Microsoft prepares a Copilot super app powered by agents

Microsoft wants to make Copilot the entry point to its AI

Microsoft is preparing a new step for Copilot: a “super app” expected before the end of the year, designed to bring together AI conversation, programming assistance and agentic capabilities within a single interface. The information, confirmed by Microsoft to The Verge, signals less the launch of an isolated feature than a shift in positioning. Copilot would no longer be merely a brand applied to several of the publisher's products: it would become a unified gateway to artificial intelligence services intended for both personal and professional use.

The term “super app” should be handled with caution. Microsoft has not yet provided a detailed product sheet, nor precisely described all the applications, models, subscriptions or services that would be brought together in this future experience. The company has also not detailed the technical architecture of the whole. But the scope mentioned by The Verge is clear: chat, code and agents are meant to coexist. This convergence responds to a logic that has become central in the industry: users do not necessarily want to choose between a general-purpose chatbot, a desktop assistant and a specialized development tool. They increasingly expect an AI capable of retaining context, moving from one task to another and, within certain limits, taking action.

For Microsoft, the stakes are considerable. The group has already deployed Copilot in Windows, Bing, Edge, Microsoft 365, GitHub and various offerings aimed at businesses. This rollout has established the brand, but it has also created a fragmented experience. Depending on the product used, Copilot does not refer to the same assistant, does not rely on the same data, does not have the same permissions and does not follow the same business model. A super app could give this whole ecosystem more visible coherence, particularly in the face of ChatGPT, whose strength lies precisely in a main interface that has become familiar to a very broad section of the public.

The project comes as the battle over the AI interface intensifies. At the beginning of the generative cycle, text chatbots served as a technology demonstration. Since then, assistants have become multimodal, integrated into work tools, gained access to files and offered search, writing, analysis and programming functions. The next competition concerns the ability to bring these uses together without forcing users to navigate among multiple services. Microsoft clearly wants Copilot to be one of these centers of gravity.

From a scattered brand to a shared experience

The difficulty is as much organizational as it is software-related. Microsoft sells products aimed at very different audiences: individuals on Windows, Microsoft 365 users, developers on GitHub, IT administrators, large enterprises and Azure cloud customers. Each of these worlds has its own interfaces, security policies and sales cycles. Bringing chat, code and agents together in a super app therefore does not necessarily mean immediately merging them into a single indistinct product. It may involve a shared layer, a central account, a shared conversation space or a delegation mechanism between services.

It is nevertheless an important change in the way AI is presented. Until now, Microsoft has often highlighted the integration of Copilot into existing software: writing in Word, analyzing in Excel, preparing a presentation in PowerPoint, summarizing meetings in Teams or completing code in a development environment. The super app introduces the reverse logic: start in Copilot, then connect the assistant to the tools needed to complete a task. The center of the experience would no longer be solely the productivity application, but the conversational agent orchestrating the applications.

This ambition recalls the role Microsoft has long sought to play with Windows: being the platform from which software, documents, communications and services are launched. In the era of generative AI, the question is no longer merely which operating system or office suite is installed on a workstation. It is which assistant receives the initial request, knows the user's preferences and becomes the primary channel for accessing information and digital tools.

A project rooted in Copilot's recent history

Copilot's trajectory has been rapid. Microsoft first popularized the name in the development world with GitHub Copilot, unveiled in public preview in June 2021. GitHub, acquired by Microsoft in 2018 for $7.5 billion, had then partnered with OpenAI to offer code suggestions generated from the developer's context. This product helped establish the idea that a generative assistant could be present directly in a work environment rather than consulted in a separate tab.

The wave accelerated with ChatGPT, launched by OpenAI in November 2022. A few months later, in February 2023, Microsoft introduced a new version of Bing incorporating a conversational experience powered by OpenAI models. In fall 2023, the group gradually generalized the Copilot brand for its generative AI initiatives. Bing Chat became Copilot, while Microsoft 365 Copilot, announced in March 2023 and then made generally available to businesses in November of the same year, brought AI assistance into Office applications and Teams.

This history explains the current situation. Copilot was not born as a single product, but as a constellation. GitHub Copilot primarily targets developers. Microsoft 365 Copilot targets productivity and work data. Copilot for the web, Windows or mobile addresses more general-purpose uses. Azure AI and related offerings concern businesses that want to build or host their own AI applications. The multiplication of these entry points has enabled Microsoft to cover many use cases, but it also exposes the company to a simple question: which version of Copilot should a user open when they simply want to ask for help?

The announced super app addresses that question. The goal is not necessarily to make specialized products disappear. GitHub Copilot retains a natural place in code editors, just as assistants integrated into Word, Excel or Teams do. But Microsoft may seek to connect these experiences around the same space. For a user, this could reduce the boundary between framing a problem, producing a document, writing a program and launching automated actions.

From chatbot to an assistant that executes

The mention of agentic capabilities is the most structuring point of the announcement. A chatbot responds to a request: it explains, writes, summarizes or generates content. An agentic system aims instead to chain together several steps to achieve an objective. It can plan a task, choose a tool, verify a result, request confirmation and continue its execution. In practice, this promise depends on several elements: access to data, permissions, connectors, the ability to use software or services, control mechanisms and action traceability.

