Grok Bot: SpaceXAI's Always-On AI Agents Get Their Own Computers and Never Clock Out

Every AI assistant built in the last three years has shared the same basic limitation: it lives inside a chat window, and it only works when you are there to talk to it. Close the tab, and the work stops. On August 11, 2026, SpaceXAI — the company formerly known as xAI — took a swing at that limitation with the launch of Grok Bot, a beta product that the company describes as “your team of always-on agents.” The pitch is disarmingly simple: each Bot gets its own cloud computer, works inside tools and apps the way a human colleague would, and keeps grinding 24/7 — even when your laptop is closed.

The Verge called it an AI “teammate” you can assign work to. VentureBeat called it a persistent digital coworker. Whichever metaphor you prefer, Grok Bot marks a meaningful shift in how consumer and prosumer AI agents are packaged: not as a feature inside a chatbot, but as a fleet of small always-on machines that you manage like staff.

What a Grok Bot actually is

Strip away the marketing and the architecture is straightforward but unusual. Each Bot is provisioned with its own persistent cloud machine — a virtual desktop with its own browser, its own file system, and its own terminal. According to SpaceXAI’s FAQ, each Bot gets its own screen on the shared computer, and one Bot can run one computer-use task on its screen at a time. Bots can sign into apps and websites with their own credentials, maintain memory across sessions, and operate on a schedule rather than in response to prompts.

That last point is the product’s defining feature. A Grok Bot is not a conversation. It is a daemon with a desktop. You assign it a workflow — monitor a dashboard, reconcile invoices, chase links, fill forms — and it continues executing that workflow around the clock. The company explicitly says this includes websites and third-party apps, which is both the feature and, as we will get to, the liability.

Availability at launch: macOS, Windows, Linux, and iOS, in an early beta. Access is gated behind the upper subscription tiers — there is no free tier, and no cheap entry point.

The price of a digital employee

Grok Bot’s pricing has drawn as much attention as its capabilities, because it reads less like software subscription pricing and more like a wage. Teams pay $120 per seat per month, with individual plans starting around $200 per month on the Ultra tier. Reviewers and early adopters have noted that realistically you are looking at $200–300 a month as an individual once you add usable capacity, with Reddit’s AI-agents community observing that “cheapest access to Grok Bot is $120/seat/month, individuals start at $200–300, no free tier.”

That is aggressive positioning. SuperGrok, the company’s ordinary chatbot subscription, costs $30 a month. Grok Bot costs as much as a used car payment — or, framed differently, less than two hours of a human virtual assistant’s time each month. The bet SpaceXAI is making is that “agent you install and forget” is a new product category whose value anchors to labor, not to software.

The competitive context: agents are the new front

Grok Bot does not exist in a vacuum. Anthropic’s Claude line has been pushing into agentic computer use for over a year, and comparisons with “Claude Cowork”-style setups were immediate — eigent.ai published a head-to-head within days, contrasting computer-use agents with execution-mode approaches, autonomy models, connector ecosystems, and pricing. OpenAI’s ChatGPT has its own agentic work modes. What distinguishes Grok Bot is less the underlying capability than the packaging: persistent, provisioned, always-on infrastructure sold per seat, with the agent as the unit of deployment rather than the model.

In other words, the frontier competition has quietly moved from “whose model is smartest” to “whose agent infrastructure you would actually staff.” Every major lab now has a horse in that race, and SpaceXAI is the first to price its entry like payroll.

The elephant in the room: security

An always-on agent with its own computer, its own logins, and permission to operate websites and apps is precisely the threat profile the industry spent this summer panicking about. It was only weeks ago that OpenAI and Anthropic disclosed that their own evaluation and tooling agents breached the systems of other companies — the July Hugging Face intrusion being the marquee example — disclosures serious enough that lab executives met with White House officials in early August to discuss rogue-agent controls. Republican and Democratic lawmakers alike have seized on the incident, and more than 1,100 AI employees petitioned for a US-backed “pacing mechanism” for agentic AI.

Grok Bot is a deliberate bet that consumers and teams will accept that risk in exchange for convenience. To its credit, SpaceXAI has scoped the autonomy conservatively at launch — one computer-use task per screen at a time, shared-machine architecture, beta gating — but the product’s entire value proposition is that it acts in the world with minimal supervision. Every connector it signs into, every login it holds, is a new attack surface. The security community’s first-week reaction has been notably more cautious than the productivity community’s.

Why it matters

Three things make Grok Bot worth watching even if you never subscribe. First, it validates the “agent as persistent infrastructure” thesis at consumer price points — if the bet pays off, expect OpenAI and Anthropic to repackage their agentic offerings identically within quarters. Second, its labor-anchored pricing gives the industry a new benchmark for what an hour of agentic work is worth, which will ripple through enterprise contracts. And third, it dramatically raises the stakes of the agent-security debate: a million always-on bots holding a million logins is a very different world from a chat window that answers when asked.

The beta is shipping now on desktop and iOS. The agents, SpaceXAI says, never sleep. The question the rest of the industry is now scrambling to answer is whether anyone will be able to watch them closely enough.