Microsoft backs open agentic stack

Microsoft spent this week tying together a set of announcements that, taken together, show how it wants to compete in the next phase of enterprise AI: not only through models, but through the operating systems, governance controls and workflow software that sit around them. The company used Open Source Summit North America and Build-related updates to push Azure Linux, container infrastructure, Microsoft 365 Copilot agents and on-device AI for Copilot+ PCs. Microsoft’s own descriptions emphasized cloud-native and AI workloads, while Linux Foundation materials pointed to a growing standards push around open agent systems. ### Why did Microsoft pair Linux releases with “agentic” AI messaging? Microsoft said on May 18 that Azure Linux 4.0 would enter public preview on Azure Virtual Machines and that Azure Container Linux had reached general availability. In its Open Source Summit post, the company described the two releases as a hardened Linux base “purpose-built for cloud native and AI workloads.” (opensource.microsoft.com) That pairing matters because Microsoft is not presenting agents as a feature that floats above infrastructure. Its pitch links agent software to the systems that run enterprise workloads: virtual machines, containers, Kubernetes and security-hardened operating environments. The company’s Azure Linux documentation says the container host is maintained by Microsoft and based on its open-source Linux distribution, formerly known as CBL-Mariner. (opensource.microsoft.com) ### What is Microsoft adding on the user-facing side? Microsoft 365 Copilot’s Researcher agent is designed for complex, multi-step research tasks and produces source-cited reports using both web data and a user’s work data, including files, emails, meetings and chats they are allowed to access, according to Microsoft documentation. Microsoft introduced Researcher and Analyst as reasoning agents for work in March 2025. (learn.microsoft.com) That gives Microsoft a second layer in the stack. One layer is infrastructure such as Azure Linux and container hosts; the other is application software that sits inside daily office workflows. Researcher is notable because Microsoft is framing it as a work product inside Microsoft 365, not as a standalone chatbot. That is an inference from Microsoft’s product documentation and launch materials. (learn.microsoft.com) ### Where does DeepSeek fit into this? Microsoft’s Windows and Azure materials show DeepSeek R1 being used as part of its model-distribution strategy rather than as an exclusive platform bet. In January, Microsoft said NPU-optimized DeepSeek-R1 distilled models were coming to Copilot+ PCs, starting with Qualcomm Snapdragon X devices, with Intel Core Ultra 200V support to follow. In March, Microsoft added 7B and 14B distilled models for Copilot+ PCs via Azure AI Foundry. (learn.microsoft.com) Microsoft’s Copilot+ PC pages also continue to stress local AI processing through NPUs capable of more than 40 TOPS, which is the hardware basis for running those models on-device. The practical effect is that Microsoft can point to AI running in the cloud through Azure AI Foundry and at the edge on Windows hardware. (blogs.windows.com) ### What does the open-standards piece add? The Linux Foundation said on May 18 that the Agentic AI Foundation added 43 new members in the past quarter, bringing total membership to 180 organizations. The group said the additions included four new Gold Members — F5, GoDaddy, Stripe and TRON — plus 27 Silver Members and 12 Associate Members. (microsoft.com) The foundation has described itself as a neutral home for open standards powering agentic AI systems, with projects including MCP, AGENTS.md and Goose. That matters because Microsoft’s infrastructure announcements arrived alongside a broader industry push to make agent systems interoperable across vendors and tools. (linuxfoundation.org) ### So what is Microsoft actually trying to own? Microsoft’s public materials point to a model in which infrastructure, governance and workflow software are bundled more tightly than raw model access. Azure Linux and container hosts cover the runtime layer; Microsoft 365 Copilot agents cover the work layer; Copilot+ PCs and Windows AI tooling cover the device layer. (linuxfoundation.org) The next concrete milestone is June 2, when Microsoft said the broader rollout of the Azure Linux 4.0 preview and Azure Container Linux announcements would be carried into Build. The Agentic AI Foundation, for its part, has scheduled additional 2026 events around AGNTCon, MCPCon and regional developer summits. (opensource.microsoft.com)