Microsoft’s Windows Hybrid Intelligence Event: MXC Security, Copilot Previews and RTX Spark Hardware

What the event was, briefly​

Microsoft held the event in San Francisco on October 7. It was the company's first major Windows showcase since May 2024, and the speaker list suggested more than a routine update. CEO Satya Nadella, Windows and Surface chief Pavan Davuluri, and NVIDIA CEO Jensen Huang were all confirmed to appear on stage. The invitation called it "a conversation on how local AI will shape the next chapter of the PC."

Thurrott's complaint is about context. To follow the show, he argues, you need to keep a lot in your head at once:

• Windows 11's year of quality fixes
• Expensive cloud AI, and local AI that never quite takes off
• The move from AI PCs to Copilot+ PCs to RTX Spark machines
• Hybrid AI orchestration and the Copilot app
• Microsoft AI's effort to depend less on OpenAI and Anthropic models
• AI agents
• Developer access to all of the above

He also says Surface has shrunk to two basic lines, Laptop and Pro. Microsoft's own announcements add a developer desktop to that picture.

In fairness, the event did have a single theme. Microsoft calls it hybrid intelligence. In his companion Windows Experience Blog post, Davuluri describes Windows as a platform where agents can run locally when that makes sense, reach the cloud when they need to, and stay under the security and management controls that organizations expect. Every segment of the show tied back to that idea, even if the presentation didn't make the links obvious.

Summary: The event made more sense than it looked, but you needed a map to follow it.

Piece one: agent security (generally available now)​

For IT pros this is the most concrete news. Microsoft says Microsoft Execution Containers (MXC) are now generally available on Windows 11. Organizations can define which files and networks an agent may access, and Windows enforces those policies at runtime.

Microsoft's security model has three parts:

1. Containment controls what an agent can reach.
2. Identity records which agent acted, so IT can tell agent activity apart from the user's.
3. Manageability feeds agent activity into Agent 365 and Intune for policy, monitoring and governance. A footnote says governance at scale may require additional services.

Microsoft says MXC already works with OpenAI's Codex, GitHub Copilot, OpenClaw, Replit, LM Studio, NVIDIA's OpenShell and Unsloth AI. Anthropic's Claude Code, Perplexity, Manus, Raycast and others are listed as coming. Meta's Muse personal agent is also promised as a native Windows app with MXC integration.

Pay attention to tense here. In the Surface blog, Microsoft says upcoming Windows capabilities will use Microsoft Entra to separate agent activity from user activity and extend Agent 365 controls to local agents. So containment ships today, but the full identity and governance setup is still on the way. Microsoft hasn't published licensing details or a complete rollout schedule for those parts.

Summary: MXC is real and usable now. Treat the Entra and Agent 365 integration for local agents as a roadmap item.

Piece two: hybrid AI for developers and Copilot (mostly previews)​

Microsoft is bringing MAI Code 1.1 Flash, its coding model announced at Build, to run on local PCs. The company gives these specs:

• 137 billion total parameters, 6.8 billion active
• 3-bit precision, which Microsoft says cuts model size by nearly 80% while preserving coding quality
• A 256K-token context window when running locally

Other models announced for RTX Spark machines include an upcoming NVIDIA Nemotron model with more than 70 billion parameters, quantized to 2-bit so it uses just over 20GB of memory, and DeepSeek V4 Flash at 284 billion parameters. All of these figures come from Microsoft. Independent benchmarks haven't been published yet.

On routing, GitHub's HydraFusion currently sends each task to the best-suited cloud model. It will now be able to use local models as well. Microsoft says this comes to the GitHub Copilot app, Copilot CLI and Visual Studio Code as an experimental preview later in October. Windows ML is also gaining llama.cpp support, which gives developers more open-source model options.

Copilot is getting three local abilities on Copilot+ PCs:

• Local context: with your permission, it can read your files and recent activity.
• Local actions: it can organize files, run diagnostics and carry out workflows.
• Local models: it can use models running on your PC.

Microsoft says these features should begin rolling out "in the coming months," and that availability will vary by device, market and silicon. Gadget Hacks noted on event day that the Windows agent and Copilot features tied to the keynote remain staged previews rather than shipping software.

