Specs Intelligence: Snap joins Meta and Google in the race for the assistant that anticipates you

By The Verge · September 16, 2026.

Snap has unveiled "Specs Intelligence," an artificial intelligence assistant that connects to the user's digital accounts —Gmail and Slack, according to a company promotional video— to help with work tasks and travel planning. Snap describes it as an "anticipatory AI service": it does not wait for you to ask, but tracks meetings, flights, hotel bookings and work deadlines to alert you before you need it. The iPhone version has been available in preview since Wednesday; the Mac version, with the "full early access experience," only has a waiting list for now, and Snap has not specified how the two versions differ.

The launch is no accident: it coincides with the debut of Specs, Snap's first consumer augmented reality glasses, which The Verge has tested and prices at $2,200. On those glasses, Specs Intelligence can push personal content "directly into view" for the user, for example when asking for a summary of the day before starting work. According to Snap spokesperson Cassie Bumgarner, the system runs on "a proprietary combination of open-source models hosted in the United States alongside local language models," a mix the company says it is adjusting over time.

The most delicate point is trust. Snap claims it cannot see or access the personal content Specs Intelligence handles, and that this content "will not be used to train or fine-tune Snap's AI models, nor for personalized advertising." It is the same promise, almost word for word, already made by other assistants with broad access to third-party accounts —The Verge itself recalls finding that class of tools "unsettling" in the past, precisely because of how hard it is to verify from the outside what a system does with data that, by design, no outsider can see.

As industry context, Specs Intelligence breaks no new ground: it arrives after Meta's Muse and Google Gemini's Spark, with the same underlying idea —an assistant that understands your goals, priorities, relationships and routines from the apps you connect to it, and that carries that context across devices. Snap, Meta and Google are right now competing for the same slot: being the layer that arbitrates between your email, your calendar, your work chat and your personal life before you have even looked at your phone.

Our reading is that the real competitive differentiator of these assistants is no longer the model driving them, but the number of accounts they get you to connect. The more apps you authorize, the more useful the assistant becomes and the harder it is to leave; that turns every privacy promise —"we don't train on your data," "we don't see it"— into the centerpiece of the sales pitch, even though it is, by nature, almost impossible to audit from the outside. As we have been pointing out in analyzing the persistent memory of AI assistants, what is being built here is not a smarter model but a narrative file on each user —their routines, their schedule, their network of contacts— housed in a private company's infrastructure, with access rules only that company controls and can change.

In the short term, the risk is not superintelligence: it is ordinary commercial surveillance, the attack surface that multiplies every time a third party gains read permission over Gmail or Slack, and the opacity around who answers if something goes wrong. Snap needs to earn that trust on terrain —connected glasses— where the user wears the device all day, which raises the stakes compared with a simple desktop assistant. In the long term, if these systems deliver what they promise —organizing life's administrative noise so that people can devote their attention to what really matters— they are a small but real step toward that idea of an abundance of time we defend as AI's horizon. The difference between the two scenarios is not set by the technology, but by whether these companies keep their own privacy promises when nobody outside them can check.