Microsoft has announced a significant expansion of its Copilot AI system, introducing new capabilities that allow the assistant to access local files and execute tasks directly across the Windows operating system. This update is part of a broader initiative to transition Windows toward a hybrid intelligence model, which balances local processing with cloud-based resources to improve performance and efficiency.
The new functionality enables Copilot to leverage local context from a user's PC, allowing it to take actions on their behalf while optimizing token usage. By utilizing local models and intelligent routing, Microsoft aims to provide frontier-class AI capabilities directly on personal computers. This shift is supported by new hardware, including the Surface Laptop Ultra, which features NVIDIA RTX Spark silicon, alongside various new devices from partners such as ASUS, Dell, HP, Lenovo, and MSI.
To address the security challenges posed by autonomous agents that can operate continuously and access sensitive data, Microsoft has introduced a new security framework centered on containment, identity, and manageability. A core component of this strategy is the general availability of Microsoft Execution Containers (MXC) on Windows 11. These containers allow organizations to define specific access policies for files and networks, which are then enforced at runtime to ensure that agents operate within secure boundaries.
The integration of MXC is designed to support a wide range of AI agents, including tools from OpenAI, GitHub, and NVIDIA. Furthermore, Microsoft confirmed that Meta’s personal AI agent, Muse for Windows, will soon be available as a native application with MXC integration. This ecosystem expansion is intended to provide users with greater choice while maintaining the security standards required for enterprise and personal environments.
Beyond agent management, Microsoft is enhancing the local compute capabilities of Windows by deploying advanced models directly to devices. This includes the MAI Code 1.1 Flash model, which has been optimized for coding workloads using 3-bit precision to reduce its footprint while maintaining a 256K context window. The company also plans to bring quantized versions of NVIDIA’s Nemotron models to local PCs, further expanding the range of high-performance tasks that can be handled without relying solely on cloud infrastructure.