← Back to feed

SpaceXAI launched Grok Bot on August 11, an AI agent platform built jointly with Cursor that gives each bot its own cloud computer and lets it sign into apps, websites, and inboxes the same way a human employee would.

The core bet is simple and pointed: most agent frameworks break because they hit API limitations or require workflow builders. Grok Bot skips both. Each bot authenticates directly through app interfaces, retains context between sessions, and passes work back to the user only when it needs a human decision.

What It Does

Grok Bot launched in beta across macOS, Windows, Linux, and iOS, with Android listed as coming soon. Access requires a SuperGrok Heavy, Cursor Ultra, or Cursor Teams Premium subscription. Enterprise customers can join a waitlist.

Multiple bots can run simultaneously and share context with each other, which matters for longer workflows. SpaceXAI described internal usage since the product was tested under the codename Sand at Cursor from late June: engineers using it for overnight sales research, pulling expense receipts from email, fixing bugs, managing CRM updates, handling customer support for an online store, and negotiating with suppliers by voice.

Users can also walk a bot through a task once, and it saves the sequence as a repeatable routine.

Context: The Merger and the Race

SpaceXAI agreed in May to acquire Cursor for $60 billion. That deal is expected to close within days. Grok Bot is the first joint product to carry both brands, and it signals where the combined company is going: from developer tooling to general-purpose office agents.

The timing is not accidental. Claude Cowork launched in April and ChatGPT Work shipped in June. Both take a similar approach: AI that runs tasks asynchronously, works through existing apps, and surfaces results rather than prompts.

SpaceXAI shipped its first coding agent in May and Grok 4.5 in July. On last week’s earnings call, Musk said Grok 4.6 could ship within days. Grok Bot is the consumer-facing, non-developer product the company has been missing.

What It Does Not Do

Grok Bot is in early beta and the company published no quantitative benchmarks for task completion rates or error rates. Internal use cases at SpaceXAI and Cursor are vendor self-reported. Independent evaluation at the level of METR or Anthropic safety card testing has not been published.

The $60 billion acquisition brings Cursor’s 50 million developer users and its IDE into the same product surface as Grok. How that scales past developers into general office work — and how it competes with Anthropic’s $0.08/hr Claude Managed Agents — will depend on what the production version delivers.

Key Details

• Beta access: SuperGrok Heavy, Cursor Ultra, Cursor Teams Premium subscribers
• Platforms: macOS, Windows, Linux, iOS (Android coming)
• Pricing: Bundled into existing subscriptions for now; enterprise on waitlist
• Architecture: Each bot runs on its own cloud computer, signs in via UI (not API), persists context across sessions
• Internal use cases: Sales research, bug fixes, CRM updates, customer support, supplier negotiation
• Underlying model: Grok (SpaceXAI). Grok 4.6 expected “within days”