Meta rolled out "spaces" to every Muse user on November 6 — small, persistent surfaces the assistant builds on request. Describe what you want and you get a shopping hub, a school dashboard, a music tracker or a trivia game, all running inside Muse rather than in a separate app. Meta says spaces can be pinned to the sidebar or the main tab, adjusted by asking, checked off, annotated, and scheduled. The company's own example: keep your weekly planner up front during the workweek, then swap it for weekend plans on Saturday.
The technically interesting part is the plumbing. Connect a space to a relevant app and Muse refreshes pricing, dates and other details in the background. That makes a space something more than a view over data fetched once: it is a small, continuously running process with permissions, holding connections and pulling changes with nobody watching it.
It ships alongside a "specialized" version of the Muse Wonder model. Wonder arrived in October with two properties that point directly here: Active Memory, meant to carry context across weeks and months, and Active Focus, meant to keep working on a task past obstacles. An assistant that remembers a child's school calendar can build a dashboard out of it — and then has to be able to say which version of that calendar it read.
Why this is more than a feature: an assistant that generates a surface per question and throws it away is a chat. A surface that persists, has a schedule, refreshes its data and may eventually be shared is a platform. And it sits between the user and the services — where the operating system and the app store sit today. If spaces can invoke third-party apps, Meta becomes a party to how those services are reached.
Meta says it is exploring ways to share spaces with other people and collaborate on them. That is precisely where the awkward questions start. A shared, generated surface needs rules for misleading presentation, for data provenance, and for content the assistant assembled itself. An app store has a review process for that. A prompt does not.
The error surface is the other problem. Automatically refreshed prices and dates can go stale, get attached to the wrong item, or come from the wrong source — and the user sees a tidy card, not the fetch path. Someone reading a generated view as fact is making decisions on the basis of a surface whose provenance they cannot inspect. It is the same question as with any agent, only more visible.
Two caveats. First, the product claim is not independently tested: whether background-refreshed spaces are actually reliable will only be clear after the rollout, and Meta has published no error rates. Second, monetisation is unstated and obvious. A place where people build shopping hubs and watch prices is a natural advertising surface for an advertising company. Anyone building a space today is building it inside a product whose business model has not been announced.
