---
title: "Unique Friends | MuleRun’s Shu Junliang: Agentic AI Has Plenty of Demos—What It Needs Is a Market"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2025-11-07T11:35:00+00:00"
canonical: "https://ffcap.cn/en/research/unique-research-2025-11-07-01"
source: "https://uniqueresearch.substack.com/p/unique-research-2025-11-07-01"
language: "en"
---

# Unique Friends | MuleRun’s Shu Junliang: Agentic AI Has Plenty of Demos—What It Needs Is a Market

_Original · Unique Research · 2025-11-07_

_Editor’s note: This complete English edition preserves the original reporting and Shu Junliang’s interview in full. User figures, developer earnings, product capabilities and compliance practices are claims of the report and interviewee, not independently audited findings. The world’s largest marketplace is MuleRun’s stated positioning and ambition, not a verified ranking. Relative dates and the reference to a market of 7 billion people are retained as stated in the original article. The source identifies the interviewee’s university only as Jiao Tong; no fuller institutional name is inferred._

If the internet taught us one thing over the past decade, it is that platforms can change an individual’s destiny. The wave of Agentic AI beginning to unfold in 2025 may well take this idea a major step further—not simply giving you an account or a stall, but enabling you, on your own, to truly possess an end-to-end loop from capabilities to business.

That is precisely the somewhat audacious undertaking MuleRun has set out to pursue.

It is not merely building another AI application. Instead, it is attempting to rebuild a market for the Agent era: connecting Agent developers around the world on one side with ordinary users willing to pay for intelligent capabilities on the other, so that technology is no longer just model parameters and demos, but products and revenue that can be created, distributed, traded, and scaled.

Behind this is the choice of a quintessential technology entrepreneur: moving from cybersecurity to the Agent marketplace, Shu Junliang shifted from being someone who builds products to someone who builds the stage for others.

I. From Foundation Models to Agents: The Technological Revolution Is Redefining What One Person Can Do

Shu Junliang’s judgment is straightforward: the defining keyword of 2025 is Agentic AI.

The breakthroughs in foundation models have given us, for the first time, an infrastructure we can converse with. In his view, this revolution is changing individuals on at least two levels.

The first is that the barriers to knowledge are being shattered.

In the past, if you wanted to make up for a gap in your education, you either enrolled in a course, wrestled with a thick book, or painstakingly pieced together information through a search engine. Today, anyone with a solid foundation in general science and logical reasoning, as long as they are willing to ask questions, can use conversations with a foundation model to lift themselves up and begin exploring the fundamentals of different fields.

From “I need to go to school to learn” to “I can ask anytime,” the acquisition of knowledge has shifted from linear to parallel.

The second layer is that the boundaries of labor have been rewritten.

When Agents can reliably write code, run tests, write documentation, develop proposals, create images, and edit videos, many tasks that once required a small team to divide up and complete are beginning to become a matter of one person directing the work and evaluating the results.

You do not necessarily need to be proficient at everything, but you can have a group of Agents queue up behind the scenes, freeing your time from repetitive work and allowing you to focus on making judgments, weighing options, and setting direction.

In this sense, “Pioneering Intelligence | The Individual Era” is not merely a polished conference slogan, but a very practical question:

As tools become increasingly intelligent, what truly needs to be redesigned is the individual as a unit—how large can one person become? How much complexity can one person take on? What enables one person to make a name for themselves in global competition?

MuleRun’s story unfolds at the very center of this question.

II. What MuleRun Wants to Build Is Not Another Application, but a Two-Sided Marketplace

MuleRun was founded quite late—it only got started in the first half of 2025—but it moved very quickly: it officially launched to the public on September 15; registered users surpassed 200,000 in less than 10 days and exceeded 500,000 within a month; the platform already had thousands of developers and more than 200 live Agents.

But more revealing than these figures is its positioning:

The world’s largest AI Agent marketplace.

There are two assumptions behind this statement:

Agents will become increasingly powerful and solve more and more real-world problems;

What truly needs to be rebuilt is not another smarter Agent, but infrastructure that connects Agents with users.

Over the past two years, quite a few teams have tried building Agent directories and Bot marketplaces, but why has no truly useful product emerged?

Shu Junliang’s answer is highly pragmatic: most people saw only the traffic opportunity, without digging in to tackle the dirty, difficult work on both the developer and user sides.

For an independent developer or small team, the real pain points of building an Agent are often not whether they can get a model working properly, but:

How do you integrate the various model APIs through a unified interface?

How do you deploy an Agent and ensure reliable access for users around the world?

How do you collect payments overseas? How do you handle settlement?

How should regulations and data compliance be handled across different countries?

How do you turn a pile of scripts into a product that people are willing to pay for—and keep using?

MuleRun’s choice is to treat all of this as part of its own responsibility.

The platform handles model integration, deployment and operations, cross-border payments, infrastructure, and productization tools, allowing developers to spend more of their time figuring out what users want instead of being dragged down by API documentation, servers, and compliance issues.

At the same time, the platform takes responsibility for traffic distribution and user education, bringing useful Agents in front of ordinary people and helping them make the transition from hearing that AI is powerful to actually using several Agents every day to get things done.

This is, in fact, a very straightforward positioning—though one that is extremely difficult to maintain:

It is not about taking developers’ livelihoods; it is about helping them finish cooking the meal and serve it on tables around the world.

