Microsoft is expanding Windows beyond cloud-based assistance, combining local models, secured AI agents and new hardware in a strategy designed to make the PC an active and manageable AI runtime.
Microsoft has set out a broader Windows strategy built around what it calls “hybrid intelligence”, a model that splits artificial intelligence work between the cloud and the device itself. The aim is to let each task run in the most suitable place while keeping control with the user and IT administrators, a direction that now spans Windows 11, Copilot+ PCs, developer tools and systems designed for local AI workloads.
A central part of the announcement is the general availability of Microsoft Execution Containers on Windows 11. The feature is designed to let organisations define and enforce which files and networks an AI agent may access while it is running, rather than relying only on policy set before execution. According to Microsoft’s presentation, the system also integrates with Agent 365 and Intune, so administrators can separate an agent’s identity and actions from those of the person using the machine. WindowsReport said the containment model can extend across process and session isolation, WSL, virtual machines and Windows 365 for Agents, giving companies different levels of control depending on their security requirements.
Microsoft is also pushing larger AI models on to local hardware. The company said MAI Code 1.1 Flash uses 3-bit precision to shrink the model by nearly 80% while keeping a 256K local context window, a technical balance intended to preserve useful model behaviour on a PC. Windows ML is being extended with llama.cpp support across GPU, NPU and CPU hardware, and Microsoft is testing GitHub HydraFusion to route workloads between cloud and on-device models. Experimental previews for the GitHub Copilot app, GitHub Copilot CLI and Visual Studio Code are expected later in October, suggesting that the company wants developers to treat the device itself as a serious AI runtime rather than only a thin client for remote inference.
The hybrid approach is also being folded into Copilot on Copilot+ PCs, where Microsoft says the assistant will, with user permission, use local files and recent activity, perform actions in Windows and call models running directly on the device. Those capabilities are due to roll out over the coming months across Copilot Home, Code and Autopilot. Separately, Windows Search is being turned into a more active control surface: Windows Insiders will be able to use it to switch settings, manage windows and send messages without opening separate menus, according to reports from Windows Central. Microsoft describes the point of the redesign as reducing friction by bringing search, system actions and Copilot into one place.
The hardware side of the strategy is moving in the same direction. Microsoft and Nvidia have unveiled new AI-focused Surface and developer machines, including the Surface Laptop Ultra and the Surface RTX Spark Dev Box, both tied to Nvidia’s RTX Spark platform and aimed at heavier on-device workloads. Axios reported that the launch is part of Microsoft’s attempt to reset its AI PC push after earlier Copilot+ machines were complicated by delays to Recall. Microsoft is now presenting the wider ecosystem, from local agents to high-end workstations and data-centre-class systems, as a single platform for secure, mixed cloud-and-device AI. The company has also said DGX Station for Windows systems should arrive later this year, with support for models of up to 1 trillion parameters, underscoring how far it wants Windows to move from a general-purpose desktop operating system towards a managed AI computing layer.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
