Meta is testing a “human concierge” layer for Muse, its new personal AI agent, Reuters reported exclusively after reviewing internal company posts. In the experiment, contractors quietly handle some phone calls that users believe they are placing through the digital assistant.
Muse is built on Meta’s own models and is meant to compete with ChatGPT and Claude. Sensor Tower figures cited in trade coverage put early iOS traction in the millions of downloads, including a stretch when it overtook ChatGPT on the charts after a fast first week.
The concierge test is a classic Wizard-of-Oz pattern: ship the agent interface, then fill reliability gaps with people. It also creates a disclosure and privacy problem. Users talking to “an AI” may not expect a contractor on the line, especially as Meta faces broader pushback over glasses and always-on assistants.
Mark Zuckerberg is expected to show more AI-era hardware and software this week. The Reuters exclusive suggests the company still needs humans in the loop for real-world tasks — phone calls included — even as it markets Muse as a personal agent.
Key takeaway. Muse’s early growth is real, but Meta is still routing some calls to human contractors — a reminder that “agentic” products often hide people behind the model.
Photo: Unsplash. Sources: Reuters exclusive via BNN Bloomberg; Economic Times; SiliconReport / Sensor Tower figures (Sept. 22–23, 2026).
