Meta has launched the next stage of its agentic AI push: the Muse Wonder series. The new models add processing power to its existing Muse personal AI so it can take on longer and more complex tasks.
Meta introduced Muse Wonder and Muse Wonder Full as "our most capable models that deliver state-of-the-art performance on personal and professional agentic tasks."
"Muse Wonder brings together Active Memory and Active Focus to keep track of what matters to you and work diligently toward what you want to accomplish," Meta said. That allows the agent to handle tasks that take days to complete.
Meta says the models apply specialized knowledge to work and daily life, "from understanding a lease to making sense of financial or health information," with evaluations covering law, finance and health. The company also reported new high benchmark scores.
Extreme Tech reported that a Muse user in Toronto handed the full Facebook Marketplace sales process to his agent — not realizing that meant it would sell the item and give a buyer his home address. That was human error, not a system failure, but the concern is how easy it is to switch on such broad permissions.
Handling financial and health tasks means handing that data to Muse. Given Meta's history with data misuse, Social Media Today argued these agents may need a longer testing period before users can trust them.
As agents shop, sell and send inquiries on users' behalf, more of a brand's "customers" will be software. For Meta's work on agent payment standards, see Meta Is Building an Agentic Commerce Protocol for Muse.
Muse Wonder and Muse Wonder Full are Meta's most capable models for its Muse personal AI agent. Active Memory and Active Focus let them work on complex tasks that can take days.
Meta says it applies specialized knowledge to everyday and work problems such as understanding a lease or interpreting financial and health information, with evaluations in law, finance and health.
A Toronto user who delegated a Marketplace sale found the agent had shared his home address with a buyer. Granting permissions in stages and reviewing sensitive steps helps reduce risk.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
Buzzvil's BD strategy team has zero developers but runs 32 AI workflow automations — from 7-minute revenue drop analysis to 90% fewer manual steps. How they scaled from personal hacks to team systems.
Pew cloned thousands of panelists as AI digital twins and found a 12.4-point average error. Five ways synthetic respondents fail — and where marketers can still use them in research.
As every brand mass-produces AI content, individual brands get harder to see. Pion Corporation CEO Jung Bum-jin explains why marketing AI must be built as an operating system, with real cases.