Stay connected with KayaToday, follow us on Instagram and Facebook for the latest news and reviews delivered straight to you.
The promise of AI that keeps working after you close your laptop has become a competitive battleground among the world’s largest technology companies. Microsoft stepped further into that fight on September 25, announcing a substantial redesign of its Copilot platform that introduces a persistent background agent, an in-product application builder with managed hosting, and tighter spending controls for enterprise buyers. The ambition is clear: make Copilot the operating layer through which employees delegate not just tasks but ongoing responsibilities.
What Microsoft actually demonstrated, however, and what remains aspirational, are two different things. Enterprises evaluating the announcement will need to hold both in view.
Autopilot Is the Agent That Does Not Clock Off
The most consequential piece of the redesign is Autopilot, which is the new public name for Scout, Microsoft’s earlier personal-agent initiative introduced in June. The rebrand is not merely cosmetic. Microsoft is positioning Autopilot as a system capable of holding a sustained assignment over days, not just responding to a single prompt.
Each Autopilot instance carries its own identity, memory, computing environment and workspace inside an organisation’s Microsoft 365 environment. It can monitor Teams conversations, track email threads, pull inventory data from Dynamics 365 and surface updates to colleagues without being prompted each time. Employees interact with it by mentioning it in documents or conversations, much as they would tag a human colleague.
Microsoft illustrated the concept through a retail scenario in which an Autopilot named Dot monitored Black Friday preparations, identified a shipment problem affecting 18 stores, surfaced the issue to relevant employees and reported that the group had resolved it. The demonstration is instructive about the intended workflow. It does not, as Microsoft’s own announcement materials acknowledge implicitly, establish how reliably Autopilot handles comparable situations across real customer deployments at scale.
Governance sits with Agent 365, which provides permissions, audit records and administrative controls around what Autopilot can access and act upon. That layer matters considerably when software begins contacting people or modifying business data. The launch documentation does not, however, spell out a complete action-by-action approval policy or explain how administrators should handle every category of failed execution, which leaves a meaningful gap for enterprise risk teams to fill in themselves.
Speaking during a September 23 presentation provided to VentureBeat ahead of the announcement, Microsoft CEO Satya Nadella framed the governance question directly: “Every agent has to have an identity. Everything it does needs to be observed.” The principle is sound. The implementation details will determine whether it holds in practice.
Code and Home Round Out a Three-Destination Redesign
Alongside Autopilot, Microsoft reorganises Copilot into three destinations. Home combines conversational chat with Cowork, the longer-form collaborative assistant Microsoft introduced in June. A feature called Today, planned for October private preview, will aggregate signals from email, calendars and task lists to surface follow-up drafts or calendar adjustments for users to review. Office in Copilot embeds Word, PowerPoint and Excel editing directly into the Copilot interface, keeping files synchronised with colleagues working in the standalone Office applications.
Code is the third destination, aimed at employees who want a custom tool without involving a developer. Users describe an application, dashboard or automation in plain language and Copilot generates the implementation. Microsoft says Code shares foundational technology with GitHub Copilot and draws on organisational context and connected services. Professional developers can continue using GitHub Copilot while business users build through the Copilot interface.
The hosting infrastructure behind Code is Copilot Managed Runtime, a Microsoft-operated platform that keeps generated applications within the organisation’s Microsoft 365 boundary. Earlier iterations of App Builder, introduced in October 2025 for Frontier participants, already used Power Platform infrastructure for hosting. The new runtime extends that approach with editable code, Git version management and an SDK that opens the same deployment path to compatible third-party development tools.
Jacob Andreou, Microsoft’s executive vice president for Copilot, described the intended simplicity during the presentation: “What this means at its core is hosting and sharing an application literally becomes as simple as saving and sharing a Word document.” That framing captures the aspiration well. Whether the runtime delivers that experience consistently for non-technical users building genuinely complex tools is a question that customer deployments will answer over the coming months.
