Microsoft Rebuilds Copilot as an Enterprise AI Super App

Microsoft is rebuilding Copilot around a broader enterprise bet: one product that can write code, handle productivity work, and delegate multi-step tasks to agents. The company introduced the updated app on Friday as it tries to turn its Office footprint into a stronger distribution advantage against Anthropic and other agent platforms, according to CNBC.

One interface for code and action

The redesign combines the Home, Code, and Autopilot experiences. Code is aimed at making app and dashboard creation more accessible to non-software engineers, while Autopilot lets users create custom agents that work across Teams and Outlook. The product is meant to reduce the fragmentation created by Microsoft’s separate 365 Copilot and GitHub Copilot offerings.

Fortune reports that the super app also connects to Office workflows, allowing users to ask an agent to work with Excel spreadsheets or change a Word document. Autopilot agents can be assigned a role and goal, then used for project updates, deadline tracking, preparation work, and partner outreach. The features are rolling out through testing rather than a broad general release, with Autopilot scheduled for private preview by the end of the month.

Usage-based agents change the buying model

Microsoft says customers will pay for the Code and Autopilot capabilities based on usage, rather than treating every advanced action as part of a standard per-seat subscription. That is a meaningful shift for enterprise buyers. It ties the cost of the product to how much work agents perform, but it also makes permissioning, budgets, and observability part of procurement rather than an afterthought.

Microsoft still has a distribution problem to solve. CNBC reports that fewer than 7% of its more than 450 million commercial Office 365 seats have licenses for the AI addition. The new packaging is an attempt to make Copilot feel less like an optional assistant and more like an execution layer inside tools companies already pay for.

Builder impact

For AI builders, Microsoft’s move reinforces a pattern already visible in Google’s private AI infrastructure work: the product is becoming the governed system around the model. Agents need scoped permissions, audit logs, isolated execution, model routing, and clear human control. The model is only one component.

The practical lesson is to design agent workflows around explicit capabilities, not a single giant assistant. Separate read and write access, meter expensive actions, preserve approval checkpoints, and make every external side effect traceable. Microsoft is betting that enterprise AI wins when coding, context, and action live in one place. Builders will need to prove that this convenience does not erase the controls that make automation safe to deploy.