Cathay Capital led the new round of financing for DeepWisdom.
On January 14th, Cathay Capital officially announced leading the current round of financing for DeepWisdom (product name: Atoms). The funds will be primarily used for the continuous R & D of multi - agent systems, large - scale product implementation, and global market expansion. As of now, DeepWisdom has successfully completed Series A and Series A+ financing, totaling $31 million.
DeepWisdom is a company centered around "basic agent technology". Its goal is to enable AI to truly complete complex software development.
Atoms (atoms.dev), launched by DeepWisdom, doesn't aim to improve single - point coding efficiency. Instead, it enables AI to directly build, launch, and operate real - world businesses in a team - based manner. By systematizing and engineering the startup process, Atoms significantly reduces the marginal cost from an idea to an operational business, allowing individuals and small teams to have complete startup execution capabilities and upgrading Vibe Coding to Vibe Business.
Atoms is not a traditional AI coding tool. It is an AI - native startup platform capable of running real - world businesses. In Atoms, users don't dispatch a single model but an autonomous team composed of AI product managers, architects, engineers, researchers, and growth and data analysis agents to complete the entire business process from research, design, development to launch and operation. This means that the definition of a "company" is being rewritten: starting a business is no longer a privilege of capital and elites. Anyone can start the golden age of individual entrepreneurs with a vision and AI.
Wu Chenglin, Founder and CEO of DeepWisdom, said, "In the past, we verified the feasibility of multi - agent systems in complex tasks through open - source and academic research. However, the real challenge lies in enabling these systems to operate stably in the real business environment for a long time. What Atoms aims to do is to transform the startup process into a set of engineerable and scalable systematic capabilities, allowing an individual to have a complete and operational startup team. For us, Cathay Capital is not just an investment institution but also a partner who understands the in - depth transformation of 'industry + technology'. They agree with our judgment on the future of multi - agent systems and are willing to accompany us in continuous exploration at the intersection of scientific research breakthroughs, product iterations, and business verification. Such support gives us more confidence to turn the 'one - person company' from an ideal into a scalable reality."
Cai Mingpo, Founder and Chairman of Cathay Capital, said, "In the new global AI wave, teams that can achieve scientific research breakthroughs, product iterations, and business verification simultaneously are extremely rare. DeepWisdom is forging an independent and clear path with its own rhythm and exploration. The value of Cathay Capital lies in promoting Chinese innovative forces like DeepWisdom to the world stage and becoming a systematic solution for global industrial upgrading."
Introduction
In the narrative of the AI industry, people often focus on parameter scale, model capabilities, or whose applications are more dazzling. However, what truly determines how far a company can go is never the tool itself but the underlying organizational structure.
How a team absorbs knowledge, verifies it in practice, and then iterates its cognition through feedback determines whether it can continuously evolve.
DeepWisdom's exploration aims to software - enable this "organizational ability". It doesn't just answer the question "Can AI write code?" but verifies whether different agents can collaborate like a team to organize R & D and operation processes, forming a new form of collective intelligence.
From the initial open - source frameworks MetaGPT and OpenManus to the officially commercialized Atoms, what it presents is a path from experiment to reality and a preview of future production methods…
From a Lonely Experiment to the Globally Popular First - Generation Product MetaGPT
In the wave of artificial intelligence, code generation was one of the earliest miracles to be noticed. Since the emergence of ChatGPT in the winter of 2022, the world has directly felt the shock of "AI can write code" for the first time. Some cheered for the soaring efficiency, while others worried that programmers would be replaced. However, Wu Chenglin and his team posed a more imaginative question: If AI is not just a code - writing assistant but a "virtual team" capable of division of labor and collaboration, can it develop and operate a complete product like a company?
This question didn't come out of thin air but was a persistent idea after Wu Chenglin's repeated verifications. On November 30, 2022, right after the release of ChatGPT, he asserted during a discussion with a group of CEOs/CTOs, "Natural language programming will soon arrive." A few days later, he searched on GitHub for libraries or scaffolds that could support natural language programming but found nothing.
"Code is deterministic, while natural language is probabilistic. The process from probability to determinism is just a translation," Wu Chenglin firmly believed this was an inevitable direction. However, after reviewing more than 70 open - source libraries, including the well - known LangChain at that time, he found that the logic was overly complex with multiple layers of packaging, and none could meet the requirements. Finally, he rolled up his sleeves and decided to write it himself.
It was an almost lonely exploration: He created a project named meta_programming in PyCharm, which means "generating programs with programs".
On June 30, 2023, he launched MetaGPT on GitHub. In July, this "lonely experiment" topped the GitHub Trending list 17 times. Even more unexpectedly, the open - source community spontaneously "crowdfunded" to complete a paper (https://arxiv.org/abs/2308.00352), introducing the concept of MetaGPT to the world.
The core idea of MetaGPT is to assign different AIs to different "positions": Some are responsible for market research and requirement documents, some for system architecture, some for writing code, and some for testing, advancing in a coordinated manner like an assembly line. In this way, the product development process that originally required the cooperation of multiple people is abstracted into a virtual team composed of AIs. Its popularity is not only because it is "fun" but also because it intuitively proves a cutting - edge concept for the first time: AI is not just a single - point agent. It can be organized and team - based, thus demonstrating collective intelligence beyond individuals.
