Microsoft has taken the wraps off Project Solara, a bold new initiative that reimagines the enterprise endpoint as a device built to run AI agents, not legacy software. Announced at the Build 2026 conference on June 2, the chip-to-cloud platform is built on the Android Open Source Project (AOSP) and aims to redefine how organizations deploy, manage, and interact with purpose-built hardware. Think of it as a blank canvas for AI-driven productivity, where the operating system fades into the background and context-aware agents handle tasks traditionally reserved for sprawling suites of packaged applications.

Satya Nadella, during his keynote, described Project Solara as "the next logical step in our intelligent edge strategy. We're moving from an app-centric world to an agent-centric world, where the device becomes a thin, secure conduit for AI-powered workflows." This marks a significant departure from Microsoft's Windows-first heritage, signaling a willingness to diversify the endpoint portfolio with a lightweight, security-hardened OS that carries zero legacy baggage.

AOSP as the foundation, not Android

Project Solara is not your typical Android tablet. By stripping away the Google Mobile Services layer and forking directly from the Android Open Source Project, Microsoft gains full control over the user experience, security policies, and update cadence. There is no Play Store, no pre-loaded consumer apps, and no bloatware. Instead, the platform exposes a minimal runtime optimized specifically for hosting AI agents developed with Microsoft's Copilot Stack and third-party tools.

The decision to build on AOSP is strategic. It grants Microsoft access to a mature, widely adopted Linux kernel with robust hardware abstraction layers, a vast driver ecosystem, and proven power management capabilities—without the license fees or engineering complexity of maintaining a proprietary kernel. At the same time, by removing any dependency on Google services, Microsoft avoids the entanglement of Android's certification programs and can deliver a truly "Microsoft-managed" endpoint, akin to what has been accomplished with Windows but on a drastically leaner footprint.

Hardware requirements are expected to be modest. The reference design, according to internal documents glimpsed during Build, points to ARM-based SoCs with integrated neural processing units (NPUs) capable of 40 TOPS or higher—enough to run substantial on-device models for low-latency agent invocation. Qualcomm, MediaTek, and even Intel are reportedly engaged in developing optimized silicon for Solara-certified devices. These machines will likely come in familiar form factors: ruggedized tablets, conference room controllers, warehouse scanners, and headless compute modules for industrial IoT.

Chip-to-cloud: a sealed trust chain

The phrase "chip-to-cloud" is not marketing fluff for Project Solara. Every device will ship with a hardware root of trust that ties directly to Azure Active Directory (now Microsoft Entra ID) and Intune for zero-touch provisioning, policy enforcement, and attestation. The moment a Solara device powers on, it establishes a cryptographically verified identity, pulls down its configuration from the cloud, and deploys the appropriate agents based on the user's role, location, and security posture.

This end-to-end integrity model borrows from Microsoft's Secure-core PC initiative but extends it to a custom Android-based platform. The Verified Boot process will check the integrity of the kernel and the agent runtime before any user interaction is allowed. If a device falls out of compliance—say, a firmware downgrade is detected—conditional access policies immediately revoke its network access and token refresh rights. For IT administrators, this means a single pane of glass in Intune can manage Solara devices alongside Windows PCs, iPhones, and Linux servers, with identical compliance and configuration profiles.

Moreover, the agent-centric architecture reduces the attack surface dramatically. There is no web browser, no email client, no office suite in the traditional sense. Instead, all data processing occurs through curated agent interactions that are logged, auditable, and often restricted to server-side manipulation. The device becomes a glorified input/output terminal with a flicker of local intelligence.

Farewell apps, hello agents

The most radical shift in Project Solara is the abandonment of the app paradigm. Users do not install executable binaries; they subscribe to agents. An agent in this context is a trusted, AI-driven digital worker that understands the user's intent, accesses relevant cloud data, and presents results in a conversational or structured interface. For example, a warehouse associate might have a "Inventory Reconciliation Agent" that listens for voice commands, scans barcodes using the built-in camera, updates Dynamics 365 in real-time, and alerts a supervisor if discrepancies exceed thresholds.

