriptions to cancel. This is the play with the fastest visible payback, because it returns money in the first run and needs no ongoing schedule.
4. Guest and meeting briefs. Rachitsky built a Bot that sends him briefs for upcoming podcast guests, and the pattern moves directly onto sales calls, investor meetings, and interviews. One page, arriving the night before, on whoever is on the calendar tomorrow.
5. Receipts into a sheet. xAI's own operations Bots process invoices arriving in Gmail. The version to build reads the mail, extracts vendor, amount, and date, writes one row per document, and flags anything from a vendor it has not seen before.
6. Weekly report assembly. Three dashboards, one recurring format, one file. Nobody enjoys this task and it has an unambiguous right answer, which makes it the ideal first hire for anyone who wants to trust the output later.

Tier two: six plays that run on a schedule 7. The overnight brief. A fixed list of companies, topics, or tickers, scanned against named sources, delivered before the day starts. Discipline beats ambition here. Forty items with sources reads better than four hundred without them. 8. Demo readiness. One of the sharpest patterns from inside xAI: a demo-readiness Bot checks the environment overnight, fixes broken seeds and stale data in the product interface, and drops a checklist before morning calls. Any team that has ever opened a demo on a broken account understands the value immediately. 9. Nightly click-path check. The Bot walks the same path through a product or dashboard every night, screenshots each step, and files a ticket when something changed. This is the job computer use was made for, because it needs no integration at all. 10. Change watcher. Competitor pricing pages, documentation, terms, job boards. The Bot loads them nightly and reports only the differences, which turns a task nobody does into a task that runs on its own. 11. Content repurposing queue. One long source, several formats, drafted overnight and waiting in a folder by morning. The Bot handles extraction and first drafts. The human handles voice and the decision to publish. 12. Matchmaking against criteria. Rachitsky's first listed use was matchmaking people looking for jobs with companies who are hiring. The same shape works for vendors against requirements, properties against a brief, or grants against a profile. Give the Bot the criteria and a source of listings, and let it score.

Tier three: five plays that use more than one Bot 13. The chief of staff. Matt Shumer, who tested the product for weeks before launch, created separate researcher and writer Bots, then created a Chief of Staff Bot and told it to coordinate the other two on a project. He described the product as an agent for everything, not just code. Coordination through a manager Bot beats coordination through a human when the work is routine and the format is fixed. 14. The relay. Researcher produces the source pack, writer produces the draft, editor produces the final. Each Bot has one job, one input, and one output, which makes a broken link in the chain obvious rather than mysterious. 15. The on-call responder. An insider session on the product demonstrated an on-call agent that monitors Datadog and PagerDuty, reacts to incidents, and creates its own monitoring routines, which is the jump from scheduled work to event-driven work. 16. Repro and fix. xAI runs a Bot that reproduces bugs in the product interface, files the ticket, and hands the fix to a debugging Bot. Two Bots, one pipeline, no human in the middle until review. 17. Records that maintain themselves. Internal Bots at xAI keep CRM records and org charts clean, flag stalls and commit risk, and update CRM notes from call transcripts. The value is not the writing. It is that the record exists on the days when nobody feels like writing it.

Tier four: four plays worth building once you trust the roster 18. One Bot per account. Claire Vo, who set up five Bots for a solo episode on the product, named multi-account connectors as the killer feature and the reason she actually used it. Anyone juggling several clients, several inboxes, or a personal and a business identity gets a clean separation here that most agent tools do not offer. 19. Screenshot evidence as a habit. Ask every Bot to save a screenshot at each decision point and drop them beside the output. Review time collapses, because checking a claim becomes looking at a picture rather than rerunning the task. 20. The self-report. One Bot whose only job is to summarize what the other Bots did this week, what they finished, what they escalated, and what took longest. A roster that reports on itself is a roster that can be tuned. 21. Routines as a library. Demonstrated tasks get saved as routines, which means every workflow taught once becomes an asset that can be scheduled, copied, and adapted. Teams that treat these as a growing library rather than one-off setups compound faster than teams that rebuild each time.

The craft tricks Demonstrate on messy data, not a clean example. A routine learned on a tidy sample will meet a real one at 3am. Walking the Bot through the ugly version once builds a routine that survives. Write a brief, not a prompt. Where the inputs live, which tools to open, what the finished output looks like, where it gets saved, and what to do when something is missing. That last line does most of the work, because an agent without an escalation rule will fill the gap itself. Set the approval line in advance. Any amount over a threshold, any new vendor, any message leaving the company, any change to a live system. Below the line the Bot finishes and logs. Above it, the Bot waits. A well-tuned roster produces a short morning queue and a long list of completed items. Route by speed. Grok Bot ships a catalogue of around 220 plugins with none installed by default, covering Google Workspace, Slack, Microsoft 365, Salesforce, Notion, Glean, GitHub and Jira, and where no plugin exists the Bot works through a browser instead. Connector work moves fast enough to schedule hourly. Browser work moves at human speed and belongs overnight, where the wall clock cost is invisible. Expect to type the password yourself. When a login or a two-factor prompt stops the Bot, the user takes over and enters the credentials. This is a feature worth planning around: do the first sign-in pass for every tool in one sitting, then let the sessions carry. Name Bots by function. A roster of five named workers with five written briefs can be audited in a morning. A roster spun up conversationally cannot be reconstructed three weeks later. Keep one account per trust level. Personal admin on one, client and production work on another. It costs an extra seat and it makes every later question about access easy to answer. Close the laptop on purpose. The most common mistake in week one is sitting and watching. The product is built for work that continues after the app is shut, and the habit of assigning at night and reviewing in the morning is where the time actually comes back.

What it costs and where to get it Grok Bot has been in beta since August 11, 2026, and there is no standalone plan, so access arrives bundled with something bought for another reason. The Bot page lists three paid doors: Cursor Ultra at $200 a month, SuperGrok Heavy at $300 a month, and Cursor Premium Teams at $120 per seat a month. Premium Teams adds centralized billing, a team skills and plugins marketplace, shared usage analytics and single sign-on, which makes the team seat cheaper per person than the solo plan. On platforms, the common misreading is worth correcting: the desktop app covers macOS and Windows, the phone app needs iOS 18 or later, Linux desktop is not supported at launch, and the Linux in the coverage is the managed virtual machine the Bot itself works on. Two details shape the real invoice more than the sticker price does. Each subscription includes a weekly Grok Bot usage allowance, extra usage is billed from model and token cost, no Grok Bot spend cap has been published yet, and there is no model picker to steer work toward something cheaper. Reports from late August describe an expanded eligibility list and a limited free trial announced through an X post rather than the documentation release notes, while a check of the official pages on August 20 still placed Cursor Pro, Pro Plus and SuperGrok Plus outside the eligible list. Check the live page before buying rather than trusting any roundup, including this one.

A first weekend that sets the whole month Saturday morning: one account, one Bot, one task from tier one, demonstrated on real data and run once with a human watching. Saturday evening: the same task put on a schedule and left overnight. Sunday: check what came back, fix the brief, add the second and third Bots on unrelated jobs so their outputs stay in separate files. By the end of week one there are three roles running, three files to check, and roughly an hour a day that nobody has to spend. By the end of the month, the interesting number is not how much the roster produced. It is how many tasks stopped appearing on anybody's list at all.