Microsoft seems to be quietly taking Copilot in a new direction.

The focus is no longer just on helping users with quick responses. Instead, the company is working on turning Copilot into something that can actually take over tasks and see them through on its own.

Speaking at the GitHub Constellation India 2026 event, Microsoft EVP Jay Parikh shared that the company is exploring ideas inspired by OpenClaw, an open-source AI agent framework. The goal is to build AI that can stay active, handle multi-step work, and require far less back-and-forth from users.

What’s interesting is that Microsoft isn’t launching this as a separate product. These changes are being tested quietly within Microsoft 365 Copilot, suggesting a gradual shift rather than a sudden rollout.

Parikh pointed out that frameworks like OpenClaw show how AI can work across platforms and remain available in the background. It can plug into everyday apps, adjust to how a user works, and handle both simple and more serious tasks depending on how it’s set up.

The company has already started moving in this direction with tools like Copilot Cowork and Copilot Tasks. These features are aimed at reducing everyday workload—things like managing emails, scheduling meetings, and handling routine office work.

With Copilot Cowork, users can go a step further and hand over more demanding tasks. These AI agents can work independently, connect with different systems, and keep things moving without needing constant input.

Another big piece of this is how Copilot connects with company data. It can pull information from emails, messages, and documents, making it easier to sort through problems and get things done faster.

At the same time, Microsoft is being cautious. Since tools like OpenClaw can run locally and sometimes involve untrusted code, there are real security concerns. Microsoft is trying to avoid those risks by building stronger safeguards into Copilot, especially as it focuses on enterprise users where security matters the most.

Also Read: Microsoft Launches Three Models to Reduce Reliance on OpenAI