This distinction matters especially in the Microsoft universe. The potential value of an agent does not lie solely in the quality of its text response. It depends on its ability to interact with documents, calendars, files, mailboxes, code repositories, project management tools or business applications. Microsoft has a vast portfolio of software used daily in organizations. If a Copilot interface succeeds in connecting these environments with properly managed permissions, it could become a coordination tool rather than a mere text generator.

But the wording remains deliberately general at this stage. The confirmation reported by The Verge does not allow the conclusion that the future application will have access to all of a user's data, nor that it will be able to perform actions without validation. The term “agentic” today covers very different realities, from guided automation to more autonomous agents. Access conditions, safeguards and actual availability depending on accounts or regions will be decisive in assessing Microsoft's ambition.

The move toward agentification also raises a trust issue. An incorrect response is already problematic; an incorrect action can have more direct consequences, for example if it modifies a shared document, transmits information, changes a configuration or executes a development operation. In businesses, the usefulness of agents will therefore depend as much on supervision interfaces as on their capacity for initiative. Microsoft is familiar with this requirement in its professional offerings: access rights, logging, compliance and administration are essential components of AI deployed at scale.

ChatGPT, Cursor and the battle over AI interfaces

The future super app places Microsoft in broader competition than that of office suites. The first competitor is naturally ChatGPT. OpenAI has established a conversational interface that has become, for many users, the initial reflex when they want to write, seek an explanation, analyze a file or produce code. Microsoft has a distribution advantage through Windows, Microsoft 365, GitHub, Edge and Azure. But distribution alone does not guarantee regular use: a super app will have to be easier to understand than the current stack of Copilots and useful enough to become a daily destination.

The relationship between Microsoft and OpenAI makes this confrontation unusual. Microsoft invested $1 billion in OpenAI in 2019 and subsequently announced, in 2023, a multiyear, multibillion-dollar extension of its partnership. For several years, OpenAI's advances were at the heart of Microsoft's AI strategy, notably in Azure and the first Copilot products. But the announcement of a centralized application also shows that Microsoft wants greater control over the direct relationship with the end user, the identity of its assistant and the way its services are distributed.

Emancipation does not imply a break. Microsoft remains a major partner of OpenAI and Azure plays an essential role in the company's ecosystem. On the other hand, the group's strategy is no longer limited to integrating OpenAI technology into its own software. Microsoft has also introduced its own models, including MAI-1-preview and MAI-Voice-1 in 2025. These announcements do not make it possible to state which models will exactly power the super app announced for this year, but they confirm that the company is seeking to expand its internal capabilities and reduce its dependence on a single source of models.

This move is strategic. Owning an interface, a model, cloud infrastructure and an application portfolio provides greater leeway over costs, iteration speed and product priorities. In Microsoft's case, this vertical integration remains partial and evolving, but it outlines an ambition: to control the layers that connect computing, models, enterprise data and end uses. As assistants become intermediaries between people and software, this control may be as valuable as the language model itself.

Software development, a decisive front

The second front is code. GitHub Copilot was one of the first generative AI products to find a visible place in development teams. However, the market has become more crowded. Tools such as Cursor have popularized an approach in which the development environment and AI assistant are conceived as a whole, with interaction centered on the code repository, file changes and programming tasks. Major IDE publishers and model providers are also developing their offerings.

In this context, including coding assistance in a super app may seem paradoxical. Developers generally prefer to work in their editor, where technical context is immediately available. But Microsoft is not necessarily seeking to move all programming into a general-purpose application. A unified interface could instead be used to initiate a project, formulate a complex task, consult documentation, make tools interact or monitor agents tasked with certain operations. The code environment would remain specialized, while Copilot would play the role of a coordination layer.

This hypothesis aligns with the market's evolution. Generating a line or a function has become a relatively widespread feature. Value is shifting toward understanding an entire project, consistently modifying multiple files, running tests, reviewing changes and integrating with enterprise tools. For Microsoft, GitHub represents a major asset in this chain, because the platform sits both at the heart of collaboration among developers and source code management.

Competition with Cursor and other players will therefore not be determined solely by the raw quality of suggestions. It will be determined by the ability to limit friction: sign-in, access to context, control over changes, compatibility with existing workflows and transparency about AI actions. A developer will not adopt an agent because it is presented as autonomous, but because it saves time without introducing risks that are difficult to assess.

The promises and limitations of a super app for organizations

For businesses, the value of a unified Copilot interface would be to bring together tasks that are still handled in separate tools. A single request may today require research, reading documents, drafting a note, preparing data, sending a message and, in technical teams, modifying code or creating a ticket. An assistant capable of maintaining the thread of this request and proposing the right steps could reduce the context switches that characterize digital work.

Microsoft is particularly well positioned to defend this vision among its existing customers. Microsoft 365 is established in many businesses, Teams has become a central collaboration tool for a significant share of them, and Azure provides widely used cloud infrastructure. This presence does not mean adoption will be automatic, but it makes it easier to integrate an assistant into identities, security policies and environments already managed by IT departments.

