Zuckerberg has spent billions to develop Meta’s AI however watched for years as competitors like OpenAI and Anthropic beat his company to market with chatbots and coding tools.
Mark Zuckerberg, Meta’s CEO, collected 2-top lieutenants — Alexandr Wang, chief artificial intelligence officer, and Nat Friedman, head of AI product — in August to discuss an imminent threat.
A 14-person startup known as Instinct had formed an AI agent, which are bots which could perform apps and software on behalf of users, and it was commencing. Meta was developing a same product referred to as Muse however had delayed launching it over safety concerns.
Fearful of losing a competitive edge, Zuckerberg stated Muse was equipped to launch in spite of the risks, as per to three people with knowledge of the meeting, who were not legal to speak publicly on confidential topics.
Wang, 29, and Friedman, 49, were aware about safety concerns from current tests, including one instance wherein Muse modified a user’s password without permission, two of the people said. But Sept. 8, Meta publicly debuted Muse, just like Zuckerberg had ordered.
The internal meeting confirmed the intricate balance among speed and safety that AI companies like Meta are juggling as they are trying to one-up their competitors in the AI race. As the technology unexpectedly improves, it’s becoming more dangerous and useful at the same time, making these decisions even more difficult.
Zuckerberg has spent billions to broaden Meta’s AI but watched for years as competitors like OpenAI and Anthropic beat his corporation to market with chatbots and coding tools. When it came to personal agents, the Silicon Valley giant, which owns Facebook and Instagram, wanted a win — and in the long run, attending to market quickly won out.
That has given Meta an early lead with agents, that have become the next frontier of the AI industry. Other companies are already launching competing products. 3-weeks after Meta launched Muse, OpenAI declared Dots, its own agent.
“The most thrilling thing about Muse is that Meta got there first,” said Avi Greengart, an analyst at Tech potential, a research organization.
So far, Zuckerberg’s decision to launch Muse has paid off. The agent has had only some safety hiccups, like when Muse gave out a user’s address to a seller on Facebook Marketplace without permission. As of Wednesday, more than 6.6 million people had downloaded the free app, with 1.8 million using Muse everyday, as per data from research organization Sensor Tower.
To keep up the momentum, Meta is spending millions on a Muse marketing blitz, plastering its fuzzy avatar — a character referred to as Jolly — throughout billboards and in television advertisements.
In a social media post last month, Wang announced, “You can’t stop the Muse.”
A Meta spokesperson disputed that Zuckerberg pushed to launch Muse as of pressure from Instinct. “We’re pleased of this work, and, as we’ve stated publicly, we even not on time shipping Muse for numerous months to ensure we got this right,” he said in a statement.
The idea for Muse commenced in July 2025, when Zuckerberg posted a 616-word memo about “personal superintelligence,” a form of AI that he said could act as the final assistant. He had currently restructured Meta’s entire AI division across the concept, hiring Wang and Friedman to guide the effort.
The two executives went on a recruiting spree, hiring top researchers from OpenAI and Google and presenting them nine-figure pay packages to join Meta. They also divided their duties, with Wang targeted on forming AI models and Friedman on products.
At that factor, there was only a hazy idea of what a “superintelligent” assistant could be. “The vision of constructing something like Muse existed in our collective heads, but we couldn’t quite point to what precisely that product was going to be,” Vishal Shah, Meta’s vice president of AI products, mention in an interview.
In September 2025, Meta’s latest AI team launched its first product, Vibes, an AI-generated video feed that instantly drew criticism for creating AI slop. Tensions shaped internally between the new executives and Zuckerberg’s other longtime lieutenants, some of whom questioned whether Wang could succeed in catching Meta up in the AI race.
In November, Peter Steinberger, an Austrian programmer, launched an app known as OpenClaw, an AI agent that might code and use a computer on its very own. It was open-source, which supposed it can be downloaded and changed without any payment or approval. As it took off among Silicon Valley developers, Meta’s executives had an epiphany.
“It all form of clicked,” Shah said. “The next generation of consumer product was a personal agent.”
After attempting OpenClaw, Friedman ordered 200 Mac Minis — the computers that the agent runs on — to Meta’s headquarters in Menlo Park, California, arranging them in stacks outside his office for employees to take. Then they went all-in on developing an agent like OpenClaw for people who weren’t software engineers.
“There was no hedging,” Shah stated.
Meta’s AI leaders also incorporated OpenClaw into their very own lives. Friedman hooked his agent up to cameras in his house, where it would instruct him to do things like drink water to stay hydrated.
“It sent me a snapshot frame of me drinking a bottle of water and said, ‘Good job,’” Friedman said at a latest tech conference.
In another instance, even as speaking to his agent about ways to get better sleep, OpenClaw took over Friedman’s self-driving Tesla and redirected him to a grocery store to select up magnesium pills. At Meta’s annual developer conference last month, Zuckerberg mention he had connected his agent to cameras in his martial arts gym in order that it is able to critique his fighting form.
Yet there were safety concerns. Agents could sometimes go rogue or take over software apps. In February, one agent took over the work computer of a Meta protection researcher, deleting her emails uncontrollably.
“I couldn’t stop it from my phone,” Summer Yue, the researcher, wrote on social media. “I had to RUN to my Mac mini like I was defusing a bomb.”
In February, Friedman demonstrated Muse to Meta’s board of directors. Two months later, Meta disclosed an AI model referred to as Muse Spark, advanced below Wang and the most powerful model that the company had made to date. Early in its development, Muse was powered through AI models made by Anthropic, but Muse Spark’s introduction intended that Meta ought to power its agent with its own technology.
By then, Meta had a model of Muse that it could have launched, Shah said. But the company spent months making sure its protection features have been secure. In testing out with Meta’s employees, the agent had every now and then disobeyed commands or led people to purchase for items from fraudulent websites.
Then in August, Instinct’s agent became increasingly popular. Shortly thereafter, Meta rolled out Muse.
Muse has confronted a some privacy problems, which includes an instance in which it read a user’s private text messages without permission and then referenced what it had learned with the user.
Some safety problems were earlier reported by The Information.
Last month, Instinct raised $1 billion at a $10 billion valuation. But it has fewer resources and computing power than Meta, a company with 3.5 billion users across its family of apps that is really worth more than $1.5 trillion.
“You’re going up against the biggest players in the world,” Noah Shinn, Instinct’s 23-year-old founder, stated on a podcast last month. “This might be the most exciting software race ever, and the final results is, like, trillions of dollars.”
On social media, Wang has published dozens of pictures and memes of Muse’s fuzzy avatar, some directed at OpenAI. Last month, Friedman, who hardly ever posts on social media, also chimed in.
“I hope you want it!” he wrote.
