Report: Meta Is Testing Human Contractors to Handle Calls for Its Muse AI Agent

Analytics India Magazine reports that Meta is testing whether human contractors can handle calls on behalf of its Muse AI agent. The story is labeled as a report, meaning it rests on media reporting rather than an official Meta announcement.

As described, the arrangement follows a human-in-the-loop model: the agent carries most of the interaction and hands off to a person when it cannot proceed or when judgment is required. Voice calls are among the hardest scenarios for conversational agents, since background noise, accents, interruptions, and multi-turn follow-ups all multiply errors.

Using outside contractors rather than an internal team is the most concrete detail in the report. It suggests the work is decomposable and can be trained at scale, and it also suggests the company is willing to trade labor cost for experience quality while agent capability remains uneven.

Meta has not publicly commented on the report, so the number of people involved, the regions covered, and the duration of the test cannot be confirmed. The product form and availability of the Muse agent itself also fall outside what the report describes.

The significance of the report is that it points to a realistic path for agent deployments: not one-step full automation, but automation first with human fallback. For products where voice is the main entry point, that hybrid model may reach an acceptable standard faster than chasing full autonomy.

What to watch is whether Meta responds, and whether the human fallback share falls as the models improve or hardens into a permanent part of the product structure. The same question applies to every vendor pushing voice agents into production.