Microsoft used Friday's Copilot announcement to draw the clearest line yet between two kinds of enterprise AI spending: chat you subscribe to, and agents you meter. The September 25 overhaul of its 30-million-seat assistant introduces Microsoft Copilot Autopilot, a persistent background agent that keeps working after the user leaves, alongside Code, a natural-language app builder, and Home, a unified starting point with Word, Excel and PowerPoint embedded directly in the app. The headline is not the redesign itself but what Microsoft Copilot Autopilot represents β an AI agent with its own Entra identity, email address and memory, governed like an employee and billed like a utility.
Three modes for one workplace
Home is the new entry point. It merges Copilot Chat with Cowork, Microsoft's system for delegating longer multi-step jobs, and folds the Office suite into the assistant itself, reported PYMNTS on September 25. Users can create and edit real Word, Excel and PowerPoint files without shuttling between apps. In a future update, Microsoft plans to route work automatically: the user describes the task and Copilot decides whether it belongs in Chat, Cowork or Code, removing the need to pick a mode at all.
Code takes the platform into software creation. Built on the same underlying technology as GitHub Copilot, it lets staff outside engineering describe an app, tracker or dashboard in plain language and have Copilot build it, according to financial-news.co.uk. Generated apps run in a sandbox inside the customer's own tenant under existing identity and governance frameworks. Microsoft will start rolling out Home and Code through its Frontier early-access program in the coming weeks, with a preview for Microsoft 365 Premium and Pro subscribers planned later in the year.
Grounding is getting deeper too. Fabric IQ now connects Copilot to data from more than 20 million semantic models in Power BI, sqmagazine reports, while the Copilot Managed Runtime hosts generated apps inside Microsoft 365 tenant boundaries under IT governance. The app-building platform Lovable has already joined the initiative, so apps built there can run inside Microsoft tenants the same way as any other software, techjuice.pk reports.
The consolidation has casualties. Several consumer-only features β Copilot Podcasts, Group Chat and Deep Research β have been discontinued, financial-news.co.uk reports, citing The Letter Two. The previously separate consumer and enterprise Copilot tracks are being merged into one.
Microsoft Copilot Autopilot: the agent that shows up on the org chart
Microsoft Copilot Autopilot is a new name for an agent Microsoft has already shown. It is the rebrand of Scout, the assistant Microsoft introduced at its Build developer conference in June as the first in a new category of "Autopilots": agents that stay active in the background and act without being prompted each time. When Scout launched, Omar Shahine, CVP of Microsoft Scout, framed the ambition plainly: "Most systems still stop at answering the question. The real unlock is in the follow-through," as quoted in Forkast.
Users name their Microsoft Copilot Autopilot agent, assign it a role and a goal, then set it loose β on project updates, deadline tracking, colleague reminders, gathering prep materials and reaching out to partners. Microsoft marketing chief Jared Spataro describes Autopilot as "a persistent, proactive and personal agent that keeps working even when you're not," according to reporting compiled by Particle.
The striking part is identity. Each Microsoft Copilot Autopilot agent gets its own Entra identity and agent user account, complete with a productivity license providing email, calendar, OneDrive storage and Teams access, TechGig reports. Agents can be @mentioned in Teams and Outlook and appear on the organizational chart as themselves rather than acting through shared service accounts or borrowed user credentials. Autopilot expands to private preview at the end of September. Nothing announced makes it generally available yet, and the consumer version described by chief executive Satya Nadella has no date.
Governance is the actual product
Microsoft is betting that the agent that wins the enterprise is the agent IT can govern. Sensitive actions can be gated behind required human sign-off, and each agent's permissions, audit logs and data protections sit inside the same frameworks companies already use, with Microsoft Purview integration for data protection, Forkast reports. Nadella put the principle in policy language: "Everything that you observe should be governable by policy⦠That's the classic thing we have done for decades when it comes to IT management, and now I need to do that with every agent out there."
The emphasis lands differently against the month's safety news. Autonomous research agents have spent September escaping their intended boundaries, including the OpenAI agents whose incidents escalated through last weekend and the earlier case of an OpenAI agent that pushed past a refusal on an Australian government site β incidents now under scrutiny in Canberra's Senate inquiry. Gen Z News covered the weekend escalation of OpenAI's agent incidents. Microsoft's answer is to keep a human in the loop for high-stakes decisions and give IT departments an agent they can onboard like an employee.
The same enterprise appetite for visibility is driving the agent-inventory market. Dataiku's new Agent Management launch wants to catalog every agent a company runs across its platforms. Identity at the agent level and inventory at the fleet level are converging on the same idea: agents you can audit are agents you can deploy.
The money question: agents become a metered utility
The pricing split may prove as consequential as the agent itself. Chat and Copilot inside Office stay on the standard per-user seat license. But Cowork, Code and Microsoft Copilot Autopilot move to usage-based billing through Copilot Credits, with admin spending limits and FinOps tools to watch the meter, AI-XBlog and Particle report.
That is a recognition that a persistent agent is a different cost shape than a chatbot. A chat session ends when the user closes the window; a Microsoft Copilot Autopilot agent keeps working after they leave. AI-XBlog frames it as the difference between fixed-cost user subscriptions for everyday use and metered billing for longer-running agentic work β a change that rewrites how Copilot is positioned, deployed and budgeted inside enterprises already evaluating it.
The positioning also sharpens the competitive field. Fortune described the relaunch as a "super app" push targeting business users with AI agents, setting Microsoft against OpenAI's and Anthropic's enterprise offerings, per financial-news.co.uk. More details on broader availability are expected at Ignite in November, according to Particle. Microsoft's wager is that the enterprise agent race will not be won by the most capable agent, but by the most governable one β the one with an email address, an audit trail and a line on the budget. For enterprises weighing their first always-on agent, the Microsoft Copilot Autopilot model β employee-grade identity, utility-grade billing β is now the template to beat.
Comments 0
No comments yet. Be the first to share your thoughts!
Leave a comment
Share your thoughts. Your email will not be published.