"Taobao of the AI world" MuleRun: 210,000 users flocked in within 10 days of launch, aiming to become the world's largest labor outsourcing company
“Let AI mules do those repetitive, trivial, and exhausting tasks, and humans can do more 'human' things.”
On September 16th, the world's first AI Agent trading market, MuleRun, was officially launched and opened to all users. The logo of MuleRun is a pixel-style mule, and the platform integrates multiple Agents of different types.
Most of the Agent creators are experienced people in various fields who understand a specific process. They turn their skills into workflows and then create Agents. Users can find the corresponding Agents on the platform according to their own needs, rent them as needed, and pay on demand. For example, the 3D desktop character creation is actually an Agent based on Nano Banana, which costs 50 credits (about $0.5) each time. As a third-party platform, MuleRun is responsible for traffic, transactions, collecting US dollars, and continuous dividend distribution. Therefore, MuleRun is also called the Taobao or Xianyu in the AI world by many people.
One very crucial point is that creators can make money here. A typical case that has been mentioned many times - the 3D desktop character creation mentioned earlier. When Nano Banana was gaining popularity, the creator deployed this Agent on MuleRun. Users only need to upload a photo and then click "run", and they can generate a figurine picture in two steps, which allowed the creator to earn $1200 in 3 days.
A small and specific trendy demand, combined with an extremely low usage threshold, brings both traffic and money. According to the MuleRun team, as of September 25th, the number of registered users on MuleRun has reached 210,000, and more than 4,000 people have applied to register as creators, with more than 500 of them having passed the review.
Chen Yusen, the founder and CEO of MuleRun, actually understood this point long ago - the real AI that can break through the circle is not to show how powerful AI is, but to solve a specific and real problem.
●MuleRun promotional image
Chen Yusen was once a top hacker in China. When he was an undergraduate in the Qiushi Science Class of Zhejiang University's Zhukecheng College, he formed a team to participate in the CTF network security technology competition. Later, he joined the internationally renowned team "Blue Lotus", which has the most outstanding achievements in this field and is mostly composed of Tsinghua students.
In 2014, Chen Yusen and three other members of the Blue Lotus founded Changting Technology to engage in Web security protection. It was acquired by Alibaba five years later. After that, Chen Yusen successively founded the data security company Serval and the game company Jiaogemaoba. However, he always felt that this track was not big enough and wanted to try in a field with a higher ceiling. So, at the beginning of this year, he started to form a team to start a business in the AI field. MuleRun received early investment from Alibaba.
Shu Junliang, the CTO of MuleRun, and Fu Cheng, the CPO, have also started businesses in the fields of network security and the metaverse respectively. Together with Chen Yusen, the three former CEOs began to look for a larger and newer track.
During the exploration process, they were all very clear that AI is still far from being sufficient in helping people do things because of hallucinations. However, Chen Yusen later found a path with a high degree of certainty. That is to combine a large proportion of SOP with a small amount of model capabilities, such as 80% SOP + 20% large model, which can at least push the Agent to be truly usable, reusable, and monetizable.
Extending this idea, they want to use it to replace "human computer operation work that is repetitive and does not require a very high threshold" and become the world's largest labor outsourcing company. "Let AI mules do those repetitive, trivial, and exhausting tasks, and humans can do more 'human' things: read books, listen to music, accompany cats, and live a life." Chen Yusen said.
During the Yunqi Conference on September 25th, the founding team of MuleRun, including Chen Yusen, Shu Junliang, Fu Cheng, and Li Xinyu, the CMO, for the first time, received a group interview including the "Future Human Laboratory". After that, we asked Chen Yusen and Li Xinyu some additional questions to complete this interview.
The following is the content of the interview, organized and released by the "Future Human Laboratory" -
On opportunities: This is a market with almost no upper limit
When did you start the MuleRun project? How was the direction of the platform determined?
Chen Yusen (CEO of Mulerun): The team started preparing in January and February this year. At that time, we still wanted to do vibe coding. We thought we could make a better bolt.new or lovable, but after we made it, we found that we could only do it about as well as the existing ones, and there was almost no chance for a latecomer. So we started looking for a new direction.
We saw an opportunity where there was already a rich supply of creative Agents, but there was a gap in the Agent deployment and trading platform. So this is where we saw the opportunity.
Many people say you are the Taobao of Agents. What kind of platform do you want to make MuleRun?
Chen Yusen: Our core value and the problem we want to solve in the early stage is that if you have a specific scenario in your work or a high-frequency repetitive scenario in your life, you can find a solution on this platform and use it every day. For example, for the junior recruitment HR position in a company, they post a large number of resumes on Boss Zhipin and LinkedIn every day. This process is actually highly repetitive and can be easily replaced by AI.
We hope to become a standardized application platform for professionals to solve some highly repetitive scenarios in their work. The value of this part is very large, and the penetration rate is very fast. I have many years of experience in To B entrepreneurship, and it was too slow. In the AI era, the penetration rate of To B is obviously lower than that of To C.
