Microsoft Copilot Super App: Code, Autopilot, and What It Means
Microsoft on September 25, 2026 unveiled the most significant redesign of its Copilot platform since the product launched. The new Copilot is not an incremental update: it is Microsoft’s attempt to build a single enterprise AI workspace that combines conversational AI, citizen-developer coding, and persistent autonomous agents under one interface, with a governance model designed specifically for large organizations. The company is explicitly positioning it as the successor to Office for the AI era, and the comparison signals how seriously it is treating the bet.
Here is what the new Copilot is, why it matters, what it is competing against, and what enterprise teams need to decide.
What Microsoft Shipped Today
The new Copilot is organized around three tabs: Home, Code, and Autopilot.
Home: The Unified Entry Point
Home merges Copilot Chat and Cowork, Microsoft’s existing background agent, into a single starting screen. More importantly, it embeds Office directly into Copilot: Word, Excel, PowerPoint, Teams, Outlook, and calendar are now accessible without leaving the app. A feature called Today, arriving in private preview in October, will surface prioritized emails, meeting requests, overdue commitments, and draft responses proactively, functioning as an AI-managed inbox and task queue.
Home is powered by Microsoft IQ, the company’s unified organizational intelligence layer. Fabric IQ connects 20 million Power BI semantic models to Chat and Cowork, so when an employee asks for a revenue summary, Copilot draws on live business data rather than a cached document. Dynamics 365 and Power Platform context are rolling out through Home in public preview over the next month.
Code: Natural Language Development for Everyone
Code is the most strategically interesting addition. It lets any employee, not just engineers, build small applications, dashboards, trackers, and automations by describing them in plain language. The underlying technology is GitHub Copilot’s stack, running in a sandboxed environment. Apps built in Code can be hosted inside a company’s Microsoft 365 tenant through the new Copilot Managed Runtime, with IT managing access, data connections, and deployment.
Microsoft’s Chief Marketing Officer for AI at Work, Jared Spataro, framed Code directly against ChatGPT’s Codex and Claude Code: “Vibe coding and building small apps is becoming as basic a skill as writing a memo or modeling a budget.” He acknowledged to CNBC that Code is Microsoft catching up on a category where OpenAI and Anthropic have already established users.
The practical implication is significant. Enterprises that today require a developer ticket for any internal tooling now have a path to let operations, finance, and sales teams build lightweight workflow tools themselves, with the security and governance controls that prevent shadow IT: sandboxed execution, tenant-hosted runtime, audit logs, and IT-governed plugin registries.
Autopilot: The Persistent Personal Agent
Autopilot is a renamed and elevated version of Scout, a personal AI assistant Microsoft previewed earlier in 2026. It runs as a continuous background agent inside a company’s Microsoft 365 tenant. Each Autopilot instance gets its own identity, memory, and cloud compute environment. Employees can name it, assign it a role, and task it with recurring work: monitoring Teams channels, generating weekly status reports, handling follow-ups, building supplier review schedules, or reaching out to stakeholders on behalf of the employee.
The enterprise-grade framing is deliberate. Autopilot requires explicit permissions, maintains a full audit log, and integrates with Microsoft’s identity governance. It surfaces where employees already work: Teams channels, Outlook, document threads. The @mention model makes it feel like adding a teammate rather than configuring a bot.
Autopilot enters private preview at the end of September. Broad rollout depends on early tester feedback.
The Pricing Shift: From Seats to Usage
The new Copilot introduces a material change to how Microsoft charges for AI at work. The existing $30/user/month Microsoft 365 Copilot license covers the user subscription layer: Chat, Office integration, and standard model routing. Starting with the new Copilot, Code, Autopilot, Cowork, and access to frontier models (Astra, Fable) move to usage-based billing. Enterprises pay for actual agentic compute consumed, not a flat per-seat fee.
Microsoft describes this as matching cost to value: “everyday AI” at a fixed subscription, “agentic work” at consumption pricing. A new FinOps for AI suite helps IT admins set spending policies by group, route budget requests through approval workflows, and monitor which Cowork tasks are generating business outcomes versus burning credits. Cost management coverage is expanding from Cowork and Work IQ to include Code and Copilot Managed Runtime, with Copilot Studio agent support planned for October.
For enterprise finance leaders, the shift has two implications. Predictable operational overhead moves to variable cost tied to agent activity, which is harder to budget but more accurately reflects actual value delivered. And the FinOps tooling Microsoft is building suggests it understands the risk: organizations that deployed ChatGPT Work or Cowork without usage controls have encountered surprise spend. Microsoft is making governance the selling point.
The Competitive Context
Microsoft’s new Copilot competes directly on three fronts.
