07/10/2026
Meta launched Muse this month: an AI agent that, in their words, "doesn't just answer questions, it actually does the work." Emails, forms, bookings. It's US-only for now.
The bit worth paying attention to isn't the agent. It's the second one. Meta runs a separate agent, the Sentinel, and nothing Muse does reaches the internet unless it approves it. It checks with the person before anything sensitive, like sending an email or buying something.
That's the shape that makes AI usable in a real business. The machine does the task. Something with judgement signs it off.
In mine, that something is me. Pulse drafts the reply. I decide if it's sent.
AI isn't the risky part. AI with nobody checking is.