Red Xiao and the art of starting again

The restaurant bill is an unusual place to begin an AI story. Red Xiao has a theory about it. At university, he recalled, he could afford to take talented classmates out for Northeastern Chinese food. Recruit the same people after graduation, he joked, and the restaurant might need a Michelin star. It is a founder’s calculation with an excellent side effect: everybody gets dinner.

The anecdote gives Xiao a more useful introduction than an acquisition price. He knows that a product requires people, that people have choices, and that timing changes the cost of almost everything. By the time Manus appeared in March 2025, he had spent roughly a decade learning those lessons through companies that made rather different kinds of software.

His public name is Red Xiao. His Chinese name is Xiao Hong, 肖弘. He is the founder and CEO of Manus, the AI agent company operating from Singapore. The route there passed through a Wuhan university, software for the WeChat ecosystem and a browser assistant called Monica. It also passed through repeated starts. Anyone looking for a straight line should probably bring a pencil.

The tuition was dinner

Xiao entered Huazhong University of Science and Technology in 2011 to study software engineering. He joined its student innovation team and became a deputy leader. The team worked on campus products, including help with university applications and a student marketplace. These were projects with an audience close enough to complain in person.

That proximity matters. Campus software has no need to invent an abstract user. The user may be sitting at the next table, asking why something takes six clicks. A student team gets a compact education in the difference between a clever idea and something another person will bother using. The restaurant anecdote fits that world: collaboration could begin before anybody needed a recruitment department.

In 2015, while still in his final university year, Xiao registered his first company, Nightingale. Graduation followed that year. He went directly into entrepreneurship, exchanging one kind of deadline for several others. There would be customers, cash and a product to finish. The ceremony had an end date; the new workload was less considerate.

Years later, when he returned to speak at his university, he encouraged students to solve concrete technical problems and make good decisions when things were going well. It is revealing advice. Success, in that telling, requires judgment too. A little money and applause do not relieve a founder of the obligation to think.

A business in somebody else’s backyard

Nightingale developed Yiban Assistant and Weiban Assistant, tools for people working in WeChat’s publishing and business ecosystem. This was a useful place to learn commercial software. Publishers needed help producing and managing content. Businesses needed help handling customer relationships. The work already existed; a tool could make it less cumbersome.

It was also a setting with an awkward ownership question. WeChat supplied the platform. An independent company supplied an application within it. What happened if the platform changed its rules, or offered a similar feature itself? Investors could ask that question in a meeting. A founder had to answer it while keeping a business running.

Xiao’s relationship with ZhenFund partner Liu Yuan runs through those years and into his later ventures. Liu has discussed five investments in Xiao. Another conversation follows their relationship across Nightingale, Monica and Manus. A repeated investment is a more interesting character reference than a congratulatory post: the investor has had time to see the less photogenic parts.

The pattern deserves attention because it reappears in AI. An application built around another company’s technology must demonstrate why customers need it. Every improvement upstream can help the product, but it can also change the competition. Xiao had encountered the argument before Manus had a name. He had already learned to work while the ground rules belonged partly to somebody else.

Four million people, one browser window

In 2022, Xiao founded Butterfly Effect and began another chapter in AI applications. Monica put an assistant into the browser. In 2023, the company acquired the ChatGPT for Google extension. The browser was a sensible address for practical software: people were already there, reading, searching and trying to finish things.

At Xiao’s October 2024 university appearance, Monica’s audience was reported as four million users. That figure preceded the Manus launch. It makes the company’s later visibility easier to understand. An application business and a route to users were already in place. The agent arrived after a considerable amount of quieter product work.

The distance between a model’s capability and a user’s convenience can be surprisingly long. Someone may have to copy text, explain its context, move an answer into another tool and check whether it works. Each step is small. Together, they become the job. Xiao’s companies have repeatedly occupied that inconvenient space between possibility and everyday use.

His appetite for an international market came with some self-mockery. He joked that his team’s English had peaked in high school. He also argued that a larger global market could pay for the founders’ education as they learned to serve it. The ambition included an admission that experience would have to be acquired on the way.

“the market itself will provide the tuition fees for founders to learn.”

Red Xiao, translated interview excerpt

The awkward part after the demo

Manus launched in March 2025 with a promise to carry out computer tasks. Research, code and finished documents gave the idea a concrete shape. An agent takes a request through a sequence of actions, rather than leaving all the intermediate work to the person who asked. It is a tempting proposition for anybody with a browser full of unfinished business.

The early experience also produced an instructive problem. In its June 2025 product reflection, Manus said some users did not know which complex task to assign. Early execution was costly and slow. A tool with broad capabilities could still leave somebody staring at an empty input box. Apparently even a digital assistant benefits from a properly explained job.

The response included a conversational mode and playbooks with examples. That is familiar application work: helping people get started, narrowing the first decision and making experimentation affordable. The lesson sits comfortably beside Xiao’s earlier tools. A new technical capability still has to meet the habits of the people expected to use it.

On December 17, 2025, Manus reported more than $100 million in annual recurring revenue. Its total revenue run rate, including usage-based and other revenue, exceeded $125 million. These are annualized measures, rather than a statement that the company had already collected that amount over a completed year. The distinction matters. A business can grow quickly without its accounting vocabulary becoming interchangeable.

An exit with another entrance

On December 29, 2025, Manus announced that it was joining Meta. Xiao welcomed the prospect of a stronger foundation for the product. The announcement said Manus would continue selling subscriptions through its app and website and continue operating from Singapore. For a company aimed at a global audience, Meta offered an obvious route to a much larger one.

The transaction did not remain a neat final chapter. On April 27, 2026, Chinese authorities prohibited the foreign acquisition of Manus and required the parties to withdraw. The decision turned an announced exit into another operating problem. Software may travel easily; ownership has a more elaborate itinerary.

In August, Manus told users about the transition back to an independent company. Some accounts needed data backed up and restored. That practical detail belongs in the story as much as the deal itself. For a user, a company’s corporate arrangements eventually become a very ordinary question: can I still open my work?

On September 1, 2026, Manus formally resumed independent operations. Its founding team continued to lead. Later that month, reporting described a proposed $500 million financing, with terms still under discussion. A proposed round belongs in a different column from money already raised. Xiao’s next chapter was underway, but a headline could not settle all its terms.

The people still in the room

There are quieter scenes in Xiao’s public record. He wrote about hosting a young contributor, Akira, and his family at the Singapore office, presenting a Best Young Contributor award. In January 2026, students from his university visited Manus during a Singapore study trip. Xiao discussed the company’s development and encouraged them to explore their interests and keep learning.

These encounters bring the scale back down to people who can fit around a table. The student founder who once bought dinner was now explaining his choices to another generation. A large company story can make its participants look like job titles. These moments restore a few faces.

On September 28, Manus introduced version 2.0, including a workspace for creating and editing, and tools for video and games. Its emphasis was the work after an initial result: the revisions that make something usable and personal. That is a fitting subject for Xiao’s company. He has spent years dealing with what happens after a capability first appears.

Xiao shared the launch with two words: “We’re back.” They make a modest announcement beside the length of the career behind them. There is still software to improve, a company to run and somebody waiting for the next useful result. The restaurant bill has grown more complicated. The business of getting people together to build something continues.