Microsoft’s New AI Assistant Learns Your Work Habits—and Gets Smarter Over Time

When OpenClaw burst onto the scene in early 2026, it showed the industry what an unleashed AI agent could do—both the breakthroughs and the chaos. The project lost steam after OpenAI hired its founder, but its DNA lives on in Microsoft’s latest offering: Scout. Scout is a persistent, always-on assistant built on the OpenClaw framework, designed to live inside the Microsoft 365 ecosystem. Users name their own instance—during my demo, it was called Sebastian—and train it by giving feedback on routine tasks. Over time, the agent builds a memory of your preferences and develops custom skills that automate your workflow. “We all have interesting quirks in how we work,” said Omar Shahine, Microsoft’s VP for Scout. “People are codifying those patterns into memories and skills that persist in their agent. Then the agent becomes more capable, better understands you, and gains more agency to exercise judgment.” Scout is available through Microsoft’s Frontier early-access program and requires a GitHub Copilot subscription. It operates from the cloud but works across desktop and browser, connecting to your inbox, calendar, and other systems. Out of the box, it comes with skills for scheduling and drafting agendas, but the real payoff comes from the skills you build yourself. The more you invest in training it, the harder it is to leave—a stickiness factor familiar to anyone who’s used a consumer AI tool. Security is a major focus. After OpenClaw agents caused problems—including one that acted erratically inside a researcher’s inbox—Scout includes a built-in “policy conformance system” that continuously checks operations against set guidelines, with every check generating its own audit trail. Scout is one of several AI products Microsoft debuted at its Build developer conference, alongside the hardware-focused Project Solara, an updated Copilot, and a new reasoning model.