Grok Bot: I Tested It on Real Work

xAI released Grok Bot on 11 August 2026. The pitch is a team of AI agents that keep working while you are asleep, sign into your existing tools, and see a job through to the end.

I built one the day it launched and pointed it at a real task, not a demo. Here is what it did well, where it fell over, and whether any of it matters if you are not a developer.

What is Grok Bot?

Most AI tools work like a conversation. You open a window, ask something, get an answer, close it. Nothing continues after you leave.

Grok Bot works more like hiring someone. You create a Bot, give it a job and a name, connect it to the apps you already use, then message it the way you would message a colleague. It carries on when you are not there.

Two things make that possible: each Bot gets its own computer in the cloud, so work does not pause when you step away, and each can sign into your tools directly, including ones without a tidy programming interface, which is how they reach the messy apps most people actually live in.

If the difference between this and a chatbot is not obvious yet, our guide to AI agents versus chatbots covers the distinction in plain terms, and the glossary entry on AI agents is shorter still.

What it costs, and what that buys

There is no standalone price. Access is bundled into three existing subscriptions:

Plan	Price
SuperGrok Heavy	$300 per month
Cursor Ultra	$200 per month
Cursor Teams Premium	$120 per seat, per month
Enterprise	Waitlist

This is not a tool you try for ten dollars to see what the fuss is about. The cheapest way in is $120 a month, and it is a team plan.

It runs on macOS, Windows, Linux and iOS. Android is listed as coming.

SpaceX acquired Cursor, the company behind the coding tool of the same name, and Cursor has merged with xAI. The company now signs itself SpaceXAI, and inside the product almost every connection is labelled a Cursor plugin. If you were wondering why a Grok product keeps saying Cursor, that is why.

The bot I built: link-swap outreach

I did not give it a toy task. I run outreach for this site, contacting other publishers about swapping links, and it is exactly the sort of repetitive work people claim agents will absorb. I named the Bot Link Swaps and told it that was its job.

Setup was a conversation. It asked what to focus on, offered lettered options I could pick from, and read the scope back to me before starting. No workflow diagram, nothing to configure.

What it proposed, unprompted, was a four-step loop: pick a priority list, find a contact for each site, draft a short personalised email naming a specific post of theirs, and move each one through a tracker as the status changed. Then it asked how sending should work.

I chose the option labelled draft only, I send. It confirmed it would not send anything without my go-ahead, and it stuck to that.

I did not expect that. This site already runs a rule that outreach is drafted by machine and sent by a human, and it exists because a wrong email cannot be recalled. The Bot arrived at the same arrangement on its own and offered it as the default.

How does it actually connect to your tools?

Through a plugin marketplace, and it is far bigger than I expected. Gmail, Google Calendar, Drive, Notion and Slack sit at the front, with categories running through data analytics, infrastructure, sales, scheduling, finance and productivity. Hundreds of entries, from Salesforce and HubSpot down to niche developer services.

Onboarding starts by asking which apps you use every day: Workspace, Slack, Notion, Microsoft 365, Zoom, LinkedIn, Jira, GitHub, Figma, Canva and so on. You pick, it connects, and the Bots can then act inside those tools.

Most of these connect through MCP, the open standard for letting AI tools talk to other software. If you have read about Grok adding connectors earlier this year, this is that idea taken considerably further.

Creating a Bot takes seconds. Pick a colour and shape, give it a name, and it is running. The suggested presets tell you who this is aimed at: Night Shift works overnight and preps a morning digest, Inbox Triage sorts email and drafts replies in your voice, Chief of Staff manages your other Bots and pulls you in for decisions, and Negotiator researches fair pricing and haggles on your behalf.

One screen asks you to give each Bot a job with examples like Invoice Chaser, Weekly Standup and Sales Forecast. Everything about the setup, from the names to the presets to the language, is designed to make you think of these as hires, not tools.

Where it fell over

Partway through, the Bot told me it was fixing what it called the invisible-file issue, and that it would put the tracker rows and draft emails straight into the chat so I could read them.

