xAI gives its agents their own computers

The agent now gets its own computer

Grok Bot keeps working after you close your laptop.

xAI’s new beta gives AI agents cloud computers where they can sign into websites and apps, handle multi-step jobs, and ask for approval when needed. Inside SpaceXAI, bots are already updating CRMs, processing Gmail invoices and reproducing software bugs.

The important part is access. Giving an agent its own computer lets it work with software that lacks an API or MCP connection, which opens up far more ordinary office work than tightly integrated agents can reach today.

xAI also says multiple bots can coordinate, with one managing specialists across areas like recruiting, expenses and engineering. Users can demonstrate a workflow once, then save it as a routine for later runs.

For now, this is an early beta limited to SuperGrok Heavy, Cursor Ultra and Cursor Teams Premium subscribers.

I’d watch the approval layer closely.

A bot clicking around your actual accounts at 2 a.m. is useful only when you know which decisions it should make alone.

Google’s medical AI can now examine what it sees

A cough sounds different when the system can actually hear it.

Google Research and DeepMind have extended AMIE, their research medical AI, into real-time video consultations using Gemini, Project Astra and multiple AI agents. It can observe visible symptoms, listen to patients and guide virtual physical exams while reasoning through a diagnosis.

That matters because text strips useful clinical information from medicine. A system that can notice gait, discomfort or a cough during a live call gets much closer to the information doctors actually use during telehealth visits.

In a randomized study with patient actors, clinical evaluators rated AMIE favorably on diagnostic accuracy, history-taking, treatment decisions and communication. The actors also preferred video consultations to text chat.

The evidence still comes from simulations.

Google says AMIE remains a research system, so the harder test is whether those results survive messy clinics, unfamiliar patients and symptoms that do not present cleanly on camera.

Nvidia needs its GPUs to age well

Nvidia wants Wall Street to finance $500 billion of AI infrastructure on the assumption that its chips can remain valuable for years.

The financing would help startups and neoclouds buy expensive GPU capacity. Nvidia says CUDA updates can keep older hardware productive even as newer chips arrive.

That assumption deserves scrutiny. If GPUs lose earning power quickly, lenders could end up repossessing racks of used chips from weak borrowers and selling them into a market where prices are already falling.

China makes that risk harder to price because a surge of cheaper domestic compute could accelerate depreciation, although export controls currently limit Huawei’s reach in the U.S. Nvidia still controls more than 75% of the American AI chip market, and H100 rental rates have actually risen this year.

Investors may demand yields of 11% to 17% to carry the risk.

That sounds expensive.

The question is how those GPUs look after three years of humming inside a data center.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.