Microsoft has begun to remove or rename some of its Copilot integrations, but this should not be seen as a step away from its commitment to artificial intelligence. At the Acer conference during IFA 2026, the company explained its plan to integrate AI more deeply into Windows. In recent months, Microsoft has sought to shift the focus away from the Copilot branding in Windows 11. The name is gradually disappearing from certain applications, while the company emphasizes a return to the core features of the operating system. However, this does not signal a retreat from AI. Instead, Microsoft has clarified its vision for AI within Windows: fewer features labeled as Copilot, more processing done locally on the PC, and a long-term goal of creating a system capable of hosting AI agents without constantly relying on the cloud. A concrete example of this strategy is Project Zenith, a new category of PCs designed to develop and run large AI models locally. These machines are not just for developers or high-end configurations; Microsoft aims to bring what it calls "unmetered intelligence" to all users, ensuring that AI can operate without relying on cloud-based resources whenever possible. This idea was first introduced by CEO Satya Nadella at Build 2026, and later reiterated by Mark Linton, vice president of Windows and devices, at IFA 2026. The goal is not to provide free AI with no limits, but to enable Windows to perform more tasks locally using the PC’s CPU, GPU, and NPU, reducing the need to access distant servers. This shift toward local AI is also evident in recent updates to Windows applications. For instance, the Copilot integration in Notepad is now labeled "Writing Tools," while similar changes have been made in Photos and the Screenshot tool. These updates reflect a broader effort to integrate AI more seamlessly into the user interface, moving it away from the foreground. This evolution supports another key project at Microsoft: the development of AI agents that can operate within Windows. These agents can perform tasks over time and frequently access AI models, making it more efficient for them to rely on local computing resources rather than the cloud. However, questions remain about how much autonomy these agents should have. At IFA 2026, Mark Linton addressed concerns, particularly in enterprise environments, where uncontrolled access to systems and user resources could pose risks. To address this, Microsoft has been working to define clear boundaries for AI agents. The company plans to isolate them in dedicated environments, assign them specific identities, and regulate their permissions so they can use PC resources without the same access rights as the user. These measures, including Microsoft Execution Containers and identity management systems, were already outlined at Build 2026. By the end of the year, Microsoft promises to provide more details on how these features will be implemented in Windows.