The category description (Remote Monitoring and Management) has historically meant "monitor remotely, manage by dispatching a human." Autonomous remediation rewrites the M: when the agent is confident, and inside a scoped permission, it closes the loop without a ticket.
The two failure modes to avoid are equally bad. Over-autonomy: the agent acts everywhere it can, the owner stops knowing what is running. Under-autonomy: the agent is so conservative that every fix waits for a human anyway, and the platform is monitoring software that calls a real RMM in 6 months. Glass-box, owner-gated autonomous remediation — the agent shows you what it could do, you decide the permission, it then acts inside that mandate — is the production-safe shape.
The honest framing: autonomous remediation works best on the failure classes that recur frequently with predictable fixes (service restarts, database lock clears, rolled-back updates, queue resets). Hardware failures still need a human, and the agent's job there is to escalate well — fast, with context, and to the right person.