Tech

Microsoft opens Windows to AI agents and ships an Nvidia-powered Surface

Execution Containers reach general availability, letting companies fence off what an agent can touch, as the Surface Laptop Ultra opens for pre-order at $2,600.

Bar chart of announced US prices for Microsoft RTX Spark Windows hardware from $2,600 to $6,000
Data: TechCrunch; Chart: 1% Better

Microsoft made its agent containment technology generally available on Windows 11 on Wednesday and opened pre-orders for a laptop built around Nvidia's RTX Spark chip.

Zoom in: Microsoft Execution Containers let an organization define which files and networks an agent may reach, with those policies enforced at runtime. Microsoft says traditional sandboxes were designed for apps, where containment needs hold still, rather than for agents that change what they touch with every prompt and tool call.

  • Codex from OpenAI, GitHub Copilot, Replit, LM Studio, OpenShell from Nvidia and Unsloth AI already support it, with Anthropic's Claude Code, Perplexity, Manus and Raycast listed as coming.

By the numbers: The Surface Laptop Ultra starts at $2,600, with a faster chip at $3,700 and configurations running to $5,900; Microsoft says the top model is already out of stock. A Surface RTX Spark Dev Box workstation starts at $6,000.

  • Microsoft is offering up to $1,000 off to buyers who trade in a MacBook Pro.

The backstory: Nvidia lined up Microsoft and other PC makers behind the RTX Spark chip in June with specifications and pricing still undisclosed. Dell detailed its XPS 16 Creator Edition the same day at $3,800, and ASUS, HP, Lenovo and MSI have machines coming.

What they're saying: "One of the things we realized in the last 3-4 years is that just having a model doesn't do much for anything," Microsoft chief executive Satya Nadella said. "You really do need to orchestrate, and you need to have memory outside of the model."

The big picture: Microsoft calls the strategy hybrid intelligence: route a task to a local model when the hardware can handle it and to the cloud when it cannot, with Copilot permitted to read local context and act on the user's behalf.

Why it matters: Agents that read files and run code need a cage before companies turn them loose, and Microsoft is betting the operating system is where that cage belongs. Running models locally also shifts AI spending off a per-token meter and onto hardware the buyer owns.

Go deeper: Microsoft

Also: TechCrunch

  • microsoft
  • windows
  • nvidia
  • ai agents
  • surface

Newsletter

Get the daily brief

The day's stories and one idea to get 1% better, in a free five-minute email every morning.

Get Started
All news