Microsoft just did to AI agents what Apple — as explored in the interface layer wars reshaping consumer tech — did to mobile apps in 2008. At Build 2026, Satya Nadella unveiled the Windows Agent Store — a curated marketplace where AI agents live as first-class citizens in the operating system, with an 85% revenue share for developers.
Alongside it: the Windows Agent Framework (WAF), open-sourced under MIT license, providing the runtime, registration, communication bus, and memory services that let agents operate natively inside Windows. Adobe and Zoom are already building agents for it.
Agents as First-Class Citizens in the OS
Nadella’s framing was explicit: “Windows is no longer a platform for human users only — agents are now first-class citizens in the runtime, the tooling, and the distribution model.”
What this means technically: Windows 11 26H2 “Helios” includes a native Agent Registration Service (a local daemon managing agent lifecycle, health checks, and versioning), a Cross-Agent Communication Bus built on gRPC, and a Memory Service for persistent context across sessions. Agents can appear in the taskbar, inside Copilot, in Teams, and in Outlook.
Adobe is building an InDesign agent that learns layout habits and automates repetitive design workflows. Zoom is building an agent that joins meetings autonomously, summarizes action items, and creates tasks in Planner. These aren’t chatbots in a sidebar. They’re persistent software entities that operate alongside — and sometimes instead of — the human user.
85% Revenue Share: The Platform Economics
The 85% developer revenue share is aggressive. Apple’s App Store takes 30%. Google Play takes 30%. Even Nvidia’s new agent marketplace hasn’t published economics yet. Microsoft is clearly using economics to win the developer land grab — establishing Windows as the default platform for agent deployment before Apple or Google can build their own equivalents.
The bet: if Windows becomes the place where agents live, Microsoft captures distribution even when it doesn’t own the model. A developer building an agent powered by Claude that runs on Windows and monetizes through the Agent Store still flows through Microsoft’s infrastructure — as explored in the economics of AI compute infrastructure — . The platform always wins.
AgentGuard: Enterprise Governance
The enterprise play is AgentGuard — a governance layer that gives IT admins role-based permissions, data loss prevention, and audit logging across all agent interactions. Rules like “an agent may never send financial data to an external API without explicit human approval” can be enforced at the OS level.
This is the feature that lets CIOs say yes. Without governance, agents in the enterprise are a security nightmare — autonomous software entities with access to email, calendars, documents, and APIs, operating without human oversight. AgentGuard makes the agent controllable enough for regulated industries to deploy.
The GitHub Copilot Endgame
Build also previewed GitHub Copilot’s next phase. Autonomous Agent Mode arrives in July for Enterprise users — agents that write, test, and commit entire feature branches without human intervention. Autopilot mode runs in the background handling dependency updates, test coverage gaps, and documentation drift. No developer present required.
The live demo was remarkable: a developer typed “Build a .NET MAUI cross-platform settings page with dark mode toggle and MQTT telemetry.” The agent generated a pull request with 12 files, passing unit tests, and an auto-recorded Loom walkthrough — in 10 minutes.
Microsoft is pricing all of this into a new M365 E7 + Agent 365 SKU at $99/user/month — bundling E5, Copilot, Entra Suite, and the full agent platform. At 400 million M365 users, even 10% adoption of E7 means $4.8 billion in new annual revenue.
The agent economy just got its App Store moment. The question is whether developers will build for Windows first — or wait to see what Apple announces at WWDC on June 8.
For the full structural map of the AI economy, read The Map of AI Redrawn on Business Engineer.









