Microsoft Scout (OpenClaw-inspired assistant)
microsoft‘s cloud AI assistant (launched at Build 2026), built directly on the OpenClaw framework and integrated with Microsoft 365. The clearest sign yet that the self-driven / autonomous-assistant pole of claude-code-channels-vs-openclaw is going mainstream — a big vendor productizing OpenClaw’s experimental model.
What it is
An “always-on agentic assistant” across desktop/web, connected to email/calendar/systems; users name their instance and shape it with ongoing feedback. The three OpenClaw inheritances:
- Autonomous operation — independently performs tasks (self-driven).
- Persistent identity — consistent personality/style across sessions (durable-agents).
- Learning & memory — develops skills/memories from user patterns, “increasingly capable over time” (self-improving-agents) — “harder to abandon as users invest in training it.”
Features & governance
Prebuilt skills (calendar management, agenda drafting) + user-developed custom skills (agent-skills). Notably ships a “policy conformance system” that continuously audits whether Scout stays within guidelines — directly addressing the unsupervised-agent reliability risk the wiki keeps flagging (governance as a first-class feature). Requires a GitHub Copilot subscription; via Microsoft’s Frontier early-access.
Why it matters
The OpenClaw lineage is now forked and productized: OpenClaw → hermes-agent (open-source successor) and → Microsoft Scout (enterprise product), with openhuman a third, local-first open-source entrant on the same self-driven pole. So the autonomous, durable, self-improving personal assistant is no longer fringe — it’s a mainstream vendor bet, arriving with an explicit governance/audit layer. microsoft now has two agent plays here (Scout + the Agent Framework of agents-that-build-agents-ms). (Caveat: launch coverage; early-access.)
Related
microsoft · claude-code-channels-vs-openclaw · hermes-agent · openhuman · durable-agents · self-improving-agents · agent-skills