Ask once, get the briefing on Monday
You open your inbox, check the calendar and discover that you accepted two meetings for the same hour. Meta wants to take over that routine. Its assistant can now connect to Gmail and Google Calendar, assemble a daily briefing and flag double bookings or last-minute changes.
The real change is repetition. According to Meta's announcement, users can set a task once and keep receiving weekly meal plans, sneaker-drop alerts or an afternoon update on a topic they follow. There is no need to return every morning with the same prompt.
It can also find restaurants, check which evening is free and suggest options for a birthday dinner. Ask for a half-marathon plan and it will adjust the schedule around your availability, then deliver the coming week's program every Monday. These are Meta's own demonstrations rather than independent tests, but they make the ambition clear: Meta AI is trying to stop waiting inside a chat window.
Research and slides happen inside the same assignment
Meta AI can research the web, combine papers and articles with posts from creators and communities, and turn the result into a presentation. Users can steer the work while it is happening—cutting a section, changing the emphasis or asking for a different tone before the report is complete.
Everything the assistant creates remains available inside the app to revisit, expand or share. Meta also demonstrates a renovation project where the agent learns a user's taste, searches Marketplace and builds a mood board that respects the budget. That is Meta's advantage: years of photos, videos, shops and social connections already live inside its products.
None of that automatically makes Meta AI the most capable agent. Axios notes that systems from OpenAI, Anthropic and Google still handle a broader range of longer tasks. Meta may not need to win every benchmark, however, if it can place the assistant across phones, Instagram, Facebook and eventually WhatsApp.
Muse Spark 1.1 has been waiting for this job
The new features run on Muse Spark 1.1, the model Meta introduced on July 9. It supports a one-million-token context window, tool use, computer control and multi-agent orchestration. In practical terms, it can remember a long assignment, decide when automation is faster than clicking through an interface and delegate parts of the work without turning every action into a new conversation.
Meta says in its Muse Spark 1.1 technical overview that the model can write a script when that is efficient and return to direct interface control when an app requires it. Developers can also access the model through the public preview of Meta's new Model API.
The company says it tested the system against prompt injection and attacks hidden in external data. Connecting an assistant to email and calendars still changes the cost of a mistake. A bad answer is annoying. An altered appointment, private detail copied into a deck or action taken inside an account is harder to shrug off.
There is no confirmed timeline for Spain
The features are beginning to roll out in “select markets” through the Meta AI app and meta.ai. Meta has not published a full country list or confirmed when Spain and the rest of Europe will receive them. Reuters reports that more regions and WhatsApp will follow, without a firm date.
Incognito chats remain available for private conversations, and Meta says users choose how they interact with the assistant. We still need to know which permissions Gmail and Calendar will request, how long access persists and which actions require confirmation. Those details matter more than a polished demo once an agent starts moving through a real schedule.
The pitch is already clear: ask once and receive the work when it is due. If Meta can bring that to WhatsApp without turning every morning into a sponsored notification, Zuckerberg may have found a home for AI that is much harder to ignore than another chatbot tab.
The conversation starts here
Sign in with a supporter account to comment. Sign in
Nobody has commented yet. Want to go first?