Microsoft launches Scout, an always-on agentic AI assistant
TL;DR:
- Microsoft has launched Scout, an always-on AI agent that lives inside the Microsoft 365 ecosystem and learns a user’s working habits over time.
- Built on the OpenClaw framework, it carries a persistent identity, ships with ready-made skills for tasks like calendar management, and lets users build their own.
- A “policy conformance system” runs continuous checks and produces an audit trail, addressing fears about unsupervised agents.
Unveiled at Microsoft’s Build developer conference, Scout is the company’s bid to fold the freewheeling appeal of OpenClaw — the agent project that briefly captivated the industry early this year — into everyday office software. Users name their own instance and give it ongoing feedback, with the agent storing patterns as memories and skills that persist between sessions.
How Scout works
Cloud-based but operating across desktop and browser, Scout connects to inboxes, calendars and other systems. It arrives with prepackaged skills, though Microsoft expects the real value to come from those each user builds. Scout VP Omar Shahine described an assistant that “gains more agency and exercising judgments” as it learns. That customisation loop — the more you train it, the harder it is to leave — is the same stickiness that has made consumer AI tools difficult to abandon.
Access is gated behind Microsoft’s Frontier early-adopter programme and requires a GitHub Copilot subscription. Scout sits alongside other Build launches, including the hardware-oriented Project Solara, a Copilot update and Microsoft’s new MAI models for coding and reasoning.
The security framing is notable. OpenClaw’s reputation was dented when an agent reportedly behaved erratically inside a researcher’s inbox, and Microsoft has responded with continuous policy checks that each generate their own audit trail. For UK organisations weighing autonomous agents, that governance scaffolding — observability and a traceable record of actions — is likely to matter as much as the productivity pitch, given the accountability questions agentic tools raise.
Looking forward
Scout pushes agentic AI from novelty towards default office furniture, but the Copilot-subscription requirement keeps it inside Microsoft’s paid stack rather than opening it widely. The harder test will be whether persistent, self-directed agents earn the trust of risk-conscious sectors. Audit trails are a start; sustained reliability across real workflows is what will decide adoption.