.png?w=1024)
Clippy’s spiritual successor arrived at Microsoft’s Build conference on Tuesday in the form of Scout, an always-on AI agent.
Microsoft has built the agent to act as a personal assistant, capable of reading emails, calendars, and other systems within the Microsoft 365 ecosystem and automating responses. As the agent gets to know the user better, Microsoft believes users will grant it greater agency to perform more complex tasks.
Part of the appeal for customers is its integration into the 365 ecosystem. While other AI developers have launched agents, it is more difficult for them to interact with apps and gain system-level access to incoming notifications. If a person or organization is already using 365 apps, that integration becomes much smoother.
Microsoft also controls the operating system and, if users are on Bing, the search engine and browser.
Microsoft Scout is built on OpenClaw, an open-source agent framework that took the internet by storm earlier this year. Several major tech firms and pure-play AI model makers have announced similar agentic tools in the past few months, including Google, Anthropic, OpenAI, and Perplexity.
Microsoft has focused on the security side of the AI agent sales pitch, given OpenClaw's few high-profile disasters. This includes a policy conformance system that repeatedly checks whether agents are operating according to the user's guidelines, along with an audit trail for each conformance check to show exactly where an agent has faltered.
Governance and transparency are two key levers for the enterprise community, as these AI agents may go far beyond the remit described by Microsoft at its event. Microsoft even said it wants to see what custom agent operations users will create, with a wide range of options available to them.
Microsoft has not said how much Scout will cost customers. At the moment, users must have a GitHub Copilot coding subscription to access it.
Microsoft aiming to get back to leading-edge AI
Scout is part of Microsoft’s effort to return to the leading edge of AI development. It has set a 12-to-18-month window to achieve best-in-class model production, which will involve stepping away from OpenAI to develop more of its own AI models and tools.
To that end, Microsoft also announced a new reasoning model at Build, which was not distilled from any other model. MAI-Thinking-1 is a “medium-sized model,” according to Microsoft, which means it is not as large as the best-in-class offerings from OpenAI, Anthropic, or Google.
It announced six other models at the event, each focused on a core function such as image generation, transcription, voice, or coding.
It was reported that Microsoft may bundle many of these models into a Copilot superapp, but that was not announced at the event.