These agents are built with Microsoft Copilot Studio, Azure AI Foundry, or third-party frameworks and deployed via the Solara Agent Catalog—a curated marketplace that replaces the traditional app store. IT administrators control which agents are available to which device groups, enforce data loss prevention (DLP) rules, and set usage quotas. Agents can run locally on the device's NPU for sub-second response, or seamlessly burst to the cloud when heavier reasoning is required, using a lightweight execution runtime called AgentOS.

AgentOS is the secret sauce. It manages agent lifecycle, memory, context switching, and permission boundaries. Unlike Android's sandboxed app model, AgentOS provisions a shared semantic workspace where agents can collaborate. An HR onboarding agent, for instance, can hand off a new hire's details to a facilities agent that books a desk and a security agent that provisions a badge—all without the user switching contexts or granting explicit permissions. The system resolves intent conflicts through an arbitration layer rooted in enterprise policy, not user whims.

Critics will rightly ask: what about offline scenarios? Microsoft acknowledges that Project Solara is designed for connected spaces. However, the platform supports a "graceful degradation" mode where critical agents cache essential models and data, allowing basic workflows to continue when Wi-Fi or 5G drops. This is expected to cover about 80% of frontline use cases, particularly in environments like factory floors where intermittent connectivity is the norm.

Developer ecosystem: building for agent-first hardware

For developers, Project Solara represents both an opportunity and a mindset change. Instead of writing Android activities or Windows Presentation Foundation (WPF) interfaces, teams will design declarative agent manifests that define the agent's capabilities, required scopes, and UI patterns. The UI itself is rendered through Adaptive Cards, which Microsoft has been honing across Teams, Outlook, and Power Platform for years. This means a Solara agent's interface automatically adapts to the device's screen size, input modality (touch, voice, pen), and accessibility settings.

Microsoft is betting that the Copilot ecosystem's maturity by 2026 will make this transition palatable. Hundreds of thousands of developers are already familiar with Copilot Studio for crafting conversational agents. With Solara, those agents gain first-class hardware access: cameras, geolocation, NFC, biometrics, and even custom USB peripherals via a safe agent-to-peripheral bridge. This opens doors for domain-specific agents that, for instance, control medical devices in a hospital or calibration equipment on an assembly line.

Monetization will follow a familiar SaaS model. Agents can be offered on a per-user, per-device, or consumption basis, with Microsoft taking a 30% cut of revenue—mirroring the traditional app store model but with recurring subscription potential. Early partners, as teased at Build, include SAP, ServiceNow, Workday, and several line-of-business software vendors that see value in pre-packaged agents for industries like logistics, retail, and healthcare.

Enterprise readiness and the Intune/Entra tandem

No enterprise endpoint stands a chance without robust management, and Project Solara is deeply enmeshed in the Microsoft Endpoint Manager stack. From the Intune admin center, IT can define agent assignment policies that are scoped to Azure AD groups, geographic regions, or compliance states. Enrollment is expected to be fully automated via Windows Autopilot-like self-deploying profiles, where the device arrives at the user's desk pre-provisioned with its identity and core agent set.

Conditional access extends to agents themselves. A risk score calculated by Microsoft Defender for Endpoint (which will also support Solara) can influence whether a high-privilege agent like a financial approval bot is launched at all. If the device's health attestation fails—maybe a rogue kernel module is detected—that agent simply won't appear in the user's launcher. All session data is encrypted at rest using Azure-managed keys, and transient caches are wiped on lock or sign-out.

Data residency and sovereignty concerns are addressed through Azure's global infrastructure. Organizations can configure their Solara fleets to process agent telemetry, transcripts, and generative responses within specific geographic boundaries, complying with regulations like GDPR, HIPAA, or the EU AI Act. Microsoft has also committed to providing a detailed bill of materials for each agent's AI models, allowing security teams to audit the provenance of training data and guard against supply-chain attacks.