Against Anthropic’s Claude. CNBC headlined its coverage “Microsoft packages business AI in single app as it tries to compete with Anthropic,” reflecting the narrative that Microsoft has lost ground to Claude in enterprise AI adoption. Anthropic’s Claude Cowork and mobile agents, launched in July 2026, have attracted enterprise customers who value stronger reasoning, broad MCP tool integration, and Anthropic’s Enterprise API. Microsoft’s counter is native Microsoft 365 integration, existing identity governance, and a zero-switching-cost path for Office-heavy organizations.
Against OpenAI’s ChatGPT Work and Codex. ChatGPT Work, launched in July 2026, gave OpenAI a direct enterprise agent product. Codex, OpenAI’s coding agent, is the explicit benchmark for Code. Microsoft benefits from OpenAI models (Astra, Sol, Luna are available in Copilot via Microsoft Foundry) while building its own surface area so that the interface relationship is Microsoft’s, not OpenAI’s.
Against its own fragmentation. The most underreported competition is internal. For several years, Microsoft offered Copilot in too many flavors: Copilot for Microsoft 365, GitHub Copilot, Copilot Studio, Azure AI Foundry, Windows Copilot, and various embedded versions. Customers consistently told analysts they were confused about which product to buy and how they related. The super app is a consolidation bet: one Copilot brand, three clear tabs, one pricing model.
What Has Not Changed
Several important limitations remain in the September 25 launch:
No general availability date. Home and Code launch to the Frontier early-access program “in the coming weeks.” Autopilot goes to private preview. Microsoft 365 Premium and Pro subscribers get Code in preview later in 2026. Broader GA depends on early program outcomes.
Model availability is not uniform. Inside the Home tab, Microsoft has no internal model available: it uses GPT-6 family models and auto-routing. Inside Code, users can manually select models. Anthropic’s Fable is available on usage-based billing. This means enterprises choosing Copilot are still highly dependent on OpenAI model supply.
Autopilot scope is still constrained. Like all persistent agents at this stage, Autopilot is best suited for well-structured recurring tasks. Complex multi-system workflows, judgment-heavy decisions, and tasks requiring real-time awareness of external data are still edge cases for production. Microsoft’s own messaging emphasizes project updates, schedule management, and meeting follow-ups: clearly defined, bounded tasks.
A Comparison Framework for Enterprise Teams
| Dimension | Microsoft Copilot (new) | Anthropic Claude Cowork | OpenAI ChatGPT Work |
|---|---|---|---|
| Integration depth | Native Microsoft 365 and Azure | MCP tool ecosystem, broad connectors | OpenAI apps, some third-party |
| Coding capability | Code tab, GitHub Copilot stack | Claude Code (API) | Codex agent |
| Persistent agent | Autopilot (preview) | Cowork (GA) | Background agents (GA) |
| Pricing model | USL + UBB for agentic | Usage-based | Subscription + usage |
| Identity governance | Azure AD, Microsoft 365 native | API-level, enterprise tier | OpenAI Enterprise |
| Frontier model access | GPT-6 Astra, Sol, Luna, Fable (UBB) | Claude Opus 5.5, Fable 5.1 | GPT-6 Astra, Sol, Luna |
| Rollout status | Early access, no GA date | Generally available | Generally available |
The selection principle for most organizations is simple: if your identity estate, productivity tools, and data infrastructure live in Microsoft 365 and Azure, Copilot’s native integration is a structural advantage no competing product can replicate today. The governance story (audit logs, tenant-hosted runtime, IT spending controls) is designed for exactly the compliance posture large enterprises require.
If your team runs on a more heterogeneous stack, or if model flexibility and third-party tool connectivity are primary requirements, Anthropic’s Claude and OpenAI’s products offer features at general availability today that Microsoft’s new Copilot is still rolling out.
Why This Launch Matters Beyond the Features
Microsoft’s new Copilot is significant beyond its feature set for two structural reasons.
First, it signals the end of the AI features-in-apps model. Microsoft tried for two years to sell Copilot as an embedded feature inside Word, Teams, and Excel. The super app is an admission that users want a unified AI workspace, not AI scattered across 20 products. That validates the bet OpenAI made with ChatGPT Work and Anthropic with Cowork: the interaction layer for enterprise AI is its own application category.
Second, the usage-based pricing shift normalizes outcome-tied AI spend in enterprise software. When agentic work is billed by consumption, customers have a direct incentive to measure what their agents actually accomplish. That creates demand for the ROI measurement frameworks enterprises have been building since 2025: you cannot justify usage-based spend without tracking business outcomes per AI action. Expect FinOps for AI to become a meaningful budget line for enterprise IT in 2027.
Internal Links
For enterprise teams building governance around agentic deployments, see Claude Cowork and ChatGPT Work: Two Models for Enterprise AI Workspaces for a detailed comparison of the competing workspace approaches. The OpenAI Agents API and Managed Codex post covers OpenAI’s enterprise agent infrastructure in depth. For Enera’s own take on what enterprise AI adoption readiness requires, book a call with the team.
Sources: Microsoft Official Blog, Jared Spataro; The Verge; Fortune; CNBC; Reuters; Tom’s Hardware