Summary: The developer tools arrive as previews this month. The local Copilot features have no firm date.

Piece three: Windows Search gets actions​

This is the most everyday change in the lineup. Windows Search on the taskbar is gaining what Microsoft calls thousands of actions. Microsoft's example is typing "Text Elisa I'll be there soon" to send a message straight from Search. The actions began rolling out to Windows Insiders in the experimental channel on October 7.

Separately, Microsoft will add an opt-in option to connect the new Copilot to taskbar Search. You'd get quick answers inline and could jump to the full Copilot app for a longer conversation. That option goes to select markets first, later this year.

If you were worried about Copilot taking over your search box, Microsoft says it's opt-in. Expect forum threads asking where the toggle lives.

Piece four: the hardware (preorders open, with dates)​

Surface Laptop Ultra is the headline device. Microsoft lists these specs:

Microsoft also lists a new thermal design, user-removable storage, and magnetic USB-C charging it calls Magnetic Connect. A footnote says the GPU can address less than the full 128GB of memory, depending on configuration.

The Surface RTX Spark Dev Box costs $5,999. It's sold only through Microsoft.com in the U.S. and ships in November. It comes set up with VS Code, Git, GitHub CLI, GitHub Copilot, WSL, Python and Node.

Microsoft says RTX Spark laptops from ASUS, Dell, HP, Lenovo and MSI opened for preorder October 7 and also ship October 16. RTX Spark mini desktops and dev boxes come later this year. DGX Station for Windows, built on NVIDIA's GB300 Grace Blackwell Ultra Desktop Superchip, also arrives later this year. Microsoft lists up to 748GB of coherent memory and support for models up to a trillion parameters. This is a deskside system for enterprises, not a feature for ordinary PCs.

Some third-party coverage disagrees on these details. One roundup from ExplainX listed the Laptop Ultra's price as not confirmed in the coverage it found, with pre-event reports pointing to about $3,000. It also placed partner PCs later in 2026. Microsoft's own posts are more specific: $2,599 and an October 16 ship date. Go with Microsoft's figures, and expect prices and dates to differ by market and configuration.

The benchmark caveat​

Microsoft claims RTX Spark PCs deliver up to 2.1x faster time to first token than a 16-inch MacBook Pro with M5 Pro and 64GB of memory. Here's what that number rests on:

• Microsoft commissioned the testing in September 2026
• It used preproduction units with 64GB of memory
• The test ran a single model, Qwen3.5 27B at Q4K-Medium in llama.cpp
• It used a fixed 8,192-token prompt

Microsoft itself says performance "varies significantly." Time to first token mostly measures how fast the machine processes your prompt, which is a GPU strength. It doesn't tell you sustained generation speed, battery life under load, or how retail units behave. Wait for independent reviews.

What came before October 7​

Thurrott is right that much of this didn't start last week. Microsoft's May 31 post had already introduced RTX Spark. That post described Windows scheduler tuning and Prism emulator work for the chip, TensorRT in Windows ML, and a higher limit on how much system memory the GPU can use, which lets larger local models load. It also said the Surface Laptop Ultra and RTX Spark were already known when Microsoft confirmed the event in September. October 7 mostly filled in prices, dates and software.

The verdict​

The "drunken meander" line has a point. Microsoft tried to cover security, models, Copilot, Search, gaming and three hardware tiers in one show. Each audience probably cared about one slice and sat through the rest. BGR's live blog put it a little more kindly: the event didn't seem interested in appealing to a wide audience, though Microsoft appeared aware the public has trust issues with Hybrid Intelligence.

The structure underneath is still clear once you sort it:

• Available now: MXC, Search actions in the experimental Insider channel, and preorders for hardware arriving October 16
• Previews this month: HydraFusion local routing in GitHub Copilot and VS Code
• Coming months: local Copilot features on Copilot+ PCs and opt-in Copilot in Search
• Later: the Entra and Agent 365 controls for local agents, DGX Station, and RTX Spark mini desktops

Microsoft says it will share more at Ignite in November. For IT admins, the best next step is to test MXC policies now, before agent identity controls arrive and before users start installing agents on their own.