However, the current large model capabilities are not yet able to directly use the model as a product to solve 80% of human work. We think this time window will last for 2 - 3 years. To solve specific problems with a high success rate with the current large model capabilities, what needs to be done is a large amount of SOP plus a small amount of large model.
●Screenshot of the MuleRun official website
About half a year ago, I talked with the person in charge of quality control in an autonomous driving manufacturer. He gave me a very surprising figure. They used a large model plus some well - accumulated SOPs to do unit tests within their enterprise, which is the work of writing unit tests when writing code. They generated about 40 million lines of code in about a year, which is equivalent to replacing the work of 200 people in a year and saving the company tens of millions of RMB.
At that time, the coding model capabilities were not as good as they are now. I asked him how he did it when we couldn't even make the code writing stable. The most core thing for them was that they had accumulated a large number of standards in the quality testing process. The Agent they finally produced for writing unit test code was about 80% SOP plus 20% large language model. This gave me a good inspiration, and we said we should solve problems in a similar way stably.
Under such a premise, people who only understand knowledge in a specific field but have no coding ability can turn the knowledge in the specific field into applications in the specific field. This is a huge opportunity. We hope to become the world's largest labor outsourcing or labor supplier in the AI era. This is a market with almost no upper limit.
On key decisions: Framework neutrality and letting some people get rich first
Did you have any key decision - making points during the process of this project?
Chen Yusen: We made a key decision a few months ago. We decided not to invest much in the creator tools and to be "framework - neutral". First of all, most people can now directly use vibe coding to develop Agents. The most user - friendly way for all humans is to clearly describe their needs, and AI will do it directly for you. Let our product manager, Fu Cheng, talk about the framework neutrality.
Fu Cheng (CPO of MuleRun): There are already too many tools in the current AI and Agent market, but this market lacks platforms or platforms specifically for monetization. We don't make a native tool bound to this platform. We just support the things created by other tools to be sold on our platform.
This can save a lot of time and industrial resources. We do a good job in part of the deployment work and the complete transaction link. At the same time, we turn the tool - making companies into our upstream and downstream partners. So we don't want to compete with dify or n8n. We want to do joint promotion with them. Good creators in their communities can get monetization and benefits on our platform. This is a win - win decision, which is quite crucial for us in the early stage when our team is small and resources are very limited.
If we made tools, we might still be in the R & D stage today, and there wouldn't be a situation where the platform is already online and has a large number of user visits.
●At the 2025 Yunqi Conference, Chen Yusen is the second from the left.
How to attract users and Agent creators to settle in?
Chen Yusen: This flywheel must have some initial impetus to start turning in the early stage. You still need to solve some certain problems, including finding some points that excite both sides, or letting a group of people get rich first, so as to promote this thing forward.
At the beginning, we had some seed creators, such as the developer who did the 3D desktop character creation based on Nano Banana. He really earned more than $1,000 in 3 days after the launch. This is a very real case, so it will penetrate into the market of many developers. Moreover, we will hold many workshops for creators offline to let everyone know that there is such a place where you can create things and we can help you sell them globally. We also have an incentive plan. The more you earn, the more incentives we will give you.
In the past, many Agent platforms focused on how to help you develop Agents, but there are actually many options at this level. In China, what we are more concerned about is where the users are. We want to make a platform that is very easy for users to use.
For some previous open - source projects, the threshold for users is very high. For example, you need to download a docker from the cloud, run the environment, and put in the so - called description file of the Agent. This is okay for people who know a little about technology, but for ordinary people, for every additional step, 90% of them will give up and won't use these things. So when you penetrate into a wider group of people, every step of reducing the threshold will bring a ten - fold expansion of the AI user group.
With the current capabilities of large language models, many problems can be solved by Agents. But when I talk to people around me who are not in the AI field and ask them what they have used AI for, they say they have asked a few questions on DeepSeek or Doubao, and they don't use overseas products. Their use of domestic AI is limited to asking a few questions. This cognitive gap is very large. If we can break this cognitive gap, we can greatly expand the user base of many of our Agent products.
We are still in the early stage now. Our infrastructure also needs a certain amount of time to accumulate, and there are not enough good - enough Agents on the platform. We are waiting for a time when the Agents are ready. After a period of time, just like the group - buying war and the e - commerce war in the mobile Internet era back then, we will definitely have to go through a more brutal and difficult battle. Everyone has a deep belief and understanding of this track. We think it is a very worthwhile thing to do. In the future, in this track, if there aren't dozens of manufacturers, it means we have misjudged this thing.
Li Xinyu (CMO of MuleRun): We have officially launched for 10 days, and the total number of users is 210,000. Nearly 4,000 creators around the world have signed up to use our platform, and we are in the process of reviewing them. Their quality is very high, including experts from all walks of life. So we think the bilateral effect has started at this stage, and the pace of adding Agents to our platform will be faster and faster in the future.
What kind of Agents and creators can enter your platform? What is the review mechanism?