OpenAI, Google, and Anthropic Are Building a FINRA for AI. History Says Self-Regulatory Bodies Don't Stop the Industries That Fund Them.

On September 25, 2026, Microsoft unveiled what it is calling "a new OS for work" — a redesigned Copilot product that consolidates the company's previously fragmented AI product surface into a single unified application. The announcement introduced three new sections: Home, which brings Chat, Cowork, and the full Office suite (Word, Excel, PowerPoint) into a single entry point; Code, which lets enterprise users build custom apps, widgets, and automations through natural language without leaving the Copilot environment; and Autopilot, a persistent background agent that operates within enterprise tenant governance, security permissions, and audit controls.

The strategic framing — "a new OS for work" — is the key signal. Microsoft is not positioning Copilot as an AI assistant that helps users do work inside existing applications. It is positioning Copilot as the primary surface from which all work happens, with Office applications and AI capabilities organized underneath it. The application layer becomes subordinate to the AI layer.

The pricing structure confirms the architectural bet: a flat per-user license covers basic Chat and Office features, while Cowork, Code, and Autopilot are metered by usage. For enterprise accounts, Microsoft announced volume discounts of 30% for deployments between 1,000 and 10,000 seats, rising to 50% for commitments above 10,000 seats.

Understanding what this means for enterprise AI strategy requires unpacking each of the three new sections, the pricing model they sit inside, and the competitive dynamics the super app is designed to address.

What Home, Code, and Autopilot Actually Do

Home is the consolidation play Microsoft has been building toward since Teams was bundled into M365 in 2017. It merges the chat interface, the M365 Copilot assistant, and native Office application access — Word, Excel, PowerPoint — into a single experience. An enterprise knowledge worker using Copilot Home can move from a structured document in Word to a conversational research session to a data analysis task in Excel without switching applications or losing context. The conversation history carries across the session, so an Excel pivot table can reference the Word document without manual copy-and-paste.

The "Office in Copilot" framing is a reversal of the original M365 Copilot positioning, where Copilot was the add-on to Office. Home makes Office the add-on to Copilot. Satya Nadella has described this as the completion of the shift from document-centric to conversation-centric work: the conversation is the primary artifact; the document is the output.

Code is the non-developer coding assistant, built on GitHub Copilot's underlying technology but designed for workers who have never written a line of production code. Using natural language descriptions, Code generates apps, desktop widgets, dashboards, trackers, and automations that run inside a sandboxed environment within the organization's tenant. A sales operations manager can describe the pipeline analysis tool they need, Code builds it, it runs inside the enterprise's security perimeter, and it can be shared with the team from within Copilot.

This matters for two reasons. First, it democratizes custom software development: the organizational capacity to build tools is no longer limited by engineering team bandwidth. Second, it puts Microsoft in direct competition with no-code platforms (Airtable, Notion, Zapier) and low-code development tools (Power Apps, Bubble) by embedding similar functionality into the productivity layer that enterprise users already work inside.

Autopilot is the long-horizon agent — the feature Microsoft previewed at Q4 FY2026 earnings when Nadella described the super app concept. Autopilot handles multi-step, multi-hour tasks that span meetings, documents, emails, and external data sources without requiring the user to remain active in the conversation. It operates inside enterprise tenant governance — the same permission structure, security policies, and audit trails that govern user activity apply to Autopilot's activity.

The enterprise governance framing is deliberate: it directly addresses the concern that autonomous AI agents create audit gaps and compliance risks that enterprise information security teams cannot accept. Autopilot's tenant-native design means its activity appears in the same security information and event management (SIEM) systems and audit logs as user activity, making it subject to existing compliance frameworks rather than requiring new governance infrastructure.

The Pricing Architecture Is the Strategy

The commercial model for the new Copilot has two tiers:

• Flat per-user license: Covers Chat, Cowork context management, and Office integration (Word, Excel, PowerPoint, Outlook, Teams). This is effectively the current M365 Copilot license at $30/user/month, restructured into the new unified app.
• Usage-based metering: Covers Code execution, Autopilot task runs, and advanced Cowork operations. Billed per task, per API call, or per agent step depending on the operation type.

This architecture is a direct response to the activation problem that has plagued M365 Copilot since its 2023 launch. Signal documented Microsoft Copilot's activation gap: fewer than 15-25% of licensed M365 Copilot seats are actively used weekly, making Copilot simultaneously the most purchased AI product in enterprise history and one of the least utilized. The flat $30/user/month model created a situation where Microsoft was capturing maximum seat revenue while most enterprise users never engaged meaningfully with the product.

