Microsoft’s Applied Sciences Group unveiled Project Solara at Build 2026, a chip-to-cloud platform designed from the ground up for devices where AI agents replace traditional app interfaces. The platform runs on Android Open Source Project (AOSP), not Windows, and targets enterprise verticals including healthcare, retail, and financial services.

Steven Bathiche, who leads Microsoft’s Applied Sciences Group, framed the initiative around a simple thesis: agents reduce the cost of hardware specialization. When software no longer depends on full app frameworks, browsers, or traditional UX stacks, the barrier to building purpose-specific devices drops. “The next platform shift is from apps to agents,” Bathiche wrote, describing the move “from software you open to intelligence you invoke.”

Two Concept Devices

Microsoft showed two proof-of-concept devices during the Build 2026 keynote on June 2. The first is a wearable badge designed for information workers, frontline staff, and healthcare personnel. It provides glanceable access to a Priority Agent and can record hallway conversations through Microsoft Facilitator, the company’s existing meeting transcription agent. The second is a “desk concept” device, similar to a standalone speaker, that connects users to their PCs, monitors, and Windows 365 clients via voice.

Both are multimodal: voice input, visual output, and contextual routing to the right agent at the right moment. Neither requires users to navigate traditional app interfaces.

Enterprise Pilot Partners

Microsoft named five companies entering private pilots “in the coming months,” according to Mary Jo Foley at Directions on Microsoft: AccuWeather, Best Buy, CVS, Levi’s, and Target. Silicon partners include MediaTek and Qualcomm. No pricing, availability dates, or production hardware specs were disclosed.

The choice of retail and healthcare companies signals where Microsoft sees the first viable market: environments with frontline workers who need agent access but don’t sit at desks. A nurse checking patient status through a badge, a retail associate querying inventory via voice, a warehouse worker confirming shipment details without pulling out a phone.

Why Android, Not Windows

Project Solara’s use of AOSP instead of Windows is the most telling architectural decision. Windows carries decades of app compatibility requirements, driver ecosystems, and UX conventions that make it difficult to strip down for single-purpose agent hardware. Android AOSP provides a lighter base that silicon partners already know how to manufacture against, with existing supply chains for wearables, kiosks, and embedded devices.

Bathiche’s blog post describes the platform as “liminal,” with the operating system spanning device and cloud. Azure handles state management across what Microsoft calls “a constellation of specialized devices.” The agent runs between user intent and distributed execution, while the device itself becomes “a window into long-running intelligence.”

The Three-Pillar Architecture

Project Solara’s platform rests on three design pillars, according to Microsoft’s technical overview:

  • Enterprise readiness: Hardware and software manageability, security, and privacy controls integrated at the platform level, not bolted on after launch. Devices connect to WorkIQ and existing Microsoft management infrastructure.
  • Agent-driven interaction: “Just-in-time UI” that adapts across devices and modalities without requiring developers to redesign for each form factor. Today that means adaptive cards and structured content types. Over time, Microsoft envisions more dynamic, generative interfaces.
  • Multi-agent extensibility: Organizations can bring their own agents alongside Microsoft’s. The platform manages boundaries between data, domains, identities, and organizations.

The Specialization Bet

The deeper argument behind Project Solara is economic. Bathiche laid out a framework in his blog post: every previous computing platform shift (mainframes to PCs, PCs to phones, phones to watches) produced more specialized, more contextual devices. Each generation moved compute closer to the user and closer to the task. Agents continue that trajectory by reducing the software stack required to build a new device category.

“Historically, specialization has been expensive,” Bathiche wrote. Building a new type of computer meant constructing hardware, software, services, developer tools, UI patterns, management systems, security models, and an ecosystem from scratch. AI agents compress that stack. With just-in-time UI, fewer apps need to be written per device. With agentic coding, developer SDKs require less human-facing polish. The result: lower cost per new device category.

Whether that theory holds in practice depends on whether enterprises actually adopt single-purpose agent hardware instead of running agents on the phones and laptops they already own. Microsoft is betting that frontline workers, healthcare staff, and retail employees represent a market large enough to justify dedicated devices. The pilot results from Best Buy, CVS, and Target will be the first real test.