It had been creating files I could not see. It noticed, said so, and changed approach without being asked, which is more than most software does. But the underlying problem is real, and it is the one to watch with any agent: work happening somewhere you are not looking is work you cannot check.

The second limitation is structural: an agent that signs into your accounts and acts inside them is only as safe as the boundary you set. The approval gate is the whole product here. Turn it off and you have handed your inbox to something that does not know which of your relationships are fragile.

It is also beta software at a premium price. Beta software that can act on your behalf carries a different kind of risk from beta software that only talks.

What I still don’t know

Setup was dramatically faster than building the same thing in Claude Code: a fraction of the time, and the clearest advantage I have seen so far.

What I cannot tell you yet is whether it holds. Two questions are still open: whether the output quality matches what I get elsewhere, and whether it keeps its own file structure orderly over weeks rather than days.

The second question is the one I care about most. Plenty of models do a task beautifully the first time. A workflow you depend on needs the same instructions followed in week six as in week one, without drifting from the original objective.

Drift does not announce itself. It looks like a tracker that stops being updated, or emails that stop sounding like you wrote them. You do not notice on any single day. You notice a month later when the output no longer resembles the job you set.

So treat everything above as a first-week report. I will come back to it after a month of real use.

Is it worth it for a beginner?

For most people reading this, not yet. The price is what stops it, and the product is fine.

At $120 a month minimum the work has to be truly repetitive and truly yours before the maths gets close. Chasing invoices every month, triaging a heavy inbox, running outreach at volume. If a task takes you an hour a week and you hate it, that is roughly the threshold worth thinking about.

It is still worth understanding even if you never buy it. The industry has spent two years selling assistants that answer. This is the first mainstream product built around agents that continue, and that difference will show up in cheaper tools within a year.

If you want the wider picture of who makes what, our rundown of every AI model worth knowing puts Grok in context, and Grok pricing explained covers the subscription tiers in more detail.

What AI can’t do here

• It cannot decide which relationships matter. My outreach Bot could draft fifty emails. It could not know which contact is a friend, which is a rival, and which one message would end things with.
• It cannot judge whether the job is worth doing. It optimises the task you hand it, including tasks you should have dropped.
• It cannot own the outcome. When an agent sends something wrong from your address, it is your name on it.
• It cannot tell you what it is doing where you cannot see. Mine flagged the invisible files itself, and I would put that down to luck.

Agents are good at repetitive work. Deciding where they stop stays yours.

Common questions about Grok Bot

Is Grok Bot the same as Grok?
No. Grok is the chatbot you talk to. Grok Bot is a separate product where you create persistent agents that keep working after you leave.

Can I try it for free?
No. It requires one of three paid subscriptions, the cheapest being Cursor Teams Premium at $120 per seat per month.

Will it send emails without asking me?
Only if you let it. When I set mine up it offered a draft-only mode and respected it. Check that setting before connecting anything that can send.

Why does a Grok product keep mentioning Cursor?
SpaceX acquired Cursor and it has merged with xAI. The combined company signs itself SpaceXAI, and the plugin system carries Cursor branding.

Do I need to code?
No. Setup is a conversation and the connections are one-click. The pricing assumes a professional user, but the interface does not assume a technical one.

Is it safe to connect my email?
Treat it the way you would treat giving a new assistant your login. Start with read and draft permissions, keep the approval gate on, and check what it produces before widening access.

Sources

• SpaceXAI, Introducing Grok Bot (official announcement, 11 August 2026)
• SpaceXAI, Grok Bot product page
• Unite.AI on the Grok Bot launch (pricing and platform detail)
• My own Grok Bot account. All observations and screenshots here are from building and running a Bot on 12 August 2026.

Read next

• What is Grok AI? A beginner guide
• Grok AI pricing explained
• AI agents vs chatbots, explained
• AI agents for beginners
• Grok just added connectors
• Every AI model worth knowing in 2026
• Glossary: what is an AI agent?
• The full AI glossary