Usage-based metering changes the commercial dynamic in a structurally important way. Heavy Autopilot users generate consumption revenue; light users generate base seat revenue. The model creates a financial incentive for Microsoft to drive actual workflow adoption — revenue from Code and Autopilot only accrues when users are actively running tasks. This aligns Microsoft's revenue with user engagement in a way the pure flat subscription model did not.

The broader shift from seat-based to usage-based SaaS pricing that Signal has tracked across enterprise software confirms the pattern: companies that successfully couple a flat base license with metered consumption on high-value features consistently generate higher net revenue retention than pure flat-fee models, because expansion revenue tracks product engagement rather than contract renewal cycles.

The Enterprise Discount Signal: What 30% and 50% Mean

Microsoft's announced volume discounts — 30% for 1,000-10,000 seats, 50% for 10,000+ seats — confirm several things about its competitive posture:

The large enterprise market is the primary target. Discounts at this tier only make commercial sense if the large enterprise deals are either currently underpenetrated (big wins available) or at risk from competitive displacement (retention is the priority). The 50% discount for 10,000+ seat commitments is a substantial concession — it signals Microsoft is willing to take significant margin compression to lock large enterprises into multi-year consumption commitments before those organizations standardize on a competing AI product.

Usage volume is expected to be high. A 50% discount on the base license only pencils out if Microsoft expects metered Autopilot and Code usage to generate consumption revenue that compensates for the discount. The commercial bet is that large enterprise customers will generate enough agent task runs per seat per month to make the deal profitable despite the discounted per-seat fee.

The competitive threat from standalone products is real. ChatGPT Work, Claude Enterprise, and Gemini for Workspace are all competing for the same enterprise AI wallet. The volume discount structure is designed to make switching-cost analysis favor Microsoft: an enterprise that commits 10,000 seats at a 50% discount and then builds Copilot-native workflows faces a meaningful switching cost to migrate to a competing platform.

The Activation Problem Is Not Solved by Architecture

The structural challenge Microsoft faces with the new Copilot is the same one it faced with M365 Copilot: feature availability does not create behavioral change. The activation research across enterprise AI products consistently shows that tools with high procurement penetration but low engagement fail for behavioral reasons, not capability reasons.

Enterprise knowledge workers have established work patterns built around email, documents, and meetings. Introducing a new AI interface — even one that embeds within the tools they already use — requires changes to those patterns that most knowledge workers will not make without explicit behavioral intervention. The issue is not that Copilot Home is unintuitive; it is that current work patterns are deeply habituated, and habit change requires deliberate incentive design, social reinforcement, and manager-level change management that Microsoft cannot provide directly.

The Code and Autopilot features add complexity to this picture. Code requires users to identify workflow problems that would benefit from a custom tool, conceptualize a solution in natural language, iterate on the output, and integrate the result into their working process — a higher-order cognitive workflow that demands more than "ask AI a question." Autopilot requires trust in an autonomous agent that operates asynchronously, which is a substantial behavioral and organizational shift from synchronous human-in-the-loop work.

The enterprises that will extract significant value from the Copilot super app are those that approach it with a structured activation program: pilot cohorts with specific use cases, manager-led behavior change initiatives, and measurement against productivity metrics. The enterprises that will see low ROI are those that extend their seat licenses to the new product and wait for adoption to emerge organically.

Code vs. GitHub Copilot: Microsoft Cannibalizing Itself

One of the structurally interesting questions the Copilot super app raises is the relationship between Code and GitHub Copilot. Both are built on Microsoft's underlying coding AI infrastructure; both allow users to build software through AI assistance; and both compete for developer mindshare and enterprise budget.

The distinction Microsoft has drawn is audience. GitHub Copilot is designed for professional software engineers — it integrates into development environments (VS Code, JetBrains, Neovim), handles complex multi-file code generation and refactoring, and assumes the user understands code. Code in the Copilot super app is designed for business users who want to build tools without writing code — it prioritizes simplicity, runs in a sandboxed tenant environment, and outputs apps that work rather than production-grade code.

In practice, this distinction will be difficult to maintain. A junior developer who can build a Copilot Code app in ten minutes may not see the value in the GitHub Copilot subscription that costs separate money. An operations analyst who discovers Copilot Code can build the reporting tool their engineering team never had bandwidth to build may start building tools with increasing complexity, encroaching on what was previously engineering territory.

