A journalist’s first-person account published Oct. 9 describes a one-week test of Meta Muse and OpenAI dots during an apartment move. Muse helped her find a dining table through Facebook Marketplace; after she gave dots her Facebook login, Meta flagged the agent as a bot and temporarily locked her out of Instagram, Facebook and WhatsApp. The test also included useful background monitoring—and a truck booking she later changed.
AI agents are designed to handle or monitor tasks, rather than just answer a prompt. These results come from one journalist’s personal test, not NeoTeo’s hands-on testing.
Marketplace help, account access and shopping suggestions
Muse searched Facebook Marketplace, drafted messages to sellers for the journalist’s approval, managed their replies and helped her find a dining table. It also monitored concert listings for artists in her Spotify library.
The Facebook login attempt had a different outcome. After the journalist gave dots her Facebook credentials to search Marketplace, Meta flagged dots as a bot and temporarily locked her out of Instagram, Facebook and WhatsApp. In her setup, Muse had access to email, bank accounts, social media and Spotify; dots had more trouble accessing accounts and said it could not access some, including bank accounts.
A $50 spending prompt brought out a different contrast. Dots suggested a used book the journalist had previously asked it to find. Muse suggested a concert by Jess Williamson, based on a song saved in Spotify, and repeatedly recommended a Target membership and lamp. The journalist says that Target suggestion appeared seven times in three days.
A car-wash prompt, a truck booking and background tasks
Asked what to do about a car wash 100 feet away, dots advised driving so the car could be washed; Muse recommended walking. It was one prompt in this test, not a measure of either agent’s general reasoning.
For the move, both agents found a farther U-Haul location with 15-foot trucks, and dots booked one. The next morning, the journalist learned that 10-foot trucks were available for walk-ins and switched to one. The booking completed a task, but the journalist’s later change shows why a booking can still leave a practical decision to the person who made it.
Dots also surfaced an email that had landed in spam and a form the journalist had forgotten to complete. Those examples, along with Muse’s concert monitoring, show the kind of background assistance she used during the week.
What a one-week test can tell us
The examples make the trade-off concrete: an agent may save steps on a search or reminder, while recommendations and bookings can still need a person’s judgment. Muse’s repeated Target suggestion and the truck change were outcomes in this journalist’s setup; they do not establish a general pattern across users.
A separate early comparison includes screen recordings of both interfaces and individual observations about features and limitations. It adds visual context, but the one-person test and an early comparison do not establish which agent is more reliable across users or tasks.
