Copilot Is Learning to Do the Work. Is Your Business Ready?

When Microsoft launched Copilot in 2023, it was a helper. It drafted emails, summarised meetings and tidied spreadsheets, and the person using it decided what to keep. The updates Microsoft announced on 25 September go much further: Copilot can now carry out multistep tasks, is beginning to build apps from a plain description and, in preview, will keep working towards a goal while its owner is offline. Many leadership teams are still managing Copilot as a productivity tool, counting licences and asking whether people use it. Those measures made sense when Copilot was only a helper, but they say very little now that it can do the work itself. 

This article walks through each update in plain terms: what it does, what is available now and what is still coming, and what it means for the people running a business. Microsoft has started building the infrastructure for handing work to AI, but most organisations have not yet decided what they would hand over, who would answer for it or what it should cost. The article ends with the Readiness Check, five practical questions a leadership team can answer in a single meeting to see where the business stands and what to fix first.

The simplest way to understand the September updates is to compare what you could ask Copilot to do three years ago with what you can ask it today. In 2023 the request was “help me”: summarise this thread, draft that reply, suggest a formula. Now it can be “do this”: prepare the briefing for Thursday’s board meeting or chase the suppliers who have not sent their figures, and soon “build this”: a tracker for the project team. Some of the pieces that make this possible are already here and others are on the way, from a single starting screen to agents that run in the cloud with their own identity and budget. 

The bigger change is in what the technology asks of senior managers. A helper needs people who know how to use it, which is why most Copilot programmes have focused on licences, training and champions. Software that can do work needs something different. Someone has to decide what work it takes on, what it may do without asking, who answers when it gets something wrong and whether the result was worth the cost. Those are operating decisions, the kind usually made by a chief operating officer, a finance director or the head of a function, and the September updates put them firmly on the leadership agenda. 

It also helps to separate what is happening now from what has been promised. Some of these features are in daily use, some are in early access and some are previews that may change before most customers see them, and each section below says which is which. The direction itself is not in doubt, and it is not Microsoft’s alone: OpenAI released an Agents API in September designed to keep AI agents working for days at a time. A board deciding where to spend time and money needs to know both how far the technology has come and how much of it is ready to rely on. 

Home is becoming the starting screen in the Copilot app. It brings together Chat, for questions and drafting, and Cowork, for tasks Copilot carries out, with Word, Excel and PowerPoint built into the same

experience. Instead of choosing an app first, people start by saying what they need done. Microsoft says a routing feature will follow that decides for itself whether a request belongs in Chat, Cowork or Code. 

Two smaller additions follow the same idea. A new feature called Today, due to enter private preview this month, will pull the important items from someone’s email, calendar, Teams chats and meetings into one view and suggest what to do next. Another will let a team bring Copilot into a Teams channel, group chat or meeting, where it works with the information that group shares. Both turn Copilot from a tool one person opens into part of a team’s working day. 

For leaders, Home is an early sign of how work will start in future. If people simply describe the result they want and let Copilot pick the tool, what comes back will only be as good as the business’s processes and data, and how clearly it has defined a good result. Where it stands: Microsoft said in late September that Home would reach early-access customers through its Frontier programme in the coming weeks, with the new Teams feature entering private preview at the end of September and Today following this month. 

Cowork is the part of Copilot that is already changing how work gets done. You tell it what you need, and it plans the steps and does them for you across your email, calendar, documents and business systems, showing its progress along the way. Where Copilot Chat answers a question or drafts something for you to act on, Cowork takes the action itself and hands back the finished result. It can prepare a briefing, organise a week or research a company, and it pauses for approval before it sends an email or posts in Teams. It can also run on a schedule, or start automatically when a particular email or Teams message arrives, preparing the response and holding it for sign-off. Everything it does runs with the user’s own permissions, so it can see only what that person can see. 

The business value is in the handoffs. Much of a manager’s week goes on work that sits between other people’s work: gathering updates, preparing for meetings, following up on actions and turning scattered information into something a decision can rest on. Cowork can take on a good share of that coordination, which is where many organisations lose time without ever measuring it. The approval step keeps a person in charge of anything that leaves the building, and senior managers should treat where those approval points sit as a design decision for the workflow, not a setting each user chooses.

Where it stands: Cowork has been available to all Microsoft 365 Copilot customers since June, and it is paid for according to how much it is used, on top of the Copilot licence. That makes it the one feature leaders can try out properly today. Teams that use it now will learn which tasks it handles well and which need checking, and that experience will help them decide how to use Autopilot, Microsoft’s more independent agent, when it arrives. 

Code is designed to let people who do not write software build apps, trackers, dashboards and automations by describing them in plain language. Copilot chooses the approach, builds the result and can host it inside the organisation’s own Microsoft 365 environment through Copilot Managed Runtime, a hosting service now in public preview. Microsoft says Code uses the same underlying technology as GitHub Copilot, which professional developers will keep using for their day-to-day work. In practice, once Code reaches them, an operations manager who needs a simple tool to track supplier issues could have one by the afternoon, without raising a request with IT. 

That is good news for any business with a long queue of small requests, and it raises questions senior leaders have met before with spreadsheets and low-code tools. Who owns an app when the person who built it changes role? When does a handy personal tool become something the business depends on and needs supporting properly? Managed Runtime gives IT a single inventory of every app, with company sign-in, data protection policies and version control applied, so the technical controls are there. The ownership decisions are not, and they belong with business leaders as much as with IT. 

