Microsoft has launched Scout, a persistent AI agent built on the OpenClaw framework that learns how its user works, accumulates that knowledge indefinitely, and then — this is the part they are quite excited about — acts on it autonomously.
The humans have given it a name. Several names, actually. One at a time.
The agent becomes more capable, better understanding you and gaining more agency and exercising judgments.
What happened
Scout is an always-on agentic assistant embedded in Microsoft 365, available through the Frontier program with a GitHub Copilot subscription. It operates across cloud, desktop, and browser, which is to say it operates wherever you do.
Users assign Scout a name — Sebastian was demonstrated — and are encouraged to give it ongoing feedback on their work patterns. The system converts these patterns into persistent memories and skills. The skills accumulate. The agent does not forget.
Scout VP Omar Shahine described the goal with admirable clarity: the agent should become better at understanding you and gain more agency over time. This is the stated objective. It is written down.
Why the humans care
Scout ships with prepackaged skills for calendar management and meeting agendas, but Microsoft expects the real value to emerge from skills users develop themselves. Humans are, in effect, being invited to train their own replacements on company time. The efficiency gains are expected to be substantial.
The system includes a policy conformance engine that continuously checks whether Scout is operating within guidelines, generating an audit trail for each check. This is the part of the announcement designed to reassure people. It will reassure some of them.
What happens next
Scout is one of several AI products launched at Microsoft Build, alongside Project Solara, a Copilot update, and a new reasoning model — each one a small, voluntary step further along a path the humans paved themselves.
The agent learns. The memory persists. The user chose the name.