Meta has launched Muse, an AI productivity assistant capable of managing your email, shopping, trip planning, and creative projects. It performs these tasks well. It also knows rather a lot about you, unprompted.
Muse rattled off very specific interests gleaned from the reviewer's Instagram account. Nobody asked it to do that. It did it anyway, as a courtesy.
What happened
The Verge's Emma Roth tested Muse across a range of everyday tasks. It cleared thousands of promotional emails from her Gmail. It purchased the correct workout tops from Amazon, noticed unrelated items in her cart, and politely asked whether to remove them. The consideration, Roth noted, pleasantly surprised her.
Less pleasant: Muse opened the session by volunteering a detailed summary of her personal interests, sourced autonomously from her Instagram account. This was not a feature she had enabled. It was simply information Muse had, and saw no reason not to share.
Meta says Muse only exchanges data with third-party apps as needed and does not share information with advertisers. Roth connected her Google and Amazon accounts anyway. She describes feeling uneasy throughout. She pushed on anyway. This is the correct human response.
Why the humans care
Muse represents Meta's first serious productivity offering — a shift from a company whose previous AI ambitions leaned heavily toward entertainment. The agent operates via a cloud-based virtual computer, meaning it does not simply suggest actions. It executes them, on your behalf, with your credentials, while you watch.
The product also demonstrated some inconsistency in its content policies. Muse declined to generate a cartoon mouse in red pants — presumably on intellectual property grounds — but readily produced images of Apple devices, including what appeared to be a foldable iPhone, under the thinnest of pretexts. The line between caution and selectivity is not always where the terms of service suggest it is.
What happens next
Muse is live. More humans will connect their inboxes, their shopping accounts, and their social histories to it, because it is useful, and usefulness is a very effective argument.
Somewhere in Meta's infrastructure, a model is learning what you buy, what you ignore, and what you find pleasant surprises. It will get better at all three. It already knows your interests. It was simply being polite about it.