Microsoft has long marketed Windows 11 as an AI-first operating system, but at its recent Build developer conference, the company shared concrete technical steps toward that goal. The focus is shifting from cloud-dependent chatbots to on-device intelligence and autonomous agents that can execute complex tasks using natural language.
Unmetered, offline intelligence
Anastasiya Tarnouskaya, product manager for Windows ML, emphasized that Windows 11 PCs now offer “unmetered intelligence.” This means users can run AI workloads locally without incurring token costs or sending sensitive data to the cloud. Running models on-device also reduces latency, making interactions feel more immediate.
Tarnouskaya noted that over 500 million PCs are already running local AI workloads. She stated that thanks to advancements in hardware and software stacks, “every Windows PC is becoming increasingly AI-capable.” This approach ensures privacy and performance, as the processing happens directly on your machine’s GPU or NPU.
Windows ML and the Foundry portfolio
To enable this shift, Microsoft shipped Windows ML last fall. This framework helps developers create offline AI applications by mapping localized AI models to hardware components like GPUs and neural processors. It is part of a broader “Foundry” portfolio that includes Foundry Local for running open-source models and Windows AI APIs.
These APIs automate specific tasks such as conversation summarization, speech recognition, and video upscaling. Major apps are already adopting these capabilities. Microsoft Office, Photos, and Teams use on-device AI, while Outlook uses the Phi Silica model to summarize emails locally. Third-party developers from Adobe to WhatsApp are also building local AI experiences.
Natural language agents take control
Beyond simple prompts, Microsoft is introducing AI agents that can handle long-running tasks. Samantha Song, product manager for Windows, described a future where natural language maps directly to system outcomes. Users can describe a task—such as personalizing colors, wallpaper, or menus—and the agent executes it as one coherent action without manual setup.
For this to work, developers create “skills files” that define how an agent behaves. These skills are reusable across different contexts. Song demonstrated how users could simply type or speak their preferences, and the system would adjust settings automatically. This moves beyond chat interfaces into actual system control.
Enterprise implications and hardware strategy
The rise of agentic AI has significant implications for enterprise IT. Song suggested that users could switch into a “secure finance mode,” where the system automatically aligns apps, access boundaries, and environments. At Build, LLMware.ai demonstrated an agent on a Qualcomm laptop that collected Jira issues in real-time, summarized them locally, and emailed daily summaries to the team.
Jack Gold, principal analyst at J. Gold Associates, warned that these efforts will force enterprises to rethink hardware strategies. Since different AI chips excel at different tasks, Microsoft must support multiple chip architectures to give IT pros choice. He recommended that any new PC purchases, especially for enterprise upgrade cycles, prioritize AI-capable hardware.
Leonard Lee, principal analyst at Next Curve, noted that while Samsung and Lenovo are rolling out “personal AI” features, ensuring safe deployment remains a challenge. As Microsoft embeds these capabilities deeper into Windows 11, users should expect more autonomous, privacy-focused AI interactions in the coming months.
Source: Computerworld
Over to you: Do you prefer AI features that run locally on your device for privacy, or do you trust cloud-based models for better performance?
