Manus and Lin Junyang both return, pointing in the same direction

Original Author: Joanne

Original Editor: Su Yang

Original Source: Tencent Technology

On August 12, two high-profile AI projects and figures that had drawn significant attention in the AI community both announced their return.

On August 11 local time in the US, AI startup Manus notified users that it would "soon resume operations as an independent company." As part of its separation from Meta, data generated by some users on or after December 29, 2025, will be deleted in accordance with regulatory requirements.

According to the company's notice, affected users need to complete data backups before 7:59 AM Singapore Time on August 23. Between August 23 and August 24, the relevant data will be deleted, and affected accounts will be temporarily inaccessible. Starting from 8:00 AM Singapore Time on August 25, users can restore their previously backed-up data and resume using the service.

Manus stated that this adjustment is not due to a data breach or security incident, but rather a measure taken to comply with regulatory requirements in specific jurisdictions during the company's transition back to independent operations. Unaffected users need not take any action and can continue using the service normally.

A few hours later, another significant return announcement came from Lin Junyang. The former technical lead of Alibaba's Qwen project announced the launch of Pragmatik (p7k) Labs (Yuyong Technology) in Shanghai, an AI laboratory focused on research into next-generation agents for both the digital and physical worlds.

Both are agent-focused projects—one concerning a company's return to independence and the other a technical leader's new beginning—but both point to the same shift: AI competition is extending beyond model capabilities themselves to how models can invoke tools, adapt to environments, and accomplish real-world tasks.

Meta Acquisition Enters Termination Phase

The deal between Manus and Meta began on December 29, 2025.

On that day, Meta announced its plan to acquire Manus. The company, founded in 2022, initially developed in China before relocating to Singapore, primarily building AI agent products.

Neither party disclosed the transaction value at the time, but earlier reports indicated the deal was valued at approximately $2 billion, with a potential total value of up to $2.5 billion including employee retention arrangements.

Following the acquisition, Meta planned to apply Manus's AI agent technology to its consumer and enterprise products to strengthen its AI business capabilities.

However, the deal subsequently drew regulatory scrutiny.

On April 28, CCTV News reported that a decision had been reached in the closely watched Manus acquisition case. The Security Review Office of the Foreign Investment Security Review Mechanism (National Development and Reform Commission) issued a prohibition decision on the foreign acquisition of the Manus project in accordance with laws and regulations, requiring the parties to unwind the acquisition.

Since then, Manus and Meta have proceeded with operational separation and ceased data sharing between the two parties.

According to earlier information, the two parties completed the operational separation in May. Currently, Manus is finalizing the last phase of data processing and preparations for independent operations.

This data adjustment primarily involves data generated on or after December 29, 2025.

Manus stated that the data deletion is required to meet regulatory obligations. The company has launched data backup and recovery tools to help affected users preserve their task records.

Affected users can perform multiple backups during the backup window. If users generate new task data after completing their first backup, they will need to back up again to ensure the latest data is saved.

Manus said it will not charge affected users during the backup period. After account restoration, the company will also offer return incentives.

For users who registered with Apple ID or Facebook accounts, since Manus may not have their corresponding email addresses, the company reminds users to pay attention to in-app notifications.

Manus Founding Team and Product Roadmap

The company behind Manus is Butterfly Effect, founded in 2022.

Founder Xiao Hong previously worked in enterprise software development for an extended period. According to public records, he studied Software Engineering at Huazhong University of Science and Technology. After graduating in 2015, he founded Wuhan Nightingale Technology, whose WeChat tools "Yiban Assistant" and "Weiban Assistant" serve more than 2 million business users.

Butterfly Effect subsequently launched Monica, an AI assistant product integrating multiple language model capabilities. Manus then extended the company's AI capabilities further into the domain of task execution.

Manus is positioned as a general-purpose AI agent platform, allowing users to direct the system through natural language instructions to complete tasks such as information retrieval, report writing, file processing, and code development.

Co-founder and Chief Scientist Ji Yichao previously developed an iPhone browser and founded Peak Labs, focusing on information extraction and search technology. Co-founder Zhang Tao previously led product-related work at ByteDance and Lightyear.

On March 6, 2025, Manus was officially released to the public.

Rapid Growth Before Acquisition; Future Equity Structure Still Unresolved

Before being acquired by Meta, Manus had completed multiple funding rounds and grown rapidly.

Earlier in 2025, the company raised $75 million from Benchmark, reportedly at a post-money valuation of approximately $500 million.

In December 2025, Manus stated that eight months after product launch, its annualized recurring revenue exceeded $100 million, with total revenue run rate reaching $125 million.

The company also disclosed that the platform had processed over 147 trillion tokens cumulatively and created more than 80 million virtual computers.

When Meta acquired Manus, both parties hoped to further expand the AI agent business. However, with the deal now unwinding, Manus's future equity structure and funding arrangements have yet to be announced.

According to earlier reports, during the process of unwinding the deal, the founding team explored raising approximately $1 billion to buy back company shares.

Some early investors also participated in related discussions, including institutions such as ZhenFund and Hony Capital.

Manus stated that after resuming independent operations, the company will continue to serve global users and plans to launch new product features. It will further enhance its AI agent capabilities in the future, though no specific product roadmap has been disclosed.

Regarding data storage, Manus said that after becoming independent, user data will continue to be stored in the United States and Singapore.

From Training Models to Building Agents: Lin Junyang's New Exploration

Manus's return to independent operations means this startup built around general-purpose agents will regain the freedom to independently adjust its product and funding strategies. On the same day, another key technical leader who once spearheaded major domestic large model R&D—Lin Junyang—also set his entrepreneurial direction toward agents.

Lin Junyang announced his departure from Alibaba on March 3, 2026. Previously, he had been long involved in large model R&D, focusing on reasoning capabilities, agent training, and model-environment interaction throughout the development of the Qwen series of models.

The shift from reasoning models to agent systems signals that the focus of AI competition is changing. Future technological breakthroughs will come not only from the models themselves, but also from how models integrate with tools, environments, and real-world tasks.

Lin Junyang has focused his new research direction on agent systems. The name Pragmatik Labs derives from "pragmatics" in linguistics. Lin explained that he initially studied linguistics because a friend recommended pragmatics, after which he turned to computational linguistics and natural language processing. "Pragmatik" represents returning to where things actually happen, while also reflecting the team's understanding of pragmatism—that AGI ultimately needs to solve real-world problems.

Pragmatik Labs will concentrate its research on two directions: digital agents and physical agents.

In the digital domain, the company aims to build general-purpose agents for knowledge work, enterprise operations, and industry processes, enabling AI to handle more complex work tasks. In the physical domain, the company is exploring embodied intelligent systems that can enter real-world environments, adapt to changes, and execute long-term tasks.

Pragmatik Labs describes on its website that next-generation agents need capabilities in reasoning, tool invocation, learning from feedback, and coordinated action. Unlike traditional models that primarily generate information, agents need to operate continuously in their environments, adjust strategies based on outcomes, and accomplish long-term goals.

This direction aligns with Lin Junyang's previously articulated view of "training models → training agents." In his perspective, reasoning models address how to enable more effective internal computation before responding, while agent systems focus on how models can take action in real-world environments. Future AI systems will need to handle issues such as tool selection, task planning, environmental feedback, failure correction, and multi-round task collaboration.

Pragmatik Labs is currently building around this direction, aiming to extend AI capabilities from digital information processing into real-world tasks.