Competition and the Windows question

Project Solara inevitably invites comparisons with Chrome OS. Google's lightweight platform has carved out a significant niche in education and, increasingly, in call centers and kiosks. However, Solara's agent-first approach sets it apart. Chromebooks run web apps and Android apps; Solara runs none of those natively. It is not a general-purpose device. It is a purpose-built endpoint that gets out of the way, turning work intent into cloud-facilitated action faster than any file-explorer-and-mouse paradigm could.

What does this mean for Windows? Microsoft has no intention of replacing Windows on the desktop. Instead, Solara complements the portfolio, addressing segments where a full Windows license and its associated overhead are overkill. Think of the barista swiveling a tablet to show a QR code, the nursing assistant roaming hospital corridors, or the inventory clerk wearing a ruggedized wrist-mounted Solara device. These roles don't need Excel; they need task-specific agents delivered securely.

Windows will continue to serve knowledge workers who require rich multitasking, creative tools, and hardware flexibility. Over time, edge AI capabilities will converge: Windows PCs will gain neural processors and Copilot experiences that borrow from Solara's agent model, but the underlying OS will remain. Solara is a parallel track, not a replacement.

Challenges and the road ahead

Despite the fanfare, Project Solara faces an uphill climb. Enterprise adoption of new device categories is notoriously slow. IT departments are hesitant to add yet another OS to their management portfolio, even if Intune promises a unified view. Compatibility with existing line-of-business apps is another hurdle. Microsoft's answer—that agents will replace those apps—is compelling long-term but requires vendors to re-platform decades of logic into agent-oriented services.

Pricing remains a mystery. Microsoft has not disclosed whether Solara devices will carry a per-device license, a per-user subscription, or fall under existing Microsoft 365 agreements. Cost could be a deciding factor. If Solara undercuts even the cheapest Chromebook Enterprise with a compelling ROI story—fewer management hours, lower TCO, improved task accuracy—it could find a foothold. The Build demo showed a Solara-powered inventory drone reducing stocktaking from days to hours, but such bespoke examples may not scale horizontally.

Skepticism around AI reliability also lingers. Agents hallucinate. They misinterpret context. In regulated industries, a wrong output from a Solara agent could have legal consequences. Microsoft's response hinges on grounding AI outputs in enterprise data using techniques like Retriever-Augmented Generation (RAG) and requiring a human-in-the-loop for high-stakes actions. The platform enforces a "confidence threshold" setting that admins can tune. Below that threshold, the agent won't act and will instead escalate to a human supervisor.

Then there's the Google elephant in the room. AOSP's freedom means Microsoft can fork it, but Google could still steer the underlying platform in directions that make Solara's maintenance costly. Microsoft's experience with the Surface Duo's Android flavor suggests it can manage this tension, but a fleet of enterprise devices watched over by security auditors is a different challenge. The company's heavy investment in its own silicon initiative could eventually wean Solara off ARM's standard designs, giving Microsoft full control from fabrication onward.

What's next and how to prepare

Project Solara is not a ship-it-and-forget-it product. Microsoft announced a public preview for early 2027, with general availability expected in late 2027 or early 2028. Early adopter programs will open in fall 2026, inviting select enterprises and ISVs to co-develop agents and feedback on the hardware reference designs. Build attendees were able to handle a non-functional prototype of a Solara tablet, described as "surprisingly solid, with a 10-inch 4:3 display and industrial-grade rubber bumpers."

For enterprise architects, now is the time to evaluate which frontline workloads could benefit from an agent-first model. Start by inventorying paper-based or app-hopping processes that involve data entry, approvals, and status checks. Map those to potential agent APIs. Begin experimenting with Copilot Studio to understand how declarative AI integrates with your existing systems. When Solara hardware becomes available in preview, you'll already have the agent logic ready to deploy.

Stay tuned for updates through the Microsoft 365 roadmap and the Endpoint Manager blog. Project Solara may seem like science fiction today, but it represents a deliberate, multi-year bet on a future where endpoints are defined not by their apps, but by the intelligence they channel. Microsoft is betting that the agent, not the app, will become the atomic unit of enterprise computing. If it's right, the traditional desktop may finally have a worthy heir.