Meta AI becomes an agent that acts with Muse Spark 1.1
Meta AI adds planning, connected tasks, live steering and recurring work with Muse Spark 1.1, available first in select markets.
Meta announced on July 24, 2026 that Meta AI is now powered by Muse Spark 1.1. The update changes Meta AI from a system that mainly answers into one that can make plans, take next steps and keep work moving. It is available today in the Meta AI app and at meta.ai in select markets. WhatsApp support is announced for the coming weeks, so it is not available there yet.
What Muse Spark 1.1 actually lets Meta AI do
The central change is execution across a task. Meta says users can set up tasks with Meta AI once and then leave it to handle them. That makes the product closer to an agent than a conventional question-and-answer interface. It can make a plan, follow through on the next steps and help keep the user on track.
Meta AI can connect to email and calendar apps. Those connections give it context for practical work instead of limiting it to the text entered in one conversation. One stated feature is a daily briefing that pulls what is on the user's calendar. The announcement also describes research completed in minutes, with information synthesized from across the web.
The output is not limited to prose. Meta AI can generate slides and mood boards. Users can steer it in real time while it works, which matters when a plan needs correction before completion. The combination is specific: connected context, multi-step execution, generated artifacts and live intervention.
Meta gives several examples of the intended scope. A user can plan a kitchen renovation, build a half-marathon training schedule, arrange birthday dinner reservations or prepare weekly meal plans. Meta AI can also create trend alerts. These examples span research, scheduling, reservations, planning and recurring monitoring. The announcement does not provide additional technical detail about how Muse Spark 1.1 performs these operations.
The product is free. Meta did not mention a paid tier in the announcement. That makes the first release a direct test of whether users will hand recurring personal work to a general assistant without a separate subscription choice.
What this means when choosing AI tools
Meta AI now belongs in an agent evaluation, not only a chatbot comparison. The useful test is whether a buyer needs an assistant that can act from connected email and calendar context, create plans and continue work after the initial setup. Availability also matters: the current release is limited to select markets and two Meta surfaces.
For web research, Perplexity is another tool page to compare. Meta's stated difference in this announcement is not research alone. It combines research synthesis with follow-through, generated slides or mood boards, connected apps and tasks that can be left running. Buyers should compare the complete workflow they need rather than treating every research response as the same product category.
NotebookLM is relevant when the selection starts with research and synthesis. Canva Magic Studio is relevant when presentation or mood-board output is the main requirement. Meta AI now puts those output types inside a broader assistant flow, but the announcement gives no detailed feature-by-feature comparison with either product.
For recurring workflows, Zapier provides another reference point. The practical distinction to examine is how much control a workflow needs. Meta says its agent can be steered in real time and can handle a task after one setup. The announcement does not describe a workflow builder, configuration model or integration list beyond email and calendar apps.
The Automation & Agents category is the clearest place to compare alternatives. A technical buyer should separate four requirements: access to connected context, research synthesis, artifact generation and continued execution. Meta AI now claims all four, but only its stated release surfaces and markets can be evaluated today.
Limits and what is not available yet
The rollout is not global. Meta says the features are available first in select markets and will expand to more countries and surfaces. The announcement does not name those additional countries or give a schedule for the broader expansion.
WhatsApp is also not part of today's availability. Meta says it is coming in the following weeks. Until that release happens, the active surfaces are the Meta AI app and meta.ai. A team choosing a tool specifically for WhatsApp should treat the capability as announced, not shipped.
Muse Spark itself is not listed as a separate linked product or standalone service in the announcement facts. It is the system powering Meta AI 1.1. The announcement does not state that developers can access it through an API, choose it as a model or deploy it independently. It also does not describe a paid tier.
Finally, the examples show breadth but do not establish detailed operating limits. Meta names email and calendar connections, but no other connected-app list is provided. It says tasks can be left for Meta AI to handle, but does not state duration, volume or execution limits. Those points should remain open questions rather than assumed capabilities.
The next concrete milestone is WhatsApp availability in the coming weeks. After that, the important signals are expansion beyond select markets and clarity about additional surfaces, connections and operating limits.
ai101
Tools in this post
Meta AI
FreeFree assistant across WhatsApp, Instagram and the web.
Perplexity
FreemiumAI answer engine with cited, up-to-date sources.
NotebookLM
FreeGoogle’s research notebook that answers from YOUR sources.
Canva Magic Studio
FreemiumAI design tools for non-designers, inside Canva.
Zapier
FreemiumConnect 8,000+ apps with AI-powered workflows.
Keep reading
Claude now marks text and files processed by its models
Claude now adds text watermarks and C2PA file metadata, but a detected mark shows processing by Claude, not authorship.
Meta ships Muse Code, a terminal coding agent built for multi-agent workflows
Meta released Muse Code in early beta alongside Muse Spark 1.2, with a native CLI, observable subagents and replayable event logs.
Google releases Gemini 3.6 Flash and 3.5 Flash-Lite, with Flash Cyber in limited pilot
Google released two Gemini Flash models on July 21, while its cybersecurity model remains restricted to a limited CodeMender pilot.