Microsoft will watch the actual usage overlap carefully. If Copilot Code meaningfully cannibalizes GitHub Copilot seat counts, the pricing model may need to evolve — possibly through a combined offering or a clearer capability differentiation that prevents the products from competing for the same workflows.

What Standalone AI Products Must Do Now

The Copilot super app creates a specific competitive problem for Claude Enterprise, ChatGPT Work, and Gemini for Workspace: Microsoft is now offering an all-in-one AI work platform that competes on individual feature dimensions while also offering the integration advantage of living inside the Microsoft 365 ecosystem where most enterprise workflows already exist.

Claude's advantage in long-context document reasoning and coding quality remains real, but capability advantages are fragile in a market where Microsoft can bundle competitive capabilities into the daily workflow surface that enterprise users are already in. The question is not "is Claude better at X?" but "is Claude better at X by a margin large enough to justify the friction of switching out of Copilot Home to use Claude for that task?"

Claude Enterprise's distribution expansion through platforms like Salesforce Agentforce and the Claudeforce integration is the right strategic response: embed Claude's capabilities into the platforms users are already in rather than competing for users to change which application they open first. Microsoft is betting on "one app for work" being the winning surface; the counterstrategy is to be available inside every enterprise surface, including Microsoft's own, where model quality or specific capabilities make Claude the right choice for the task.

Four Enterprise Procurement Questions the Super App Raises

For enterprise procurement teams evaluating how to respond to the Copilot super app announcement, four questions determine the strategic response:

1. What is the current M365 Copilot activation rate? If fewer than 30% of licensed seats are actively used weekly, the incremental investment in the super app architecture will not solve the root problem. Address activation first; the new features are not a substitute for behavioral change management.

2. Which teams have the most to gain from Code? Copilot Code's highest-value use cases are in functions with high workflow complexity and limited engineering support: finance (custom modeling tools), operations (process automation), sales operations (custom reporting). Identify those functions before broad rollout to focus activation energy on the highest-ROI cohorts.

3. How does Autopilot interact with data governance policies? Autopilot operates inside tenant governance, but tenant governance policies vary widely across enterprises. Validate that Autopilot's activity scope, audit trail format, and permission boundary assumptions align with information security requirements before enabling broad access.

4. What is the multi-vendor AI strategy? Committing to the 10,000+ seat volume discount in exchange for 50% pricing is a meaningful lock-in decision. Enterprises that expect to run multi-model AI workflows — using Claude for document reasoning, Copilot for Office integration, specialized models for domain-specific tasks — should evaluate whether the volume discount's financial benefit outweighs the platform dependency risk.

What to Watch in Q4 2026 and Beyond

The Copilot super app is rolling out in phases. Home and Code enter the Frontier test program in the coming weeks; Code becomes broadly available for M365 Premium and Pro in preview later this year; Autopilot entered private preview at the end of September. Full production availability for all three features for all enterprise tiers has no announced date.

The signal to watch: Autopilot adoption metrics once private preview expands to a meaningful cohort. If Microsoft reports significant Autopilot task run volumes in Q2 FY2027 earnings — alongside consumption revenue data showing metered features are generating expansion revenue — the super app architecture is working as intended. If Q2 FY2027 shows flat Copilot engagement despite the new features, the activation problem has not been solved and Microsoft faces a harder path to demonstrating AI ROI for enterprise customers.

The broader market signal: how ChatGPT Work, Claude Enterprise, and Gemini for Workspace respond. OpenAI has the operator and agent infrastructure; Anthropic has the coding model quality advantage; Google has the Workspace integration depth. If any of them launches a credible "one app" response to the Copilot super app before Microsoft achieves broad availability of all three sections, the competitive window narrows considerably.

Takeaway: Microsoft's new Copilot — Home, Code, Autopilot — is the company's most coherent enterprise AI product vision since Copilot launched. The unified app architecture is strategically sound: it creates a single engagement surface that competes with standalone AI products on convenience while leveraging Microsoft's existing enterprise install base. The pricing structure — flat license for basics, usage-based metering for agents — aligns Microsoft's revenue with actual engagement in a way the original M365 Copilot model never did. The enterprise discount structure (50% for 10,000+ seats) signals that large enterprise commitments are the primary competitive battleground, and that Microsoft is willing to take margin compression to prevent platform standardization on competing AI products. The risk is the same as it has always been: enterprise AI adoption is a behavioral change problem, not a feature problem, and architecture alone does not change behavior. The enterprises that win with the new Copilot will be the ones that design activation programs alongside the technology decision, not the ones that upgrade their license and wait.