Editor's note, 1 October 2026: This story was published in March 2026 and ends with Meta's purchase of Manus. The deal did not last. On 27 April 2026 China's National Development and Reform Commission prohibited it and required the parties to withdraw the transaction, TechCrunch reported, and on 11 August Manus said it would resume operations as an independent company, according to Caixin.

On March 6, 2025, a short demo video began circulating on Chinese social media. It showed an AI agent doing something that no software had convincingly done before: it received a vague instruction, broke it into steps, opened a browser, searched the web, wrote code, read documents, and delivered a finished result. No hand-holding. No step-by-step prompting. The agent simply worked.

Within 48 hours, the video had been viewed tens of millions of times. Invite codes to access the product were reportedly resold on Chinese platforms for up to 100,000 yuan, roughly 13,900 dollars. The product was called Manus. The company behind it was a Singapore-registered startup named Butterfly Effect. And the story of how it got there, and where it ended up nine months later, is one of the more remarkable chapters in the short history of artificial intelligence as a commercial product.

The Name and What It Means

Manus is the Latin word for hand. The choice was deliberate. The founders wanted to signal something specific about what their product was: not a mind that thinks, but a hand that acts. The distinction matters more than it might seem. The AI products that had captured public attention before Manus, the large language models from OpenAI, Anthropic, and Google, were fundamentally conversational. They answered questions. They generated text. They explained things. They were, in the most literal sense, mouths.

$100M ARR
Annual recurring revenue reached by Manus within 9 months of public launch, a growth rate that placed it among the fastest-scaling SaaS products in history

Manus was designed to be something different. It was designed to do things. To open a browser and search for information. To write code and run it. To read a spreadsheet, analyze it, and produce a report. To take a task that a human might spend an afternoon on and complete it autonomously, without requiring the user to supervise each step. The name was a thesis statement about what AI should actually be.

The Company Behind It

Butterfly Effect, the parent company of Manus, was founded in Beijing in 2022. The key figure behind the project is Xiao Hong, a Chinese entrepreneur who had previously built productivity tools for enterprise users, including AI-driven platforms integrated into WeChat. His approach, which he described as "kitbashing," involved taking existing platforms and enhancing them with AI capabilities rather than building from scratch. It was a pragmatic, product-first philosophy that would define how Manus was built.

The other prominent figure in the company's public story is Yichao Ji, known as Peak, a 32-year-old who became the face of Manus in international media. Ji spoke fluent English, gave interviews to MIT Technology Review, and was named to the Innovators Under 35 list in 2025. He was the person who stood on stages and explained what Manus was to audiences who had never heard of Butterfly Effect.

The founding team was assembled in part through the matchmaking of ZhenFund, an early investor, which connected the technical and product talent that would build the initial version of the agent. By the time Manus launched publicly, the company had already received backing from Tencent and HSG, formerly known as Sequoia China, in a ten million dollar seed round.

The Beta Launch

March 6, 2025 was the official launch date. It was not a public launch. Manus launched in private beta, with access granted through invitation codes only. This scarcity was partly practical, partly strategic. The infrastructure required to run an autonomous AI agent at scale was expensive, and the team needed to manage demand carefully. But the invite-only model also created exactly the kind of social pressure that drives viral adoption: the product was visible, people wanted it, and most people could not have it.

The demo video that went viral showed the agent performing tasks that felt genuinely new. It screened job candidates by reading resumes and generating assessments. It planned a vacation by searching for flights, hotels, and activities and producing an itinerary. It analyzed a stock portfolio by pulling financial data and writing a structured report. Manus claimed at the time that it outperformed OpenAI's Deep Research on several benchmark tasks. Whether or not that claim held up under scrutiny, it landed in a media environment that was hungry for evidence that AI was becoming genuinely useful rather than merely impressive in demos.

The reaction was not uniformly positive. TechCrunch published a piece in March 2025 questioning whether Manus was really China's second DeepSeek moment, noting that the product suffered from bugs and limitations that the viral demo had not shown. This was fair. The early version of Manus was slow, sometimes unreliable, and prone to getting stuck on tasks that required judgment calls the agent was not yet equipped to make. But the underlying capability was real, and the team was moving fast.

