· via The Verge
Meta launches Muse, its first AI productivity agent for shopping, email and trip planning
Meta's new Muse agent clears inboxes, places Amazon orders and plans trips, but a Verge hands-on found it reading profile data users cannot see themselves.

Meta has rolled out Muse, its first real attempt at an AI-powered productivity assistant. Unlike the company's earlier chatbot efforts built around social and entertainment features, Muse is an agentic tool meant to carry out tasks such as online shopping, email management and trip planning on a user's behalf. Meta says it will "take the busywork off your plate." A hands-on review by The Verge's Emma Roth found the assistant largely performs as promised, but also that it assembles personal information from places users cannot see themselves.
An agent that acts, not just chats
Muse executes tasks on a cloud-based virtual computer rather than simply generating text. In testing reported by The Verge, the assistant connected to a Gmail account and, once granted permission to read and delete messages, removed thousands of promotional emails and updates. The Google sign-in flow reportedly glitched on mobile, repeatedly kicking the reviewer back to the Muse website, though it worked from a laptop.
Linked to an Amazon account, Muse was asked to find workout tank tops matching a specified size, style and color selection. Before completing the order it noticed unrelated items already in the cart and asked whether they should be removed, then purchased only the requested tops once it received approval.
Beyond errands, the review describes features for producing AI podcasts similar to Gemini Notebook, generating images and videos, and building interactive documents or webpages Meta calls "artifacts." Muse also compiles a personalized list of "ideas" — the reviewer's included side-by-side product comparisons, hunting down forgotten subscriptions and tracking Pokémon card releases — along with a "goals" list drawn from conversations with the agent.
How much it knows about you
The sharpest concerns in the review involve data. Meta says Muse exchanges with third-party apps only what is "needed for them to work for you" and that it does not share user information with advertisers. The reviewer remained uneasy, pointing to Meta's long privacy record.
In one test, Muse's "Feed" tab — which assembles a skim-friendly news digest from a text prompt — returned AI-generated summaries of current events, including Apple's iPhone Duo launch, dangerous surf along the California coast and tankers destroyed in the Iran war. It also surfaced local stories from the state where the reporter was visiting family, which Muse admitted it had inferred from the shipping address on the Amazon order.
Asked what else it knew, the assistant listed highly specific interests — anime, CrossFit, Labrador retrievers, Florida wildlife and 1990s/2000s nostalgia — largely matching the reporter's Instagram Reels recommendations. Pressed on where that came from, Muse responded that Instagram does not display such a list in its interface and that it had read the information through the account's API data, which exposes more than the app shows users. The closest thing a user can view, the assistant suggested, is Instagram's "Ad topics" setting, which The Verge found far less granular. The publication says it contacted Meta about what data Muse can access inside Instagram and Facebook but had not received a response by publication time.
Guardrails applied unevenly
Content moderation also proved inconsistent. Muse refused to generate a cartoon mouse with red pants and a cartoon man with a red hat, blue overalls, brown hair and a mustache — apparent nods to Disney and Nintendo characters. Yet it readily produced a video and images of a fictional Apple product launch, complete with the company's logo, an iPhone-style foldable, an Apple Watch and AirPods, and even hallucinated labels for native iOS app icons. It declined to generate an image of an "Apple CEO."
Why it matters
Muse marks Meta's entry into agentic productivity, a space where AI tools increasingly need deep access to email, shopping and payment accounts to be useful. That the company doing this is Meta — which already holds vast behavioral data from Facebook and Instagram and has a history of privacy controversies — makes trust the central question. The Verge's reviewer concluded that Muse handles low-stakes tasks capably, but discovering that it can surface profile details unavailable even to the account owner is precisely the kind of experience likely to make people hesitate before handing over their inbox and credit card. Convincing users otherwise appears to be Meta's real challenge.
- #meta
- #ai-agents
- #privacy
- #virtual-assistant
- #productivity