At Build 2026, Microsoft showed an operating system and two concept devices built around a single bet: that agents need hardware of their own. The platform is codenamed Project Solara, and its most interesting design choice is hiding in plain sight. It assumes you will run several agents, not one assistant.
What Microsoft announced
Solara is a chip-to-cloud platform for agent-first devices, pitched as an operating system that spans the device and the cloud rather than living on one box. Microsoft paired it with two reference devices to make the idea concrete. A portable badge, built with Qualcomm, carries a touchscreen, a fingerprint sensor, a far-field microphone array, a side-facing camera, and 5G. A stationary desk unit, built with MediaTek, adds facial-recognition sign-in, a UWB presence sensor, and dual USB-C, and can run on its own or turn into a Windows 365 client with a display attached.
The enterprise plumbing is the part worth reading twice: device management through Microsoft Intune, identity through Entra ID, biometric sign-in via Windows Hello for Business, all on MDEP, Microsoft's AOSP-based device platform. Pilots are starting in the coming months with AccuWeather, Best Buy, CVS Health, Levi's, and Target. Microsoft published no pricing and no general-availability date.
Solara is built for multiple specialized agents with a dispatcher that activates the right one. That is an orchestration problem, now baked into the operating system.
The design choice that matters: multi-agent by default
Most agent products still center one dominant assistant. Solara goes the other way. It is built for a set of specialized agents, with an agent dispatcher and task manager that brings the right one forward based on what the user is doing. A nurse's badge can run Dragon Copilot; a developer's surface can run GitHub Copilot; Microsoft 365 Copilot answers by voice, grounded in WorkIQ data. The interface adapts to the device instead of forcing a redesign per form factor.
We have argued that capability without coordination just produces orchestration versus chatbot sprawl. Routing work to the right agent, arbitrating between them, and managing their state is the core of that problem. Seeing it promoted to an operating-system primitive, by the largest enterprise software vendor, tells you the multi-agent model is no longer a niche architecture choice.
Security is the enterprise unlock
The physical mic-mute button and recording indicators are the visible half of Solara's security story. The half that decides whether IT can deploy a fleet is Intune and Entra. An agent device that enrolls into existing management, carries a real identity, and authenticates the person in front of it is a device a security team can actually approve. We made this case about software in who's governing your AI agents: every agent needs an identity, least-privilege access, and an audit trail. Solara pushes that control plane down to the hardware.
What it doesn't settle yet
These are concept devices and early pilots, not shipping products. No price, no availability date, and a partner list that is a starting gun rather than a finish line. A device that can host five agents still does not tell you which agent should touch which system, how it proves it got the answer right, or how it ties into the line-of-business tools a company already runs. That work sits above the OS, and it does not arrive in the box.
Solara also lands inside Microsoft's stack: Copilot Studio, the Microsoft 365 Agents SDK, the Microsoft Agent Framework. Useful if you live there. Most enterprises run a mix, and the orchestration, governance, and integration that make agents trustworthy have to work across vendors, not inside one.
What we're doing about it
Foundation builds and runs multi-agent systems for clients today, and we do it vendor-neutral. Whether an agent runs on a Solara badge, a laptop, or a server in the client's office, the hard parts are the same: deciding which model handles which step, giving each agent a scoped identity and an audit trail, wiring it into real tools, and evaluating that it works. That layer ports across whatever hardware ships. Microsoft's framing, from technical fellow Steven Bathiche, is that agents will reshape not only software but the devices themselves. We have been building for that world; it is good to see the hardware catch up.
If you're sorting out what an agent-first stack means for your team, on Microsoft's hardware or anyone else's, let’s talk — we build the system that runs on top of it.