xAI launches Grok Bot, always-on AI agents with their own cloud computers

Grok Bot positions xAI directly against Anthropic's Claude and OpenAI's agent products in the emerging 'digital labor' category. Each agent is provisioned with its own cloud computer, can integrate with existing tools, retain context across tasks, and share information with other agents — enabling coordinated, multi-step knowledge work and learning workflows by observation.

The product is in public beta, bundled into existing premium subscription tiers rather than sold separately, and available on desktop and iOS. Musk framed it as a step toward autonomous teammates that operate continuously without supervision. He also teased Grok 4.6 arriving later in the week and, per one source, upgrades to Grok Image 2.0 with templates and precise editing that rank #2 on Arena text-to-image behind OpenAI's gpt-image-2.

Competitively, the always-on, tool-signing design is aggressive — and controversial. Agents that authenticate into real apps and websites raise the same security concerns that dominated this week's discourse around agentic autonomy, including prompt injection and unintended actions.

Skeptics point to the reputational baggage of unsupervised agents acting in production systems, echoing the viral Claude gym-booking incident. The bundling strategy also signals xAI is racing to lock in users before rivals mature their offerings. What to watch: real-world reliability, the security posture of agents with live credentials, and whether Grok 4.6 meaningfully closes the gap with frontier models.