The team has successively open - sourced multi - agent systems such as MetaGPT and OpenManus, which have been widely applied and verified in the global developer community, accumulating more than 150,000 GitHub Stars. Relevant research results have been published in top - tier international academic conferences such as ICLR, NeurIPS, and ACL, providing a reproducible and verifiable technical foundation for its commercialization.
Choose to Do What Is Difficult but Right, and Time Will Be on Our Side
The success of MetaGPT catapulted the team into the spotlight of the global developer community overnight. However, the popularity of open - source projects cannot be directly translated into sustainable commercial value.
From 2023 to 2024, during the capital winter and the early emergence of LLM technology, DeepWisdom was on the verge of running out of cash flow three times. To survive, the team relied on customer payments, renewals, and loans from relatives and friends to keep going. Salaries were never interrupted, and the core members didn't get salary increases or year - end bonuses for three years but still chose to move forward together.
In February 2025, MetaGPTX was launched when funds were extremely tight. Thanks to long - term open - source promotion, industry partners offered help voluntarily. Eventually, the product exceeded $1 million in ARR in its first month, winning the immediate recognition of users. Two years of difficult transformation led to a more solid start. This was also the first time DeepWisdom truly entered the consumer market and achieved the highest traffic among the vibe coding products launched during the same period.
This performance is unprecedented in the Chinese market. Globally, only a few competitors such as Cursor and Lovable have achieved a similar speed and scale. The difference is that Atoms is not a single - point "code - writing assistant" but an "AI team" built on a multi - agent framework, capable of completing more complex tasks and showing greater flexibility and completeness. This means that it not only improves efficiency but also reshapes the boundaries of software development in terms of paradigm.
The Future Path from Tool Application to Organizational Intelligence
Whether it's the open - source popularity of MetaGPT or the rapid commercialization of Atoms, what Cathay Capital values is never the short - term performance of a single product but the main line that DeepWisdom hits: the core logic of AI applications. Code is the most natural application scenario for large models - it has a standard answer and immediate feedback, forming a self - reinforcing flywheel. More importantly, code is essentially an interface between humans and the digital world, but only a few people understand it. What Atoms does is to significantly lower this threshold, allowing anyone to create applications with natural language, just like how short - video platforms allow everyone to become content creators. More importantly, this team doesn't just stay at the conceptual level. Through the global influence of the open - source community and real - world cash - flow growth, they have verified the future vision of an "AI team" as a current reality.
The competition in AI coding is entering deeper waters. Initially, people were amazed that AI could "write faster", helping developers complete code or fix bugs. Then, tool platforms emerged, enabling users to "generate an application prototype with one click". However, the real turning point is coming: AI is no longer an isolated assistant but starts to collaborate in a team - based manner. Different agents assume different roles, balancing and complementing each other, gradually forming a new form of "organizational intelligence".
In this framework, the meaning of the future becomes clearer: Only 3% of human data on the Internet is utilized by LLMs, while the remaining 97% exists in organizations, individuals, and the offline world. DeepWisdom is liberating this dormant productivity from the "black hole" and giving it to every individual with an idea. Through the agent network, these scattered data and ideas will be transformed into real value. This is not just an improvement in efficiency but a redistribution and explosion of productivity.
The old paradigm is: Starting a company requires a long process of recruitment, team - building, communication, and management, just like manually assembling a complex machine, where every step may go wrong and consume a lot of time and energy. The new paradigm is: DeepWisdom brings the "atomized startup" model - you no longer need to build a team but can call on different expert agents like assembling Lego bricks. You are not managing employees but commanding an AI army of super - capable individuals. Starting a business has changed from being resource - intensive to idea - intensive, and the relationship between humans and tools has been upgraded to the relationship between humans and an intelligent collaboration network.
Imagine that in the future, every ordinary person with a unique idea can become an entrepreneur. The concept of a "personal studio" will be completely rewritten: an individual can operate like an efficient company, continuously creating, iterating, and developing. The future business world will no longer be monopolized by large enterprises but will be an ecosystem composed of countless flexible, efficient, and creative micro - intelligent companies that can quickly respond to market demands and fill every small business gap.
More than two thousand years ago, ancient Greek philosophers conceived that the magnificent universe we see originated from tiny, indivisible particles - atoms - that contain infinite possibilities. This idea inspires us that the greatest creations all start from the most fundamental "one".
Today, in the business world, we are also witnessing an "atomic - level" transformation. This new "atom" can be any small idea, a general backend capability, or even an independent startup individual.
Since that lonely open - source experiment in the winter of 2022, DeepWisdom has forged an extraordinary path. This is not only the growth trajectory of a company but also a microcosm of AI evolving from a point - based tool to organizational intelligence. We are standing at the starting point of a new business civilization: an AI universe composed of countless "atoms" is taking shape, where every small idea can ignite productivity, and every individual can unleash the power to change the world...
This article is from the WeChat official account "Cathay Capital Newsletter", author: Cathay Capital. Republished by 36Kr with permission.