The super app can also become an answer to a very concrete difficulty: the proliferation of AI tools. In many organizations, employees already use consumer chatbots, assistants integrated into software, transcription tools, presentation generators and coding solutions. This fragmentation can complicate the management of data and subscriptions. A centralized offering is appealing to IT managers if it makes it possible to clearly define access rules, choose authorized connectors and track usage.

However, consolidation is not automatically synonymous with simplicity. A super app risks accumulating modes, settings and promises that are difficult to distinguish. Microsoft will notably have to avoid having the same Copilot name cover incompatible experiences across the general public, developers and Microsoft 365 customers. Clarity regarding functional limits, subscription terms and accessible data will be essential. Users must be able to know which assistant is acting, with which tools, on which information and under whose responsibility.

Data, security and sovereignty: particularly European issues

In France and Europe, the announcement will be viewed through the lens of data protection and digital sovereignty. An application bringing together chat, documents, code and agents can concentrate sensitive information: personal data, internal content, intellectual property, professional exchanges, commercial information or industrial secrets. The benefit of shared context is precisely what makes the issue delicate. The more the agent knows about the work environment, the more useful it can be; the more strictly what it can consult and do must be governed.

French companies will therefore need to look beyond the interface. They will be interested in data residency and processing settings, contractual commitments, identity management, separation between enterprise data and personal data, as well as auditing capabilities. These questions already exist for Microsoft 365, Azure and GitHub. They become more pressing when AI is likely to establish links between several tools and carry out chained actions.

The European regulatory framework reinforces this vigilance. The AI Act entered into force on August 1, 2024. Its obligations apply according to a progressive timetable, and rules concerning general-purpose AI models began to apply in August 2025. Most other provisions are due to apply in August 2026. Without prejudging the precise legal classification of the future Copilot application, this timetable requires major providers to consider documentation, risk management and the transparency of their offerings at a very early stage.

The GDPR remains just as central. In professional use, an assistant's ability to retrieve, summarize or connect information can have effects on individuals, even where the initial objective is purely operational. Organizations will need to define which content can be entrusted to the tool, which connectors are enabled and which human validations are required. For regulated sectors, the question will not be whether AI can help, but in which tasks it can be allowed to intervene without compromising control obligations.

This European constraint can become an advantage for providers that offer manageable and understandable mechanisms. Microsoft is accustomed to selling to large organizations with high compliance requirements. But agentification raises the bar: it is no longer enough to secure a document stored in a service; the decision-making path of an assistant capable of consulting multiple sources and triggering an action in another piece of software must also be secured.

A transition toward the intention-driven computer

The Copilot super app is part of a broader transformation of personal and professional computing. For decades, users have worked by choosing an application, then a command: open a spreadsheet, write a formula, create a folder, send an email, launch a code editor. AI pushes toward a logic of intention: describe the desired result, then let the system propose the necessary tools and steps. In this model, the chatbot is no longer an autonomous destination. It becomes a command layer above software.

This evolution will not make traditional applications useless. Specialists will continue to need precise interfaces to verify data, design a document, develop software or administer infrastructure. But the assistant could absorb part of the preparation and coordination operations. This is particularly true for tasks that span multiple products and currently require many repetitive manipulations.

Microsoft has an interest in accelerating this change, because it aligns with the nature of its portfolio. The company does not depend on a single flagship application: it sells a set of platforms and services. A successful super app could serve as a link between these assets. It would offer Microsoft a way to defend the relevance of Windows and Microsoft 365 at a time when a growing share of interactions begins in an external conversational assistant.

The risk is symmetrical. If the Copilot interface does not provide a sense of continuity, users may retain their habits: ChatGPT for general questions, a specialized coding assistant for development, Microsoft 365 for documents and other services for automation. The value of the super app will therefore depend on the quality of the bridges between these worlds, not merely on the presence of numerous functions in a single menu.

The next test will be reliable execution

The announcement confirmed by The Verge sets a course before the end of the year, but leaves open the questions that will determine the product's real impact. Microsoft will need to clarify how existing Copilot experiences will be connected, which audiences will be concerned at launch, what role GitHub and Microsoft 365 will play, and above all what agents will actually be able to do. Between an assistant that suggests a plan and an agent that executes tasks across multiple software applications, the gap is substantial in terms of usefulness, security and responsibility.

For the French-speaking market, the promise is attractive: unified AI can reduce the complexity of adoption for employees, self-employed workers and technical teams already equipped with Microsoft products. It can also further concentrate distribution power around major American platforms, at a time when European players are seeking to develop their own models, infrastructure and interfaces. Choices regarding compatibility, data governance and regional availability will therefore matter as much as productivity demonstrations.

Over the longer term, Copilot's success will not be measured by the number of announced capabilities. It will depend on Microsoft's ability to make the agent a reliable, understandable and controllable intermediary between the user and their digital environment. If this step is achieved, the super app could become the daily face of Microsoft's AI strategy. Otherwise, it will remain an additional layer in a landscape already saturated with assistants.