Microsoft is not alone in this direction. On September 16, Anthropic announced the combination of Claude chat and its own Cowork equivalent, alongside document and slide tools, with the ability to continue delegated tasks after a user closes a laptop. Microsoft’s proposed advantage centres on connecting those interactions to Office files, business systems and administrative policies that enterprise customers already use, rather than requiring them to integrate a new platform from scratch.
The Pricing Picture Is Incomplete, and That Matters
For enterprise buyers, the economics of more autonomous AI are as important as the capabilities themselves. Microsoft’s announcement extends its existing credit-based billing model across the redesigned Copilot, but leaves key usage rates unspecified for Code and Autopilot.
The base enterprise subscription remains at $30 per user per month, billed annually. Everyday work stays within that subscription. Cowork, Code and Autopilot each require the subscription plus additional Copilot Credits charged by consumption. Microsoft’s own modelling illustrates how quickly the bill can grow: a workload of 15 everyday tasks fits within the $30 subscription, but 20 everyday tasks plus five complex Cowork assignments produces a modelled total of $73 with Opus 5 or $69 with GPT-5.6 Sol, including between $39 and $43 in consumption charges on top of the base fee.
Those figures describe selected workloads constructed by Microsoft for illustration. They are not new subscription tiers and should not be read as representative bills for typical enterprise users. Microsoft tested 40 prompts 10 times each, and its rival comparisons use public subscription prices and API rates with different model and reasoning settings rather than verified customer bills.
Earlier Cowork credit estimates provide some context. Microsoft’s June guidance put light tasks at $0.70 to $2, medium tasks at $4 to $6 and heavy tasks above $15 at pay-as-you-go rates. Annual credit commitments offer discounts of 5 to 20 percent, with unused balances expiring at the term’s end. No equivalent estimates exist yet for Code or Autopilot, meaning organisations cannot fully model their exposure before committing.
The FinOps additions Microsoft announced extend spending-management coverage to Autopilot, Code and the managed runtime, with Copilot Studio agents scheduled to follow in October. Administrators can allocate budgets, impose limits and receive alerts, and metered services remain disabled until a spending policy is created. Group-specific model access, beginning with Cowork, lets administrators restrict which model families are available, which also limits the choices available to automatic model routing. Employees gain visibility into their consumed credits and remaining allowance.
Availability is staggered. Home and Office integration begin reaching Frontier participants over the coming weeks. Code is scheduled for Frontier at September’s end, followed by wider access, with a preview for Microsoft 365 Premium or Pro subscribers planned for later in 2026. Autopilot’s private preview expands at month-end, while the managed runtime and SDK enter public preview on the same timeline.
Why the Next Few Months Will Be the Real Test
Microsoft’s redesign represents a genuine strategic shift rather than a cosmetic refresh. Bundling persistent agents, application creation and managed hosting into a single enterprise platform raises the ceiling on what Copilot can do, and raises the stakes for organisations that choose to delegate substantive work to it.
For businesses in Malaysia and Singapore evaluating enterprise AI platforms, the announcement adds urgency to questions that regulators including the Monetary Authority of Singapore and Malaysia’s Securities Commission have already begun framing around AI governance: who is accountable when an autonomous agent acts on business data, how audit trails are maintained and how spending on AI services is controlled and reported. Microsoft’s Agent 365 governance layer addresses some of those concerns in principle, but the gaps in its published approval policies mean that enterprise risk and compliance teams will need to do significant work before deploying Autopilot in regulated environments.
The competitive pressure from Anthropic and others means Microsoft cannot afford to let the gap between announcement and reliable delivery stretch too long. The product roadmap is substantial. The missing usage rates for Code and Autopilot, combined with the absence of detailed failure-handling policies, are the two areas where the announcement falls short of what enterprise buyers need to make informed commitments. Until those details arrive, the redesign is best understood as a credible and ambitious direction, one whose real value will only become clear once customers can match the autonomy Microsoft is offering with predictable costs and dependable results in their own environments.
Read More: Stanford and Nvidia’s CLM-8B Wants to Be the Fast, Cheap Decision Layer Inside Your AI Agent
