Microsoft turns Copilot into an agent that works for you: what changes for businesses
With Autopilot, Code and Home, Microsoft’s assistant moves from conversation to carrying out tasks, sometimes without constant supervision. New usage-based billing, a choice of models, spending controls: African businesses need to look closely.
- On 25 September, Microsoft unveiled Home, Code and Autopilot to turn Copilot into a platform that carries out tasks.
- Autopilot is a personal agent that keeps working without constant intervention; Code builds apps from a description.
- Advanced uses (Cowork, Code, Autopilot) may be billed by consumption, with a choice of AI model.
- For businesses: budget, access security and data protection must be framed from the start.
Microsoft wants its artificial intelligence assistant to do more than answer. In an announcement published on 25 September 2026, the company presented three new Copilot building blocks, Home, Code and Autopilot, designed to move the tool from generating content to carrying out complex tasks, sometimes over time and without constant user intervention.
The announcement is part of a trend running through the whole sector: after chatbots, the big software vendors are betting on “agents”, programs able to chain actions on their own. For businesses, including African ones, that make heavy use of the Microsoft 365 suite, the consequences are concrete: new uses, new billing, new risks.
Autopilot: an agent that carries on without you
The most ambitious new feature is called Autopilot, formerly known as Scout. Microsoft presents it as a personal agent able to carry out tasks over time, drawing on the organisation’s account, memory and workspace. The user gives it a name, a role, a goal and a scope. The agent can then consult the company’s communication channels, send follow-ups, handle repetitive tasks and keep working while its user attends to something else.
Microsoft gives an example: Autopilot draws up a schedule and a work plan, prepares the necessary meetings, follows up and checks progress with the people involved. The user tracks progress, adjusts the scope or steps in when needed. Autopilot is to be extended as part of a private preview.
Code: apps without programming
With Code, a user describes in everyday language the app, dashboard or automation they need, and Copilot works out how to build it. These can be small office tools, interactive dashboards to analyse data or web apps to share with a team. The feature relies on GitHub Copilot technology, and the code produced runs in an isolated environment. A preview also allows these apps to run in the company’s Microsoft 365 environment, under administrators’ control.
Home: a single entry point
Home becomes the gateway to this new generation of Copilot. It brings together Chat, for everyday requests such as looking up information or drafting a first version, and Cowork, for complex assignments carried out end to end, such as preparing a tender response, a client briefing or an analysis of regulatory data. The user no longer has to choose the tool: they describe what they want, and Copilot routes the request. Word, Excel and PowerPoint documents can be created and edited directly from the conversation, and changes are synced with the team’s files.

The bill changes too
This may be the most important element for finance departments. Microsoft now distinguishes everyday use of AI from agents that work for long periods. The subscription covers Copilot in Chat and in the Office apps. However, Cowork, Code, Autopilot and some long-running agent tasks may fall under usage-based billing.
An automatic mode, included in the subscription, itself chooses the AI model best suited to each request based on accuracy, speed and cost. For consumption-billed uses, users will also be able to choose the model, to match spending to the value of the work. Microsoft is widening its range of models, with third-party models and new OpenAI models in Word, Excel, PowerPoint, Chat, Cowork and Copilot Studio.
To avoid unpleasant surprises, the company is strengthening its tools for tracking AI spending. Administrators can define which models are available to which groups and set consumption rules. Employees see their credit usage, balance and history directly in Copilot.
« It is about connecting the tools users know to the next-generation capabilities needed to create, customise and deploy AI. »
Jared Spataro, chief marketing officer for AI at Work at Microsoft (paraphrased)
What African businesses should look at
In many African companies and government agencies, Microsoft 365 is the standard work tool. Copilot’s new features will therefore arrive there naturally. Three points deserve particular attention. Cost first: usage-based billing, paid in dollars, can weigh heavily when the local currency depreciates, and budgets must be framed from the outset. Security next: an agent that reads messages and acts on behalf of an employee must have strictly limited access, as recent incidents involving AI agents at other companies have shown. Data finally: African data protection laws apply to the information these agents handle.
Connectivity is also an issue. An agent working in Microsoft’s cloud uses bandwidth and requires a stable connection. In African companies where the internet remains expensive or unreliable, the value of these tools will depend on the quality of the network. Finally, staff training will be decisive: a powerful tool that is poorly understood mainly produces errors, and employees must learn to set clear goals, check results and recognise when the agent is wrong. It is a human investment as much as a technical one.
Concrete uses, and limits
In a small business or a government agency, the possible uses are many: an agent that prepares a project status report every Monday from the team’s emails and documents, chases late suppliers, compiles the responses to a tender, or builds a sales tracking dashboard without going through a developer. For organisations short of skilled staff, the time saved can be real.
The limits are well known. An agent can make mistakes, misunderstand an instruction or act on outdated information. The more autonomous it is, the more its errors can spread before a human sees them. Hence the importance of the safeguards Microsoft highlights: a defined scope, progress tracking, the ability to step in. The first companies to adopt Autopilot will have to measure not only the time saved, but also the time spent checking and correcting the agent’s work.
A race between vendors
Microsoft is not alone in this field. Google, OpenAI, Anthropic and many enterprise software vendors also offer agents able to act within work tools. The battle is no longer only about the quality of the models, but about integration into the software employees already use, and about the trust of IT departments. Microsoft’s advantage is its presence on desktops around the world; its challenge will be to convince customers that these agents save more than they cost, and that they stay under control. The first previews, and feedback from pilot companies, will show whether the promise holds.
- Microsoft transforme Copilot en « IA d’exécution des tâches » et dévoile Home, Code et Autopilot, ZDNet France.
- Microsoft veut transformer Copilot en plateforme de travail, Le Monde Informatique.
Spotted an error? Write to infos@novotech.blog. Corrections are flagged at the end of the article.




