Microsoft used its Build 2026 developer conference to lay out a broad update to its computing stack, spanning Windows, AI models, hardware, and experimental platforms. The announcements reflect a shift toward systems designed to run AI workloads locally, support agent-driven workflows, and operate more seamlessly across devices and cloud environments.
Rather than focusing on a single product, the company’s updates point to a wider transition in how software is built and used, with AI moving from a feature to a foundational layer across the Windows ecosystem.
Windows moves closer to an AI platform
A significant portion of the announcements focused on Windows, which Microsoft is increasingly positioning as a platform for AI workloads rather than just a desktop operating system.
The company introduced several features aimed at simplifying how applications are built and run across devices, including support for Linux containers through Windows Subsystem for Linux (WSL), new configuration tools to set up development environments, and an Intelligent Terminal that integrates AI assistance directly into command-line workflows.
At the same time, Microsoft expanded its Windows AI APIs beyond devices with dedicated AI hardware, allowing more PCs to run AI-powered features such as speech recognition and video enhancement locally.
While these updates are developer-focused, they signal a broader shift: Windows is being adapted to support AI workloads natively, rather than relying entirely on cloud-based processing.
Microsoft also outlined a set of security features designed to support the deployment of AI agents on Windows, as these systems begin to operate with greater autonomy.
These include:
-
Agent containment through execution containers -
Identity management for AI agents -
Integration with Microsoft Entra and Intune -
Security protections through Microsoft Defender and Purview -
Support for Windows 365-based agent environments
Microsoft introduces Scout AI agent
Microsoft also introduced Scout, an always-on AI agent integrated across Microsoft 365 services including Teams, Outlook, OneDrive, and SharePoint.
Unlike traditional assistants that require explicit prompts, Scout is designed to operate continuously in the background, using information from emails, calendars, chats, and documents to understand user context and assist with everyday tasks.
According to Microsoft, Scout can help prepare for meetings, manage scheduling conflicts, draft emails, and surface relevant information without requiring manual input.
The company positions Scout as part of a new category of AI systems it calls “autopilots”, which are designed to execute tasks on behalf of users rather than simply respond to queries.
Microsoft expands its MAI model family
Alongside the new AI assistant, Microsoft introduced seven new first-party AI models under its Microsoft AI (MAI) family, expanding its in-house capabilities across text, code, image, voice, and speech.
Available through Microsoft Foundry, the models include:
-
MAI Image-2.5: An image generation model designed to create high-quality visuals from text prompts -
MAI Image-2.5 Flash: A faster, lower-cost version aimed at rapid image generation -
MAI Transcribe-1.5: A speech-to-text model for converting audio into written content -
MAI Thinking-1: A reasoning model designed for multi-step problem-solving tasks -
MAI Voice-2: A text-to-speech model focused on generating natural-sounding voices -
MAI Voice-2 Flash: A lower-latency version for real-time voice applications -
MAI Code-1 Flash: A coding-focused model for code generation and completion
New Aion models bring AI to the device
Microsoft is also pushing AI capabilities directly onto devices through its new Aion model family.
The Aion lineup includes small language models designed to handle lightweight AI workloads locally on Windows systems, reducing reliance on cloud processing.
The models include:
-
Aion 1.0 Instruct, built for Windows 11 to support on-device tasks such as text summarisation, rewriting, and accessibility features -
Aion 1.0 Plan, a 14-billion-parameter reasoning and tool-calling model designed to support more complex, agent-driven workflows even in offline environments
By enabling these capabilities locally, Microsoft is positioning Windows devices to run AI workloads without constant connectivity, addressing latency, privacy, and cost constraints associated with cloud-based models.
Project Solara points to an agent-first future
Beyond incremental updates, Microsoft used Build to showcase its longer-term vision through Project Solara, a platform designed for “agent-first devices”.
Unlike traditional systems that rely on apps, Solara is built around AI agents that interpret user intent and execute tasks across services. Instead of navigating multiple applications, users interact with agents that coordinate workflows in the background.
The platform also introduces a concept called “just-in-time UI”, where interfaces are generated dynamically based on context rather than being pre-designed for each device.
Microsoft demonstrated this approach through prototype devices, including a wearable badge and a desk-based system, both of which rely entirely on agents rather than traditional applications.
The idea reflects a broader shift across the industry, where companies are exploring ways to move beyond app-based computing toward more adaptive, task-driven systems.
Hardware built for AI workloads
To support this shift, Microsoft also introduced new hardware platforms designed specifically for AI development.
The Surface RTX Spark Dev Box is positioned as a compact system capable of handling AI workloads locally, while the DGX Station for Windows, developed in partnership with NVIDIA, is designed to run large AI models, including those with hundreds of billions of parameters.
Majorana 2 advances quantum computing efforts
Alongside its AI announcements, Microsoft also shared progress in quantum computing with its Majorana 2 chip.
According to the company, the new chip introduces a revised materials approach that significantly improves qubit reliability compared to its previous design. Microsoft said the development could accelerate its roadmap toward building a scalable quantum computer, with a target timeline of 2029.
The company also noted that its AI systems played a role in developing the chip, highlighting how machine learning is being used not just in software, but in scientific research and hardware design.
