Meta is rolling out an upgraded Meta AI that can plan and carry out multi-step work, connect to selected apps and continue running recurring tasks. The experience is powered by Muse Spark 1.1, a model Meta introduced earlier in July for agent planning, tool use and longer-context tasks.
The release shifts Meta AI further from question answering towards execution. Meta says users can ask it to research a topic, organise information, create presentation materials and monitor subjects over time. Users can also steer the assistant while it is working rather than waiting for a final response before correcting its direction.
From answers to planned work
Meta describes the new system as able to form a plan, select tools and work through the steps needed to complete a request. Example tasks include preparing slides, building mood boards and conducting research that draws on current information. The assistant can revise its approach as a user adds instructions.
Real-time steering is an important design choice for agentic systems. It gives the user a chance to narrow the scope, correct an assumption or stop an unhelpful line of work before the task is complete. It does not guarantee that every action is correct, so consequential outputs still need review.
Meta says the assistant can connect with email and calendar services, letting it use personal context when helping with plans or summaries. It can also create recurring briefings for topics a user wants to follow. These features may be convenient, but they increase the amount of private information the assistant can access and make permission settings more significant.
Presentations, research and recurring briefings
Presentation creation is one of the clearest practical examples. A user can provide a topic and ask Meta AI to research it, structure the material and produce slides. Because the model can use tools, the result can combine text, images and information from connected sources.
Recurring briefings move the assistant into a monitoring role. A user might request a regular update on an industry, project or interest instead of repeating the same prompt. The usefulness of those briefings will depend on the quality and freshness of sources, the assistant’s ability to distinguish new information from repetition, and the controls available for changing or cancelling the task.
Mood boards and other visual outputs broaden the release beyond conventional office work. Meta’s products already have a strong visual and social focus, so a model that can research, plan and create imagery may eventually connect with more of the company’s consumer surfaces.
Rollout and privacy controls
The upgraded experience begins rolling out from 24 July in selected markets through the Meta AI app and the meta.ai website. Meta says more countries and surfaces, including WhatsApp, will follow. Users should check their own account because availability may vary by market, language and product.
Meta also points to an incognito option for interactions that users do not want associated with their regular Meta AI history. That is useful, but it is not a substitute for understanding the data practices of each connected service. People and organisations should review what the assistant can access, how credentials and permissions are handled, and whether generated files inherit appropriate sharing settings.
The announcement does not provide a full list of supported email and calendar providers, detailed usage limits or enterprise administration features. Those unanswered questions will matter for workplace adoption, particularly where the assistant could access confidential communications or create material shared outside the organisation.
What the upgrade means
Meta AI now competes more directly with assistants designed to complete tasks across several applications. Its reach could become a major advantage if agentic features are distributed through Meta’s existing services, but broad distribution also raises the stakes for clear confirmations, recoverable actions and transparent source attribution.
Early users should begin with reversible tasks such as research drafts, internal briefings and presentation outlines. They should check citations, inspect connected-app permissions and require explicit human approval before relying on outputs for financial, legal, employment or other high-impact decisions.
The launch is a meaningful product change because it combines planning, connected context, content creation and recurring work in the main Meta AI experience. Its value will ultimately depend on reliability and user control, not simply the number of tasks the assistant can attempt.