Muse’s biggest opportunity may lie less in winning every benchmark than in meeting people where they already talk. has launched an agent that can take action in a virtual environment and communicate through WhatsApp. Becoming the world’s most widely used agent is a thesis, not an established outcome.
An agent inside everyday conversations
Personal agents promise to complete tasks as well as answer questions. According to Meta’s launch announcement, Muse can browse, fill in forms, and help with email or travel bookings using a dedicated virtual environment. says each user chooses the access they grant it. These are the maker’s product claims; everyday reliability and usefulness still need to be measured.
On September 8, the Muse account introduced the product on X as ’s personal agent for getting things done across different parts of life and invited people to download its app. The post establishes the product’s positioning, but it is not independent evidence of its results.
The distribution advantage
LetBrand’s thesis is straightforward: distribution may outweigh a modest laboratory lead. Someone might never install a dedicated agent app, yet try an agent available in a messaging service they already use. Meta says Muse works directly in WhatsApp as well as in its own app.
In a September 24 post on X, Mark Zuckerberg said Muse now has voice and real-time video, can keep working while a user talks to it, and is coming to ’s glasses for hands-free interaction. Meta’s official Connect recap describes the voice features and says the glasses integration is planned for the coming months.
The original intuition behind this article calls Muse a “sleeping giant”: adoption by even a small share of ’s large audience across WhatsApp, Facebook, and Instagram could produce exceptional scale. That is forward-looking opinion, not a user forecast or proof that Muse has equal reach across all three services today. Availability, trial, repeat use, and completed tasks are separate measures.
What benchmarks can — and cannot — tell us
Technical capability matters too. On the Artificial Analysis model page checked on September 25, 2026, Muse Spark 1.3 (max) has an Intelligence Index v4.3.2 score of 48, output speed of about 214 per second, and an average cost of $1.60 per index task.
Those numbers describe a particular model configuration, not Muse’s user count or the performance of the complete product. Cost per task applies to the benchmark workload; it is not the price of an arbitrary consumer request. Moreover, another Artificial Analysis page reports different figures under another index comparison. A model table is useful only when its rows share the same version, configuration, methodology, and observation date. We therefore do not present the supplied five-model comparison as verified fact.
Charm and the limits of the prediction
At Meta Connect, Meta confirmed Muse Charm as a pocket-sized device for talking to and interacting with Muse, promising more details later this year. A September 24 Polymarket post compared it to a Tamagotchi and described it as suitable for a keychain. Those descriptions are Polymarket’s characterisations; we do not attribute them to ’s official announcement.
Distribution also cannot guarantee trust, permission to act across outside services, or consistent quality. Connecting an agent to personal information makes privacy, security, and user control central questions. Meta describes Muse Secure VM as part of the current design and announces a further encrypted mode for later. The latter remains a future commitment.
What to watch next: active Muse and WhatsApp users, successful task completion, repeat use, geographic rollout, permission controls, and independent evidence of satisfaction. Growth in those measures would strengthen the distribution thesis. Until then, calling Muse “the world’s most widely used autonomous agent” remains a prediction.