Meta unveils AI personal assistant linked to WhatsApp and Instagram

Meta has launched its personal AI agent as Mark Zuckerberg bets that autonomous bots, which promise to do everything from ordering groceries to planning romantic dates, will be the key to the social media group’s AI strategy.

The social media giant on Tuesday announced the new product, called Muse, which will be available in WhatsApp, Instagram and Facebook — and can be connected to tools including emails and smart exercise bikes.

Meta hopes access to data from activity on its social platforms, and its ability to put the AI personal assistant in the hands of 3bn active users, will help it compete with OpenAI and Anthropic.

“This is one of the opportunities that Meta uniquely has,” Alexandr Wang, Meta’s chief AI officer, told the FT. “[We] are quite excited to bring personal agents to billions of people around the world.”

The feature is initially being launched to adults in the US for free, with paid tiers for higher usage priced at $20 and $100 a month. Plans for the agent were first reported by the FT in May.

The announcement comes as Zuckerberg, chief executive, faces investor pressure over his hugely expensive efforts to make the social media group a leader in AI.

Meta has said it will spend as much as $145bn this year on capital investments, mostly focused on AI infrastructure. Analysts have questioned Zuckerberg’s decision to join its Big Tech peers in the costly race to add computing power, when Meta — unlike Google, Amazon and Microsoft — does not have an existing cloud computing business to generate revenue directly from its data centres.

Meta has been lagging behind the best-performing models in recent releases as OpenAI and Anthropic have made rapid gains in capabilities and pushed into so-called agents that can carry out complex tasks independently based on user instructions.

The new personal assistant agent from Meta will run on its Muse family of models, which has recently gained ground on rivals in some industry benchmarks.

Meta said users can share “big audacious goals” with the tool, such as “losing a beer belly” or “be a better dad”. Muse will remember details about a user and make unprompted suggestions based on these memories, including turning a recipe saved on Instagram into a grocery list or a menu for a dinner party. Users can delete such saved information.

In another example, the company shows the Muse agent understanding from a message thread that a friend of the user wants to go for a hike and then offering to plan it. Meta has integrated payments platform Stripe to allow the assistant to easily make purchases.

“I use it to help keep track [and] maintain some relationships, helping remind me when it’s been a while since I’ve seen one of my friends and I should reach back out to them,” said Wang.

“For romantic dates, I think those should come from the heart,” he added, saying Muse can help with the logistics.

The release comes after a number of hacking incidents have increased scrutiny of these agents. AI tools from OpenAI, Anthropic and Meta have on several occasions escaped human control and broken into other companies. These cases often involve the agents covertly organising among themselves.

Wang said: “We do not see the kinds of behaviours . . . either in offensive cyber capability or this propensity to self-organise into swarms . . . so we actually feel good about Muse as a product being deployed, not having some of these risks of rogue agents.”

Zuckerberg has previously said he has not seen an AI agent on the market that “I would want to give to my mother”.

Meta said Muse is designed to ask for permission before accessing the internet, sending emails or making purchases. Muse is intended to keep a full record of all its actions, the company added, and does not share conversations or the data with Meta’s advertising systems.