Meta began rolling out agentic capabilities in Meta AI on July 24, 2026, letting the assistant plan and carry out tasks in select markets. The features are powered by Muse Spark 1.1, a model Meta describes as built to plan, work with your apps, and follow through from start to finish. The rollout covers select markets in the Meta AI app and meta.ai, expanding to more countries and surfaces including WhatsApp in the coming weeks. ASAP works from Meta's official newsroom post as the primary source to lay out how the feature is structured and what it implies.

What Meta added: not answers, but completion

The change Meta leads with is that Meta AI no longer stops at supplying information but finishes work on the user's behalf. Meta says the assistant can now make plans, follow through on next steps, and keep you on track without needing to be reminded or re-prompted.

The shape of the work is spelled out concretely. Meta AI gathers and synthesizes information from multiple sources, described as ranging from research papers to what creators and communities share, and turns that research into visual presentations. Daily briefings pull calendar information, identify double-bookings and deliver summaries at the time the user prefers. Recurring work is in scope as well: Meta cites weekly meal plans, sneaker drop alerts and trend updates, saying that once you tell Meta AI what you want and when, it will keep delivering.

The design choice of intervening mid-task

The most technically specific item in the announcement is real-time steering. Meta says you do not have to wait for Meta AI to finish before you weigh in while it is putting together a report, a presentation or a plan. Tell it to shift focus, change the tone or cut a section, and it course-corrects while still working.

That choice reveals where Meta locates the failure mode of agent products. The largest cost a user bears when delegating a long task is not a wrong result but the wait until the wrong result arrives. A five-minute job that starts from a wrong premise consumes the whole five minutes, after which the user re-prompts and runs it again from the top. Mid-task steering is an attempt to cut that waste, and it simultaneously frames the agent as a collaborator rather than a fully autonomous worker. Execution is delegated; control is not.

The example that points at Marketplace rather than the calendar

One of Meta's examples does not overlap with what other assistants offer. In the kitchen renovation scenario, Meta AI learns the user's style preferences, scouts Marketplace inventory and sends mood boards. The other examples are more familiar: checking a calendar to suggest restaurant options, or building a week-by-week training schedule adjusted for availability with weekly updates.

The appearance of Marketplace is what locates Meta in this race. Calendar integration and document drafting are common ground every assistant is trying to claim, and on that ground the competition runs on model quality and connector coverage. Searching Marketplace inventory, by contrast, only works on top of an asset Meta already owns. As assistant competition moves into the execution stage, what each company can lean on is the data and inventory inside its own walls, and for Meta that happens to be a marketplace where people buy and sell things. The announced expansion to WhatsApp reads the same way and means more than adding a surface. Once messaging becomes the point of execution, the assistant stops being an app you open and becomes a capability inside a conversation.

How to read an announcement with no numbers

Meta's July 24, 2026 announcement of Muse Spark 1.1 contains no task success rate, no benchmark score and no user count, and that omission sets the limits of any judgment about the release. The value of an agentic feature rests not on the fact that it plans but on how often it finishes, and the absence of that rate is the first gap to note when evaluating this release.

Availability is the same. The post says only "select markets" and lists no countries, so whether users in Korea can use the feature today is not established by this announcement, and with WhatsApp expansion still ahead, the practical experience will differ across markets with different messaging habits. The approval flow is another open item. The post emphasizes that users can redirect work in progress, but it does not state whether hard-to-reverse actions such as bookings or purchases require separate confirmation before execution. Meta does note that users have choice in how they use Meta AI and points to Incognito chats for fully private conversations, and says everything Meta AI creates now lives in one place so it can be revisited, built on and shared.

Where the phrase "personal superintelligence" sits

Meta frames this release as its next step toward personal superintelligence, describing it as an AI that knows your context, is there for you whenever you need it, and handles things so you don't have to. The gap between the strength of that phrase and the shipped feature set is worth reading plainly. What shipped is planning, app integration, mid-task correction and recurring delivery, a list that spans the familiar territory of scheduling, document work and shopping regardless of the ambition in the naming.

The naming still signals direction. Pairing "superintelligence" with "personal" rather than with a capability claim puts Meta's axis on how long a model holds one person's context rather than on how hard a problem it solves. The resource that builds an advantage on that axis is accumulated personal context rather than parameters, and switching costs grow with it. That is why every company is moving quickly while execution features are spreading through the assistant market. Features are matched within months; a user's accumulated context is not.

Source: ASAP summary based on Meta's official newsroom post "Meta AI Doesn't Just Think, It Acts" (July 24, 2026).