Microsoft is changing what it wants Copilot to be. Instead of placing the assistant in as many Windows apps as possible, the company is focusing on whether Copilot gives people a reason to return.
The shift follows a hard lesson for Microsoft and Jacob Andreou, its EVP of Copilot. The company removed Copilot from several Windows apps after finding that those placements brought traffic but delivered little value. Microsoft now has fewer entry points for Copilot, but more usage per user.
More reach did not mean more value
Andreou says he ran into the same problem as a user of Microsoft’s own consumer agent, and he expects this fall’s personal agents to face it too. People may try an agent once because it is easy to find, but that first visit does not prove the product has earned a place in their routine.
“We had to relearn that putting more people into the top of the funnel doesn’t equal delivering more value to them,” Andreou says. His point is simple: bringing people to an AI product matters less when the product does not give them a useful reason to stay.
Andreou puts the lesson even more plainly: “Usage is fleeting unless you’re delivering real value.” That view helps explain Microsoft’s pullback on Copilot in some Windows apps. The company is moving away from reach as the main measure and toward repeated use tied to something the assistant can actually do.
“Products should earn their right to exist,” Andreou says. For Copilot, that means being more than another place to ask a question. It means helping people complete work, find information, and take actions that make the product useful beyond a first experiment.
Copilot moves closer to the PC
Microsoft’s new version of Copilot addresses earlier limitations by giving it a more direct role on the computer. Copilot can now access documents and take actions on a PC using local AI models instead of Microsoft’s cloud models.
That change matters because the assistant is no longer limited to producing an answer in a conversation. It can work with documents and act on the PC itself, connecting the AI experience to tasks people already perform there. The local models also make the computer part of the Copilot experience rather than treating the PC as a simple window for a cloud service.
Microsoft has not presented this as a race to add Copilot to every possible surface. The company’s current approach puts more weight on what happens after someone opens the product. Fewer entry points can still lead to stronger use when each interaction offers a clear benefit.
Copilot Tasks in the consumer product increased Andreou’s own consumption. That detail gives the strategy a practical test: a feature that helps Copilot do more can change how often someone uses it, even when the product appears in fewer places.
Why the model is only part of the product
Andreou also says the AI model itself is not the whole story. “The model definitely matters. But having to behave as if the model really matters is a large tax to put on professionals and companies.” His concern is that users should not have to track which model leads at a given moment to get dependable value from a product.
He points to a more mixed model landscape. “Starting with 5.6, OpenAI’s new class of models are really smart,” Andreou says, but he also says no single lab’s models win at everything anymore. That makes product design more important because the best experience cannot depend on treating one model as the answer to every task.
“Performance now depends on the type of output rather than which model is currently on top,” Andreou says. In practice, that places the focus on how Copilot handles a job, not on which model name sits behind the feature at that moment.
His broader argument is that every advance in AI needs a useful way for people to access it. “Each generation of intelligence needs a product form factor for you to actually get value out of it.” For Microsoft, that form now includes an assistant that can access documents, act on a PC, and use local AI models.
The Copilot reset is therefore less about putting AI everywhere and more about making each part of the experience justify itself. Microsoft pulled back from Windows placements that produced traffic without enough value, while adding capabilities that can give users a reason to return.
That leaves Copilot with fewer doors into the product, but a clearer purpose once people arrive. The test for Microsoft will be whether local actions, document access, and Copilot Tasks turn that purpose into lasting use.
Based on
