Meta AI becomes a task-runner and the WhatsApp advantage makes it a real threat

Meta AI can now schedule tasks, connect to Gmail and Google Calendar, create slides, and handle recurring work, powered by the Muse Spark 1.1 model with a 1 million token context window. The feature rollout starts in select markets on the Meta AI app and meta.ai, with WhatsApp integration coming in weeks, giving Meta a distribution advantage over standalone AI assistants through its 3 billion WhatsApp users. Meta also opened the Muse Spark 1.1 API to US developers in public preview with $20 in free credits, positioning itself as both a product and infrastructure competitor against OpenAI's ChatGPT Work and Google's delayed Gemini 3.5 Pro.

Meta AI has moved from answering prompts to carrying out scheduled work, and the real threat is not only the model. It is the fact that WhatsApp gives Meta distribution no standalone AI assistant can match. Meta made its AI assistant more useful on July 24, and the timing is hard to miss. According to Meta's own newsroom, Meta AI can now make plans, connect to Gmail and Google Calendar, create slides, and handle recurring tasks after you set them up once. That changed things. The feature rollout is powered by Muse Spark 1.1, the model Meta Superintelligence Labs released on July 9. Meta says the model is built for agentic work, with a 1 million token context window, parallel subagent delegation, tool use, computer use, coding, and multimodal understanding. The features are starting in select markets in the Meta AI app and at meta.ai, with WhatsApp coming in the following weeks. That is the story. The model is only part of the pressure Meta's developer page lists Muse Spark 1.1 at 88.1 on MCP Atlas, 54.7 on JobBench, 62.1 on Humanity's Last Exam with tools, and 57.2 on Finance Agent v2. Those are Meta's published numbers, not independent proof that the model is suddenly the best agent on the market. You should treat them the way you treat every benchmark table from a company selling access to its own model: useful for orientation, incomplete as evidence. The harder fact is that Muse Spark 1.1 is now available through the Meta Model API in public preview for US developers, with $20 in free credits. For the first time, outside developers can build directly on a Muse Spark model rather than only watching Meta use it inside its own apps. That matters for builders because Meta is no longer only a product competitor. It is also trying to become an infrastructure supplier. OpenAI is moving in the same direction from the other side. On July 9, OpenAI announced ChatGPT Work, powered by GPT-5.6, as an agent that can gather context across connected apps and files, then produce finished documents, spreadsheets, presentations, reports, and Sites. Google is under its own pressure after Gemini 3.5 Pro missed its expected June public release window, with recent reports pointing to delays around long-horizon and coding performance. Everyone is chasing the same behavior: AI that does not wait politely for the next prompt. The WhatsApp problem for startups Meta's advantage is not subtle. OpenAI and Google still need many users to decide they want an AI assistant badly enough to download a separate product, wire up their files, and keep coming back to it. Meta can put the same class of feature in the places people already open all day. WhatsApp had about 3 billion monthly active users as of 2025, according to Statista. Meta's family of apps had 3.58 billion daily active people in December 2025, based on the company's fourth-quarter results. That is a different starting line. If you can set a morning briefing from inside a WhatsApp conversation, without installing another app or explaining another workflow to your family or team, adoption gets easier. It gets brutally easy. A standalone assistant has to be much better to justify the extra step. That is the pressure this creates for startups. Companies building consumer AI assistants, scheduling agents, daily digest tools, research helpers, and light productivity bots are now competing against a version of the same work baked into the world's biggest messaging surface. It's not a fair fight on distribution. Frankly, it never was. A niche product can still win, but it has to earn the right to exist. A scheduling agent that understands oncology clinics has a real argument. So does a research assistant built around SEC filings, or an automation layer wired into tools a consumer assistant will not touch. A generic assistant that reminds you about your calendar and sends you recurring updates has a much weaker one. Breadth is already gone. Meta is also running the consumer and business tracks in parallel. In June, it launched Meta Business Agent for WhatsApp, Messenger, and Instagram, saying more than 1 million businesses already use a Meta Business Agent on WhatsApp and Messenger to respond to customers around the clock. The business version can answer questions, recommend products from a catalog, book appointments, qualify leads, and work alongside the Meta Business Agent Platform for larger companies. That gives Meta two routes into the same habit. Consumers get daily briefings, research help, slides, mood boards, and meal plans. Businesses get sales support, appointment booking, and customer follow-up inside messaging threads they already use. The assistant becomes less like an app and more like a background function. There are still real limits. Axios noted that Meta's AI agents remain less capable than rival systems from OpenAI, Anthropic, and Google for broader autonomous work. Not every WhatsApp feature is live yet, and US-only developer access limits who can build on the API at launch. You should not confuse a rollout with victory. Still, the direction is plain. Meta is not trying to win only by having the smartest model in a lab. It is trying to make agentic AI the default behavior inside products people already cannot put down. If you are building in this space, that is the competitive fact you have to start with. Also read: A researcher says he has broken every major AI model at once and is deciding when to tell the world https://startupfortune.com/a-researcher-says-he-has-broken-every-major-ai-model-at-once-and-is-deciding-when-to-tell-the-world/ • A New Brunswick lawmaker read his AI chatbot's rewriting instructions aloud in the legislature and didn't notice https://startupfortune.com/a-new-brunswick-lawmaker-read-his-ai-chatbots-rewriting-instructions-aloud-in-the-legislature-and-didnt-notice/ • Bluesky turns Attie into a social research tool and the open protocol is the whole point https://startupfortune.com/bluesky-turns-attie-into-a-social-research-tool-and-the-open-protocol-is-the-whole-point/