Microsoft today at its Build 2026 conference launched a slew of artificial intelligence (AI) capabilities and platforms, including a family of seven “clean” models built using licensed content that can be customized using a frontier tuning tool that Microsoft has developed.
The family is based on a 35 billion active parameter model built for high efficiency and performance that consumes tokens more efficiently than rival models. Microsoft CEO Satya Nadella told conference attendees the time has come for organizations to move beyond consuming frontier AI models to building and customizing them. “We believe the time has come for moving from consuming to participating in development of frontier models,” he said.
Microsoft also launched an ambitious Project Solara initiative that enables specialized AI agents to run on a range of new classes of devices and previewed a runtime, dubbed Microsoft Execution Containers (MXC), that allows developers and IT administrators to describe agent containment requirements that can be used to safely deploy AI agents such as OpenClaw that is now available in alpha on the Microsoft platform. NVIDIA is collaborating with Microsoft to bring the NVIDIA OpenShell runtime to Windows to run any AI agent safely.
At the same time, Microsoft also previewed Scout, the first in a series of long-running autonomous AI agents that will have access to enough memory to automate tasks on behalf of organizations running workloads on the Microsoft Azure cloud.
Microsoft also pledged to make Hosted Agents in Foundry Agent Service generally available in the coming weeks, providing per-session sandboxing for untrusted code, sub-100 ms cold starts and zero idle cost in an agnostic runtime environment that includes support for OpenTelemetry to collect observability data.
At the core of that capability is Microsoft Agent Framework (MAF) 1.0, a harness for AI agents that provides access to skills, context, memory and middleware. Fireworks AI on Foundry, now generally available, also provides access to open-source models that can be integrated with a Managed Compute service that Microsoft is making available via a private preview to add additional graphical processor unit (GPU) capacity.
Microsoft also launched Web IQ, a set of application programming interfaces (APIs) that gives AI agents access to a knowledge graph through which they can discover, rank, extract and package relevant information from web pages, news, images and video data sources. Microsoft IQ, which packages Work IQ, Fabric IQ and Foundry IQ in a way that is more accessible, is now also generally available. “If you structure the context right by definition you will be more token efficient,” says Nadella.
An update to Foundry IQ, now generally available, creates a knowledge base to unify Work IQ, Fabric IQ, File Search, Azure SQL and MCP into a single retrieval endpoint. Microsoft World Grounding extends that reach to live external information alongside internal data. A preview of a procedural memory capability will allow AI to learn the “how” a process needs to occur across multiple runs.
There is also a bevy of new tools aimed specifically at developers. Microsoft is previewing a GitHub Copilot application that provides developers with access to a set of AI agents to build and test code and a set of Windows Development Configurations that makes it simpler to set up a development environment.
There is also now an open source Rayfin software development kit (SDK) and command line interface (CLI) through which developers can describe the application they want to build to invoke an AI agent, built in collaboration with Replit, to generate a backend database and support for Postgres databases on the Horizon service Microsoft provides.
Microsoft, in effect, is creating its own composable integration plane that feels unified to the developer, says Mitch Ashley, vice president and practice lead for software lifecycle engineering at the Futurum Group. “The connective tissue between layers — the agent control plane that spans Windows MXC, Agent 365 governance, Foundry hosted runtime, and the IQ context layer – was positioned as Microsoft’s durable strategic asset,” he says.
Finally, Microsoft is extending Windows application programming interfaces (APIs) to support AI applications, including those running on a new class of PCs being developed in collaboration with NVIDIA, and is making available a Linux-like command-line utility, dubbed Coreutils, to Windows.
As the pace of innovation continues to accelerate in the age of AI, the one thing that is clear is that new capabilities are being added at a dizzying pace. The one thing they all seem to have in common is they provide organizations with more control over their own AI destiny.
