Overview
- Microsoft announced at its Windows event that it is adding a hybrid intelligence model that routes tasks between on‑device AI and cloud models to balance cost, latency and capability.
- The company made Microsoft Execution Containers (MXC) generally available to isolate agents, enforce runtime policies and let administrators control which files and resources an agent may access.
- Windows will use “intelligent routing” to pick local models for simple or privacy‑sensitive tasks and cloud models for heavier workloads, a design Microsoft says reduces data‑center processing and customer cloud bills.
- GitHub Copilot and other agents will gain permissioned local context and actions, with experimental Copilot features and Copilot+ PC model support starting to roll out in October.
- Microsoft highlighted smaller, optimized models and stronger client silicon as keys to making local inference practical, but security and privacy commentators say enterprises must still vet agents despite the MXC containment layer.