The invite codes were selling for $13,900 on secondary markets. That is not hype. That is a market telling you something is genuinely different.
The signal that separated Manus from the wave of AI product launches in early 2025

The Benchmark That Mattered

In April 2025, just weeks after launch, Benchmark, the venture capital firm that had backed Uber, Twitter, and Snap, led a seventy-five million dollar funding round that valued Manus at five hundred million dollars post-money. Benchmark general partner Chetan Puttagunta joined the startup's board. This was not a routine investment. Benchmark is one of the most selective firms in Silicon Valley, and its decision to lead the round at that valuation, for a company that had been publicly available for less than two months, was a signal that serious money believed Manus was building something that mattered.

From Invite-Only to Open Registration

In May 2025, Manus opened public registration. The transition from invite-only to open access was managed carefully. The team had spent the two months since launch fixing the most critical bugs, improving the agent's reliability on complex tasks, and scaling the infrastructure to handle significantly more users. The public launch was not a second viral moment in the way the original demo had been, but it was the point at which Manus began accumulating users at the scale that would define its trajectory.

The product that users encountered in May was meaningfully better than what had launched in March. The agent was faster. It handled a wider range of tasks. It failed less often on the kinds of multi-step workflows that had tripped up the early version. The team had also made a decision that would prove commercially important: they built a subscription model. Users who wanted full access to the agent's capabilities paid a monthly or annual fee. This was not the freemium model that most AI products had adopted. It was a bet that Manus was useful enough that people would pay for it directly.

The Numbers

$100M+
Annual Recurring Revenue, reached in under 9 months from beta launch

By December 2025, nine months after the initial beta launch, Manus announced that it had signed up millions of users and was generating annual recurring revenue of more than one hundred million dollars. The company had also relocated its headquarters from Beijing to Singapore earlier in the year, a move that reflected both the international ambitions of the team and the increasingly complex regulatory environment for Chinese AI companies operating globally.

The growth from zero to one hundred million dollars in annual recurring revenue in nine months is, by any measure, an extraordinary achievement. For context, Slack took approximately three years to reach that milestone. Zoom took roughly four years. Manus did it in less than one. The comparison is imperfect, because the AI market of 2025 was a different environment than the enterprise software market of the 2010s, with different distribution dynamics and different willingness to pay. But the speed of the growth was real, and it reflected something genuine about the product's utility.

The Acquisition

In late December 2025, Meta Platforms announced that it had agreed to acquire Manus. The deal was valued at more than two billion dollars, the valuation Manus had reportedly been seeking for its next funding round. For Mark Zuckerberg, who had staked Meta's future on artificial intelligence and was under increasing pressure from investors to show returns on the company's sixty billion dollar infrastructure spending program, Manus represented something rare: an AI product that was actually making money.

Meta said it would keep Manus running independently and integrate the startup's AI agent technology into Facebook, Instagram, and WhatsApp, where Meta's own chatbot, Meta AI, was already available to hundreds of millions of users. The combination of Manus's agent capabilities with Meta's distribution would, in theory, put autonomous AI agents in front of more people than any other product in the world.

The acquisition was not without complications. Senator John Cornyn, a Texas Republican and senior member of the Senate Intelligence Committee, raised concerns about American capital flowing to a Chinese concern, citing Benchmark's earlier investment. Meta moved quickly to address the regulatory dimension, telling Nikkei Asia that after the acquisition, Manus would have no ties to Chinese investors and would discontinue its services and operations in China. In January 2026, Chinese authorities announced an initial regulatory review of the deal, examining whether AI technologies developed by Manus while it was based in China fell under national security or technology export regulations.

One Year Old

On March 6, 2026, Manus turned one year old. The company marked the anniversary with a post that acknowledged the improbability of the journey. One year earlier, it had been a private beta with invite codes trading on the secondary market. Now it was part of one of the largest technology companies in the world, with millions of paying subscribers and a product that had genuinely changed what people believed AI could do.

The story of Manus is not primarily a story about technology. The technology was impressive, but it was not magic. It was built on top of existing large language models, orchestrated through careful engineering, and improved through rapid iteration based on real user feedback. What made Manus remarkable was the combination of product clarity, execution speed, and commercial discipline that the team brought to a problem that many other companies were also working on.

The name, hand in Latin, turned out to be exactly right. In a field full of products that talked about what AI could theoretically do, Manus built something that actually did it. That turned out to be worth two billion dollars. And it took nine months.