Meta version of "Lobster" -- Zuckerberg's high-stakes "AI assistant Muse" officially released.

If the wildly popular OpenClaw showed geeks the prototype of a local AI assistant, then Zuckerberg's gamble on "AI assistant Muse" is a cloud-based "Lobster" that is open to the general public and ready to use out of the box.

On September 8, Meta officially launched its largest AI product bet to date – Muse, a personal AI agent , putting its vision of "personal superintelligence" proposed more than a year ago into practice.

This product is equipped with an independent virtual computer that can perform tasks on behalf of users around the clock, and it launches an aggressive price offensive in the personal AI agent market by offering massive free computing power.

It not only possesses the "lobster-like" ability to operate autonomously around the clock, but also minimizes the complexity of configuration, allowing ordinary people to have an all-powerful surrogate in the cloud without needing to purchase a computer and build it themselves. Zuckerberg stated:

I don't think billions of people will buy a Mac Studio, configure it themselves, and run it at home.

Upon launch, Muse offers a free weekly allowance of up to 100 million tokens and includes a confidential virtual machine system designed with the participation of Signal founder Moxie Marlinspike, which even Meta itself cannot access.

Zuckerberg stated that his ultimate business model involves taking a "very small percentage" of the transactions completed by users through Muse and offering paid subscription services to heavy users.

The release of Muse directly impacted AI startups like Instinct and Town, which were pursuing the same path. These companies lacked the subsidies from Meta's existing advertising business, while Meta set market expectations by offering near-free services—that personal AI assistants should perform a significant amount of work before requiring users to pay.

"7x24-hour work and nighttime reflection": AI with proactive network effects

Investors are currently focused on whether AI products can truly change users' workflows.

Muse's solution is "long-term memory" and "proactivity." It breaks through the limitation of a single prompt, and is equipped with a dedicated virtual machine.

In his latest interview on the Sources podcast, Zuckerberg precisely defined his ambition:

Give everyone in the world a highly capable personal agent who can understand their goals and work on their behalf 24/7.

Unlike passive response models, Muse is highly proactive. It can integrate information at night, which the team calls "learning through the night."

Zuckerberg gave an example: when he asked Muse to create a game strategy for their daughter, Muse would proactively suggest:

Now that we have a guide, would you like me to expand it so that it can also teach historical lessons about different civilizations?

What's even more noteworthy is that Meta is incorporating its expertise in "network effects" into AI learning. Zuckerberg revealed that Muse has a "fleet" concept, where underlying agents can learn anonymized insights from the entire user base.

When asked if this would be a competitive advantage in the future, he said:

Right now, I think the whole industry mostly sees agency as a single-player game... but it will only get better as more and more people start using Muse.

Free computing power in exchange for "commercial commission": Muse's ultimate business model

Regarding the "revenue growth rate and potential" that Wall Street is most concerned about, Zuckerberg gave Meta's economics for the first time: to seize the market with a highly aggressive free strategy and ultimately make a profit by taking a cut of commercial transactions.

To establish a significant competitive advantage in the personal agent market, Meta is giving away a large amount of computing power for free. Zuckerberg revealed:

When Muse launched, it offered up to 100 million free tokens per week, plus a virtual machine that represented your work.

While startups were still charging monthly subscription fees, Meta used profits from its main business to subsidize and directly raise the barrier to entry for free services in the industry.

Regarding future monetization methods, Zuckerberg outlined a closed loop with immense potential:

We are basically confident in supporting the fact that we believe this will effectively generate substantial profits for those who use it to run small businesses, make money, or conduct business transactions. Our anticipated business model, over time, essentially involves taking a very small commission from any transaction.

He further elaborated on the logic behind this payment model:

We believe this thing will actually make you money and save you money, and that's how it will pay for itself.

Meanwhile, users can also connect Muse directly to Meta's advertising system, allowing AI to automatically generate products and manage ad placement, forming a perfect B-end business collaboration . In addition, Meta will also provide paid subscription services for heavy users.

Absolute privacy barriers and the "long-term gamble" on data centers

The biggest obstacle to enabling AI agents to take over users' bank cards, emails, and business bills is trust. Zuckerberg astutely recognized this and made "privacy and security" the core competitive advantage that differentiates Muse from its competitors.

To technically "prove his innocence," Zuckerberg personally recruited Moxie Marlinspike, the founder of the well-known encrypted communication software Signal, to lead the development of confidential virtual machine technology . Zuckerberg promised:

Even Meta cannot see the contents inside.

In addition, Muse is equipped with a dedicated "sentinel" agent that monitors Muse's incoming and outgoing data, prevents prompt word injection, and requires human approval before performing sensitive operations.

In response to current market concerns about over-investment in AI infrastructure and local community resistance to data centers, Zuckerberg distanced Meta from developers seeking short-term profits. He stated that Meta is investing in a community for "decades," while speculators "don't care about the local community like we do, and they're not investing for the long term."

The recombinant laboratory prioritizes talent density, and the next-generation large model, "watermelon," is about to be unveiled.

Regarding the progress of model capabilities, Zuckerberg acknowledged that the previous Llama 4 roadmap had deviated from its intended course.

To this end, he personally reorganized the AI lab, establishing it around his workstation, pursuing "talent density" to an extreme degree, and significantly increasing investment in computing power, reaching several gigawatts of computing capacity.

Currently, Muse is powered by the Muse Spark 1.3 model, codenamed "Avocado." However, Zuckerberg has released a major announcement: the next-generation model, codenamed "Watermelon," is about to be released.

When asked about the progress, Zuckerberg confidently stated:

This is a very significant advancement. It is a considerably more advanced pre-trained model.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.