Microsoft Puts Copilot in the Windows Core With Agent 365 and Microsoft IQ

Microsoft held its first major Windows event since May 2024 in San Francisco on October 7, 2026. Satya Nadella, Nvidia CEO Jensen Huang and Windows and Surface chief Pavan Davuluri took the stage with a single thesis: AI moves off the cloud and runs on the machine. The new names are Agent 365, Microsoft IQ and a Copilot built “into the core” of the system.

Quick answer: what did Microsoft announce?

That Copilot stops being an application on top of Windows and becomes part of the system core, alongside Agent 365 (an enterprise agent layer) and Microsoft IQ. The hardware behind it is Nvidia’s RTX Spark platform, debuting in the Surface Laptop Ultra and running the Nemotron model locally. And despite the scale of the event, there is no Windows 12.

What “Copilot in the core” means

Until now Copilot was a layer on top: a panel, a button, an app that talked to the cloud. Moving it into the core changes three things at once — it gets access to the file system and settings without an intermediary, it can act offline, and it stops being removable the way an ordinary program is.

That last part is what will generate argument. Integrating a feature into the operating system is exactly the move that earned Microsoft two decades of antitrust litigation, starting with Internet Explorer. The difference in 2026 is that the direct competitor — Google, with the Googlebook launched three days earlier — is doing the same thing with Gemini.

The pieces announced

Name	What it is
Copilot in the core	Assistant integrated into the system, no longer an app on top
Agent 365	Agent layer for enterprise environments
Microsoft IQ	New intelligence layer announced at the event
RTX Spark + Nemotron	Nvidia platform running the model locally
Surface Laptop Ultra	Microsoft’s first hardware with the dedicated chip

The real contest: where AI runs

The event confirms the 2026 turn. In 2023 and 2024 the race was over who had the biggest cloud model. Now it is over who can run a useful model inside the device — Apple with its own silicon, Google with the Googlebook and Gemini, Microsoft with Nvidia and Nemotron.

The motivation is not only technical. Local AI cuts inference cost, solves part of the enterprise privacy problem and removes the dependency on a connection. For Microsoft, which pays Copilot’s cloud bill, every task that migrates to the customer’s device is margin recovered.

Why this matters

If you use Windows, the practical effect arrives in stages: first on new machines with an NPU or dedicated chip, then on existing hardware with a reduced feature set. There is nothing to do today except one thing — when buying a PC from here on, look at the memory and the AI accelerator, not just the processor, because that is what decides which features your system gets in 2027.

Frequently asked questions

Is Windows 12 coming?

No. Despite being the first major Windows event in two years, Microsoft did not announce a new numbered version of the system.

What is Agent 365?

An agent layer aimed at enterprise environments, announced alongside Microsoft IQ and the integration of Copilot into the system core.

Will the AI run without internet?

Part of it, yes. Nvidia’s RTX Spark platform runs the Nemotron model locally on the device.

Do I need a new PC?

For the local features, yes — they depend on hardware with an AI accelerator. Current machines should receive a reduced set of capabilities.

At DigitalRadar, we track where the giants are moving computing. Stay on the radar.