Looking at the new Copilot "Super App"

Summary

The presenter reviews Microsoft’s new Copilot “Super App” after several weeks of trying it, focusing on its capabilities and implications for organizations. His central argument is that AI has moved from helping with individual tasks toward handling delegated, multi-step outcomes—and that Copilot’s new structure aims to make the right level of AI available for different kinds of work.

Main features and concepts

• Home and chat: Home brings together chat and collaboration features. Chat is presented as the interactive, lower-latency starting point for everyday work such as drafting, summarizing, briefing, answering questions, and brainstorming. Users can type, dictate, or use voice mode.
• Enterprise knowledge and controls: Copilot can draw on organizational content such as email, Teams conversations and meetings, business data, and connected resources. The presenter describes this as relying on organizational “Work IQ” and business-data capabilities. He says access remains subject to existing permissions, classifications, data-loss-prevention policies, and security controls, with newly created content inheriting relevant classifications.
• Models and reasoning: Copilot’s automatic mode selects models and reasoning levels based on a request’s complexity, context, accuracy needs, and speed. Users can choose a specific model when appropriate. The presenter emphasizes that different models have different strengths rather than one being best for every task, and says the product has improved substantially compared with a year earlier.
• Agents, plugins, and connectors: Users can invoke Word, Excel, and PowerPoint capabilities from Copilot, as well as use agents such as Analyst, Researcher, and a meeting facilitator. Organizations can also create and share agents. The described plugin ecosystem includes:
  • Skills: Instructions that guide an agent through a task, with details surfaced as needed.
  • Connectors and MCP servers: Ways to access knowledge or take actions in other systems.
  • A plugin registry: A proposed central place for organizations to discover and govern plugins.
• Notebooks: Notebooks bring together selected Office and web materials from OneDrive or SharePoint, along with user instructions. Copilot can be constrained to use only the notebook’s reference materials, supporting focused work and collaboration.
• Cowork: For complex, multi-step deliverables, users can switch from chat to Cowork. It is described as using richer task planning and potentially more advanced reasoning to complete work that may take considerable time. The presenter recommends starting in chat and using Cowork when the task is clearly larger or chat does not produce the needed result.
• Code: The Code tab helps users turn ideas into applications, trackers, or dashboards. Projects can run locally, with files stored in OneDrive, or in Microsoft’s managed cloud runtime. The cloud option is described as integrating with Entra, supporting access policies, version control through Git, and app hosting for selected users or an organization. Projects can also be connected to or moved to GitHub. The presenter says these apps are managed separately from agents and can be administered through Microsoft 365 tools.
• Autopilot: Presented as an always-available digital assistant for delegated or recurring tasks—for example, identifying follow-ups in messages and meetings, preparing expense reports, resolving calendar conflicts, optimizing schedules, or monitoring news about specified clients.
• Intent routing: Microsoft is working toward having Copilot infer whether a request is best handled by chat, Cowork, or Code, rather than requiring users to choose a mode themselves. The presenter says this could reduce confusion, though routing complex work to a usage-based mode would need to account for users’ quotas and costs.

Pricing, governance, and organizational considerations

• User Seat License (USL): Chat and the associated capabilities are described as having a fixed per-user, per-month price, making costs more predictable. The presenter expects chat to cover much of many employees’ day-to-day use.
• User-Based Billing (UBB): Cowork and other compute-intensive work are described as pay-per-use, based on the work performed. This offers access to more involved tasks but makes cost monitoring and controls important.
• FinOps controls: Administrators are expected to have tenant-, group-, and user-level spending policies; controls over available model families; usage and token reports; quota notifications; and programmatic management through Graph API. The presenter highlights measuring business outcomes—such as work completed and time freed up—as important for judging the value of usage-based AI.
• Organizational opportunity: The presenter sees Code as a way for employees without traditional programming access to prototype ideas. He also stresses that delegated work still needs appropriate human oversight, especially when tasks are complex or consequential.

Review and guidance

This is a hands-on overview and analysis rather than a step-by-step tutorial. The presenter’s practical recommendation is to use chat for most interactive work, move to Cowork for substantial multi-step deliverables, and use Code when the goal is to create an application or dashboard. He notes that availability depends on rollout status and the organization’s deployment wave.

Main speaker and sources

• Main speaker: John Savill, presenting his own observations and organizational analysis.
• Other sources referenced: Microsoft blogs and product information about the new Copilot, plugin registry, AI FinOps, Copilot pricing, measuring Cowork’s value, and Copilot-managed application hosting.