# Microsoft Unveils Project Solara, Agent-First Platform > Source: > Published: 2026-06-14 12:43:32.854841+00:00 # Microsoft Unveils Project Solara, Agent-First Platform Reporting by Reuters and GeekWire describes Microsoft unveiling **Project Solara**, a chip-to-cloud platform for "agent-first" devices that run AI agents instead of traditional apps or operating systems. Reuters reports prototypes include small, screen- and microphone-equipped devices built with **Qualcomm** and **MediaTek** chips and that Solara devices will connect to cloud models to carry out tasks such as documenting medical visits. Forbes reports Microsoft also revealed a family of in-house models under the **MAI** brand, including MAI-Thinking-1 with a sparse Mixture-of-Experts design and roughly **35 billion active parameters** and a **256,000-token** context window. WindowsCentral quotes Satya Nadella saying, "You need to give them identities," in reference to AI agents. Industry outlets frame Solara as Microsoft's bid to create hardware, silicon, and runtime for agentic workflows rather than conventional apps. ### What happened Reporting by Reuters and GeekWire describes Microsoft unveiling **Project Solara** at the Build 2026 developer conference, a chip-to-cloud platform intended to host AI agents that operate across apps, screens, and workflows rather than through individual applications. Reuters reports prototypes include devices the size of a smart speaker or keycard badge built with **Qualcomm** and **MediaTek** chips that rely on cloud connectivity to carry out tasks such as documenting medical visits. GeekWire reports the platform is based on Android components and includes two working hardware designs and pilot partners. Forbes reports Microsoft also introduced seven in-house models under the **MAI** brand, including MAI-Thinking-1, which Forbes describes as a sparse mixture-of-experts reasoning model with roughly **35 billion active parameters** and a **256,000-token** context window, currently in private preview via Microsoft Foundry. WindowsCentral publishes a direct quote from Satya Nadella: "You need to give them identities," referring to governance and policy for agents. ### Technical details Forbes reports MAI-Thinking-1 and other MAI models were trained from scratch on licensed data, with Microsoft saying the training pipeline and custom inference accelerator are co-designed to improve efficiency. Reuters and Qualcomm materials describe Project Solara as a chip-to-cloud stack that combines on-device silicon, an Agent Shell runtime that dynamically loads cloud-hosted agents, and Azure cloud services for model execution and orchestration. Reporting describes three integration levels for AI in user workflows: sidebar helpers, in-app integration, and agents operating outside apps to streamline cross-app workflows; Solara is framed as targeting the third level. ### Industry context Editorial analysis: Companies that push an "agent-first" thesis commonly pursue three engineering pillars simultaneously-on-device accelerators for latency and privacy, a runtime to manage agent lifecycle and security, and cloud-hosted models for heavy reasoning. Microsoft's public demonstrations and the MAI model disclosures align with that pattern and mirror concurrent efforts from other major vendors to control more of the stack (silicon, models, runtime). For practitioners, this pattern raises engineering trade-offs between on-device efficiency, cloud orchestration complexity, and compliance boundaries when agents act autonomously across data sources. ### Context and significance Editorial analysis: Microsoft's public rollout bundles hardware prototypes, a new runtime concept, and in-house models, which-if adopted-could shift developer attention from app-centric UI flows to agent orchestration and policy. The combination of dedicated inference accelerators and models with very large context windows (Forbes' **256,000-token** claim for MAI-Thinking-1) points to use cases requiring long-context state and multi-step reasoning, such as multi-visit medical documentation or complex enterprise workflows. The public quote from Satya Nadella about giving agents "identities" underscores that Microsoft is foregrounding governance, sandboxing, and policy as part of the runtime conversation, according to WindowsCentral. ### What to watch For practitioners: track three measurable indicators-first, whether Microsoft releases SDKs or emulators for the Agent Shell runtime and the timelines for developer access; second, independent evaluations of the MAI models versus competing reasoning models (the Forbes piece cites internal ratings and awaits outside testing); third, pilot results from device partners and latency/privacy metrics when agents combine on-device compute with cloud reasoning. Observers should also watch for concrete security and policy primitives that implement the "identity" and sandboxing ideas Nadella mentioned, since those will determine practical trust and compliance constraints for agentic deployments. ### Bottom line Editorial analysis: Project Solara represents a coordinated push to reframe user interaction around persistent, cloud-backed agents running on a new class of devices. The technical claims in Forbes and the hardware prototypes reported by Reuters make this an important signal for architects building long-context, multi-step AI integrations, but independent benchmarks, SDK availability, and policy mechanisms will determine whether the agent-first model becomes a mainstream developer platform. ## Scoring Rationale Major vendor (Microsoft) unveiling a coordinated hardware, runtime, and model stack for agentic workflows is notable for architects and platform teams. The score reflects substantial product and technical claims but also depends on independent benchmarks, SDK availability, and real-world pilots. Practice with real Ride-Hailing data 90 SQL & Python problems · 15 industry datasets 250 free problems · No credit card [See all Ride-Hailing problems](/problems/datasets/mobility)