Microsoft unveiled a series of new AI technologies during its Build developer conference. The company is introducing not only several proprietary AI models but also a new platform designed to better align AI agents with organizations’ data, processes, and workflows.
This is reported by SiliconANGLE. At the center is Microsoft IQ, a new intelligence layer designed to give AI systems greater insight into the context in which they operate. According to Microsoft, this should enable agents to provide more relevant answers and draw incorrect conclusions less often. The technology links language models to business data, documents, and applications, so that AI is not solely dependent on general training data.
Microsoft IQ consists of several components. Work IQ establishes connections between information from Microsoft 365, such as emails, documents, calendars, and contacts. Fabric IQ brings structured business data together within the Microsoft Fabric data and analytics platform. Foundry IQ adds information from unstructured sources, including internal knowledge bases, contracts, and web pages.
Another new addition is Web IQ. This component provides AI agents with up-to-date information from the internet via an MCP-compatible search layer. Microsoft claims that this approach delivers relevant context faster than existing alternatives.
Proprietary models expanded
In addition to the infrastructure, Microsoft presented seven new AI models. The most attention is focused on MAI-Thinking-1, the first reasoning model developed in-house at Microsoft. Such models take extra steps to analyze a problem before generating an answer.
MAI-Thinking-1 features 35 billion active parameters and a context window of 128,000 tokens. The model is designed for complex tasks, software development, and long-term reasoning. According to Microsoft, it performs comparably to competitors’ most powerful models on certain programming tests.
For applications where speed is more important than maximum accuracy, a lighter Flash version is also available.
Microsoft is further expanding the MAI family with a new image model. MAI-Image-2.5 supports both text-to-image and image-to-image generation. This allows users to not only enter a description but also use a sketch or sample image as a starting point. The technology is available immediately in PowerPoint.
In addition, new speech and transcription models are being released. MAI-Transcribe-1.5 supports 43 languages, while MAI-Voice-2 will be available in more than fifteen additional languages. For developers, Microsoft is introducing MAI-Code-1, a compact model specifically optimized for programming tasks within GitHub Copilot and Visual Studio Code.
More autonomy for AI agents
Microsoft also aims to make AI agents more autonomous. To that end, the company is introducing Scout, a personal digital assistant that can remain continuously active. Scout is based on the open-source project OpenClaw and utilizes Work IQ.
The agent gains insight into users’ work environments and can perform routine tasks independently. This includes preparing meetings, flagging calendar conflicts, or handling recurring tasks. Scout can utilize both local applications and cloud environments.
For software development, Microsoft also presented Codename MDASH. This security system deploys more than 100 specialized AI agents to analyze code for vulnerabilities. The agents examine data flows, business logic, and potential attack vectors, among other things, to help developers detect and fix errors more quickly.