Tuesday, August 11, 2026. SpaceXAI shipped Grok Bot, and it’s the first real answer to what the SpaceXAI-Cursor merger was actually for.
An always-on AI teammate that gets its own cloud computer, logs into your apps. And finishes jobs while you sleep.
Not text. Work.
Grok Bot Features and How It Works
Here’s what threw me. Each bot doesn’t just chat back at you. It gets a persistent machine, signs into the accounts you already use, clicks through real interfaces. And hands back finished output. The release notes call it “Durable AI teammates that work on a persistent cloud computer, with messaging, approvals, connectors. And routines.”
The feature set, straight from those notes:
– Messaging. Bots talk to you and to each other
– Approvals. They pause and ask before doing something risky
– Connectors. They plug into your existing tools and apps
– Routines. They run on a schedule, not just when you poke them
The trending X post adds that the bots “have their own computer, work inside tools and apps like you do. And keep working 24/7,” and you can “place Bots in a group chat where they can coordinate on their own.”
A bot surfaces when it needs your sign-off, then goes quiet and keeps going. Particle’s reporting puts it plainly. The bots “carry out multi-step jobs end to end rather than only returning written suggestions.” VentureBeat echoes that: each bot “signs into a user’s existing apps. And completes work end to end, returning only when approval is needed.”
This is an agent platform. Not a feature stapled onto a code editor. That distinction tells you what the thing is actually for.
Grok Bot and the SpaceXAI-Cursor Merger
People spent months arguing whether the SpaceXAI-Cursor deal was about owning the code layer or grabbing distribution. The first product out answers the question.
It’s a general-purpose worker that gets a computer and logs into your life.
Not a better model for the Cursor editor. Not a benchmark win.
That’s a loud signal about where the combined enterprise thinks the market is going. They’re betting that “models that write responses” is already a commodity. And the money sits with “workers that own a task.”
I’ll be straight with you.
When the merger closed, I expected them to ship a sharper Cursor model. I was wrong. What came out is way more ambitious. And honestly I’m not sure the beta can hold the weight of the pitch yet.
For builders and small agencies, that reframes the question worth asking. The competition isn’t “which model writes better code” anymore.
It’s “which platform runs a job while I’m not looking.”
Grok Bot Pricing, Platforms, and Availability
The beta isn’t open to everyone.
Quick breakdown of where things stand:
– Supported platforms: macOS, Windows, Linux, iOS (desktop and iOS)
– Android: listed as “to follow” — no date given
– Subscriber access: SuperGrok Heavy, Cursor Ultra, Cursor Teams Premium
– Enterprise users: waitlist only, no timeline
– Status: beta — explicitly unfinished
The gap here is worth sitting with.
The pitch is AI teammates that never sleep. Reality: a beta running on the smallest possible audience, missing at least one major platform. If you’re on one of those premium tiers, you can try Grok Bot today. If you’re not, you’re reading about a product you can’t touch yet. Which means the smart move is watching what breaks before you build anything on it.
Grok Bot Security: The Question Nobody Answered
Here’s where I get nervous.
Every demo shows a helpful agent logging into your apps and finishing your work.
What the launch materials don’t dwell on is that “signs into a user’s existing apps” means this bot is holding real credentials and taking real actions inside live accounts. One analysis of the rollout described Grok Bot as “a credential-holding autonomous agent gated to their highest-priced plans,” and warned that IT and security leaders “should assume power users can now expense an unsanctioned computer-using bot straight into corporate inboxes.”
For a small agency like mine, this is the part that matters most. I build automations for clients. The moment an agent logs into a client’s email, CRM, or bank with stored credentials, the liability picture changes completely. An API integration fails closed when it hits an error. An agent clicking through a real browser session can do things you didn’t script and didn’t anticipate. Because it’s operating in a general-purpose environment.
The approvals feature is supposed to be the safety valve.
But a valve only works if you actually configure it, check it. And treat every single approval like it matters.
So who’s liable when a bot makes a wrong call inside a live account?
How do you audit what it did after the fact? None of the launch coverage touches that. That silence is the real gap right now. And it’s the thing a careful operator has to fill with their own process.
Grok Bot FAQ
What is Grok Bot? Grok Bot is SpaceXAI’s always-on AI teammate. The first joint product of the SpaceXAI-Cursor merger. Each bot gets its own persistent cloud computer, logs into the apps you already use. And completes multi-step jobs end to end. It surfaces when it needs your approval, then keeps working.
How much does Grok Bot cost? There’s no standalone price. Access is bundled into premium subscriptions: SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium.
Enterprise users can join a waitlist but can’t use the beta yet.
What platforms does Grok Bot support?
macOS, Windows, Linux, and iOS right now. Android is “to follow” with no release date.
Is Grok Bot safe to use with my real accounts? That’s the open question. The bot holds credentials and takes real actions in your live accounts. The approvals feature is the main guardrail. But launch coverage doesn’t address liability or auditing when a bot makes a wrong decision.
What I’d Actually Do
My read: Grok Bot is the most interesting agent launch of the year since it commits to a clear thesis — an AI worker should own a whole task, not just generate a response.
But “own a whole task” and “own a whole task safely inside my client’s accounts” are two very different bars.
The launch clears the first one and is silent on the second.
If you run a small operation, don’t rush in. Pick one low-stakes, reversible job. Point a bot at it. Watch every approval like a hawk. Learn what the failure modes actually are before you let anything near money, credentials, or client data.
The agents that win will be the ones whose builders took the security questions seriously. Be one of those builders. I’ll track Grok Bot as the beta opens up and report back on what holds and what breaks. If you want my notes on testing AI agents against your real workflows without handing them the keys to your business, reply and tell me what you’d automate first.