Centrica, the UK energy company behind British Gas, shows how to manage this, based on its experience with Microsoft’s tools for building apps without much coding, set out in a customer story Microsoft published in June. As hundreds of its employees started building their own apps, it introduced a simple sign-up step: anyone starting an app has to register it, recording what it is for, who it affects, how many people will use it and any financial or regulatory risks. Higher-risk apps are reviewed by business and technical leads, who also check whether a similar tool already exists, and each app’s results are tracked against what was registered. Where it stands: Microsoft said in late September that Code would begin reaching early-access customers in the coming weeks, with wider availability to follow, and that it will be paid for according to how much it is used. 

Autopilot is the most ambitious part of the announcement. Microsoft describes an agent that a manager can give a name, a role and a goal, and that keeps working towards that goal over days, following threads and channels, handling recurring tasks and picking a project back up without being asked, including while its owner is offline. Microsoft’s example is a supplier review, in which Autopilot would build the timetable and working plan, prepare the meetings, handle the follow-ups and ask stakeholders for their updates. Microsoft says each Autopilot agent will have its own identity, memory and workspace inside the organisation’s Microsoft environment. 

This is where Copilot starts to resemble a member of staff with a defined job, and senior managers will recognise the questions that follow. What exactly is the goal, and how would anyone know it had been met? What may the agent say to a supplier, and what must it pass to a person? The independent evidence argues for handing over a little at first. Model Evaluation and Threat Research (METR), an independent group that tests advanced AI systems, reported in May that the best AI agents could sometimes complete software tasks that would take a skilled person more than a full working day. But the tasks they could complete reliably were far shorter, and on harder problems they tended to overstate what they had achieved. 

That points the first uses towards coordination work whose results a manager can check at a glance: meetings booked, updates received, reminders sent. Negotiation, approvals and anything that depends on reading a relationship should stay with people, who are far better placed to judge what a supplier really meant. Where it stands: Autopilot, previously tested under the name Scout, was due to enter private preview at the end of September, and Microsoft has not yet published full detail on its limits, controls or pricing. This is the right time to plan for it and too early to build an operating model around it. 

Three less visible changes matter as much as the headline features. The first is data. Copilot can now use the business data your teams already report on in Power BI, Microsoft’s reporting tool, and links to Microsoft’s finance, sales and operations systems are starting to roll out in preview. IT is also getting a single list of approved add-ons that connect Copilot to other software, whether from Microsoft, its partners or the organisation’s own developers. The more Copilot can reach, the more useful it becomes, and the more it matters that the right people can see the right information, because an agent will use whatever it is allowed to see. 

The second is identity. Through Entra, Microsoft’s security system, an AI agent can now have its own login, separate from the person who built it, with its own access rights and its own record of everything it does. Each agent must have a sponsor: a named business owner who answers for why it exists, is asked whether to keep it or switch it off, and is the person the security team calls if something goes wrong. Agent 365, available since May, gives organisations a single list of all the agents they run. The rules are still being written, and in February the US National Institute of Standards and Technology described how to identify and control AI agents as an open question. 

The third is billing. Everyday Copilot use is still paid for with a standard licence per person, but Cowork is already charged by how much it is used, and Code and Autopilot will be too. Microsoft is giving finance and IT teams tools to set spending limits and choose which AI models can be used, and will soon show which tasks deliver the best results. For a finance director, the question changes from how many people need a licence to which work is worth paying software to do, and at what cost per result. 

It would be easy to read all of this as a reason to buy more licences and wait for the results. The best independent evidence on AI assistants argues against that. In a randomised field experiment involving thousands of knowledge workers across dozens of firms, people who used Microsoft’s assistant in their email, meeting and writing tools spent about two fewer hours a week on email and did less work outside normal hours. Beyond those personal savings, the researchers found no change in the amount or mix of work people did. Some of the authors worked for Microsoft, which makes the modest finding harder to dismiss. 

The reason is that an assistant only changes what one person controls. Meetings, handoffs, approvals and the way work moves between teams stay as they were, because nobody redesigned them. The newer capabilities can reach across those boundaries, but only if someone decides they should, and that decision sits with the leaders who own the processes. 

Premera Blue Cross, a US health insurer, shows what happens when that decision is made, in a case study Microsoft published in April. Its team used Copilot Studio to build an agent that pulls the details out of incoming contracts and presents them to a person for checking before the work is routed onward, cutting the time spent on each contract from most of an hour to a few minutes. The gain came from redesigning one high-volume workflow around the technology, with human checking kept at the point where accuracy matters most. Premera also set up a cross-functional committee on data, AI and ethics to agree guardrails, so staff could experiment within clear limits. The figures are the company’s own, but the pattern is the one leaders should look for: one important workflow, redesigned on purpose, with people in charge of the judgement. 

Copilot is becoming capable of doing a growing share of the work in a business, and Microsoft is building the controls, identities and billing that make handing work over practical. The opportunity is real for any organisation that spends too much senior time on coordination, reporting and chasing. Capturing it depends on choices the technology cannot make: which work to hand over, who owns it, where people must sign off and what a result is worth. Those choices sit with senior leaders, and the businesses that make them early will be the ones that get the value. 

The test of readiness is simple. If your leadership team can answer the five questions for one important workflow this quarter, you are ready to start. Microsoft is making Copilot ready to do the work. Making the business ready is the leader’s job.