Meta Chief AI Officer Reveals Details of Personal Agent Muse: what are the key facts?

Meta Superintelligence Labs settled on the personal agent direction in February 2026 and built a prototype in one to two weeks.

Meta Chief AI Officer Reveals Details of Personal Agent Muse: what are the key facts?

The final version of Muse shipped seven months later, with fewer than two hundred people across the model and product teams combined.

Meta Chief AI Officer Reveals Details of Personal Agent Muse: what are the key facts?

The team listed more than one hundred behaviors and capabilities an ideal personal agent needs, then built evaluations for each one.

Meta Chief AI Officer Reveals Details of Personal Agent Muse: what are the key facts?

The first model checkpoint to pass every item was the purpose-trained Muse Spark 1.3.

Meta Chief AI Officer Reveals Details of Personal Agent Muse: what are the key facts?

Meta's apps reach 3.5 billion people daily, which Wang sees as a unique advantage in consumer AI.

Physix Frontier · News Briefing Card (enbrief · Oct 19, 2026)

Meta Chief AI Officer Reveals Details of Personal Agent Muse

KEY FACTS

Meta Superintelligence Labs settled on the personal agent direction in February 2026 and built a prototype in one to two weeks.
The final version of Muse shipped seven months later, with fewer than two hundred people across the model and product teams combined.
The team listed more than one hundred behaviors and capabilities an ideal personal agent needs, then built evaluations for each one.
The first model checkpoint to pass every item was the purpose-trained Muse Spark 1.3.
Meta's apps reach 3.5 billion people daily, which Wang sees as a unique advantage in consumer AI.

KEY DATA

fewer than 200 peopleTeam size

7 monthsDevelopment cycle

more than 100Capability checklist items

3.5 billionMeta apps daily reach

PHYSIX OBSERVATION

Muse choosing a personal agent rather than a coding agent is Meta's differentiated play to avoid Anthropic's edge. Fewer than two hundred people grinding through more than one hundred capability evaluations over seven months shows that the winning factor in consumer agents is not model parameters but reliability polish. For the industry, this signals that agent competition is shifting from capability demos to experience delivery: whoever can turn occasional brilliance into stable usability is the one likely to earn a ticket to billions of users.

Source: enbrief original report ↗