At Meta's third-quarter earnings discussion on October 21, 2026, executives said more than 10 million people in the US and Canada had set up Muse, a personal AI agent launched in September. Shopping is a prominent use case. These are setup counts, not daily users or completed purchases. The more significant shift is AI taking part in shopping tasks rather than only giving advice.
Beyond the chatbot
Meta introduced Muse on September 8 as an agent that can act across websites and apps inside a dedicated Muse Secure VM. Meta says users control permissions and confirm sensitive actions such as making purchases or sending emails [1].
An agent can search products, compare offers and assist with checkout. Product descriptions, however, do not prove that every action works automatically or reliably.
What ten million measures
Zuckerberg reported more than ten million setups in the US and Canada after roughly six weeks. Coverage of the earnings discussion identified shopping as a common use case, with more than one-quarter of weekly active users engaging with shopping functionality [2][3][5]. Setup counts, activity and actual purchases are different metrics.
Completed transactions, retention, satisfaction and dispute handling will matter more for long-term adoption than curiosity-driven initial signups.
Retailers must make product data usable
E-commerce websites have traditionally optimized pages for human browsing. Shopping agents may increasingly rely on accurate structured data, prices, stock status, delivery and return terms.
Retailers can begin by auditing whether automated tools can interpret product information correctly. That does not require assuming human browsing will disappear overnight.
Trust is part of the architecture
In October 2026, Claroc described more than 6,800 Muse shared-conversation hostnames discoverable via passive DNS. Researchers inspected a subset and found some sensitive material. This was a shared-link discoverability problem, not evidence of mass account compromise [4].
Claroc reported Meta had suspended the sharing feature and said it had no evidence of third-party misuse [4]. Permissions, sharing behavior, audit trails and revocation deserve scrutiny alongside credential security.
When AI becomes a retail customer
Merchants need not redesign everything for an unproven future. Practical preparation begins with accurate catalogs, inventory, pricing, clear policies and explicit approval for sensitive actions.
As agents increasingly represent buyers, merchants with reliable, comparable information and trustworthy transactions may stand out. The key measure is meaningful delegation, not setups alone.
Sources & further reading
- [1]Introducing Muse Meta · 2026-09-08
- [2]Meta Reports Third Quarter 2026 Results Meta · 2026-10-21
- [3]Meta puts Muse agent at heart of AI strategy Reuters · 2026-10-21
- [4]GhostShares Claroc · 2026-10-20
- [5]Muse shopping adoption MarketWatch · 2026-10-21