III. The Story of Nano Banana: When Insight, Initiative, and Infrastructure Come Together

What does it mean to give individuals leverage? A specific case on MuleRun illustrates this very well.

Last month, Google’s Nano Banana image-generation model went viral overnight. The most popular use case on social media was to upload a photo of an object and have it generate a composite image combining a collectible figurine with a computer design draft—an existing physical object presented as though you were looking at a concept sheet.

Trends like this spread extremely quickly, but the buzz often lasts only a few days.

On the very day the model was made public, a developer on MuleRun had already launched an Agent based on Nano Banana.

The functionality was extremely simple: using a fixed prompt, it sent the image you uploaded to the model, which then generated an image of the figurine plus the design draft. There was no complex multi-turn interaction or lengthy interface—just an ultra-focused little toy.

But this time, he had several key variables working in his favor:

He had a keen enough feel for cutting-edge models to recognize that Nano Banana offered more than better image quality—it represented a new form of expression;

He had enough drive to turn the idea into an Agent that people could click and use at the earliest opportunity;

He no longer needed to build an entire website, payment system, distribution network, and global-access optimization setup himself; MuleRun had already taken care of all of that.

As a result, within just a few days, this Agent received a huge volume of calls on the platform, and the developer made a considerable sum of money from this small product.

Looking back, the amount of work involved in this closed loop was actually quite limited:

There was no long development cycle, no large team, no complex sales process, and the UI was extremely restrained.

The three things truly amplified are intuition, understanding, and the ability to act.

Everything else is leverage provided by the platform.

This is precisely a very important signal in the Agentic AI era:

Individuals no longer need to build every piece of infrastructure themselves. Instead, they need to learn to stand on the right infrastructure and maximize their own insights and execution.

IV. Chinese Developers’ Strengths and Weaknesses: Technology Is No Longer the Biggest Challenge—Product Is

Shu Junliang’s assessment of Chinese teams in the global AI market is, in fact, clear-eyed despite its optimism.

The optimism comes from an obvious fact:

Years of demographic dividends and investment in education have cultivated a vast pool of engineers in China. This has been evident in nearly every technology-intensive competition, and AI is naturally no exception.

On MuleRun, many outstanding Agent creators already come from China, and the products they build are fully capable of holding their own with users around the world.

But he also pointed out very candidly:

For individual developers and small teams, technology itself is no longer the biggest challenge. The real weakness is product sense.

He and his team can see a large number of Agents in the platform backend that are technically sound and built around interesting ideas, yet are unable to make any commercial headway:

The logic is flashy and technically impressive, but users get lost the first time they try it;

There are many features, but not a single memorable core use case;

They have introduced paid offerings, but have barely thought seriously about why anyone would pay that amount of money;

They have usage and early-adopter users, but cannot retain them, much less drive a second purchase.

At their core, these problems have nothing to do with whether the model has been tuned well enough. Rather, they reflect a serious lack of systematic thinking about user experience, paid conversion, and the path to retention.

Taking a broader view, Chinese teams’ participation in global competition over the past decade or more has indeed been relatively limited—not because there have been no projects expanding overseas, but because, viewed as a whole, they have remained oriented primarily toward the domestic market.

This also means that our engineers’ strengths have often been applied only on the chessboard of the domestic market. When truly going global, they still have to develop a new understanding of laws and regulations, payment habits, cultural differences, and users’ mindsets.

In Shu Junliang’s view, for Chinese engineers’ strengths to genuinely become an advantage in global markets, we must first acknowledge and make up for these weaknesses:

Compliance must be learned, product must be learned, operations must be learned, and commercial thinking must be learned as well.

Otherwise, no matter how high your technical proficiency, you are merely someone who can write code, not someone who can make the whole operation work.

V. Three Pointers for Individuals and Small Teams: The Frontier, Global Markets, and Your Weakest Link

Returning to the question of the individual era, if we narrow our perspective to the next 1–3 years, Shu Junliang’s advice is simple, but by no means easy:

First, maintain genuine sensitivity to the frontier.

This does not mean scrolling through the news every day and forwarding a few Demo links; it means putting yourself in the position of asking whether you can turn something into a usable product at the earliest possible moment.

The frontier is not merely information, but a window of opportunity—as with a model on the level of Nano Banana, every appearance creates an arbitrage window: if you discover it early and act quickly, you genuinely have the opportunity to earn in a few days a sum that would previously have taken several months to earn.

Second, set your sights on the world from Day 1 and build a business for 7 billion people.

The inherent advantage of Agentic AI is that one codebase can be used globally. If you lock your vision onto a single market from the outset, you will miss room for maneuver in many aspects of design: language, payments, use cases, pricing, and branding.

The real change lies in your mindset: you are not building a small tool, but developing a solution to a specific need somewhere in the world, and simply choosing to deliver it in the form of an Agent.

Third, use AI to shore up your own weak spots.

For most people with a technical background, those weak spots tend to be business, product, and operations;

for many people who come from a content background, they tend to be technology, structured thinking, and long-term planning.

But the good news now is that you no longer need to treat your weak spots as fate; you can treat them as bugs that can be fixed.

Use large language models to repeatedly refine your product logic, operational strategies, and business models; use Agents to assist you with research, drafting plans, and running experiments, compressing what would originally have required two or three years of trial and error into a few months or even a few weeks.

What is truly worth guarding against is not saying, “I do not know how to do this right now,” but becoming accustomed to using what you do not know as an excuse.

The platform is in place; it is time for individuals to make their choices

From a cybersecurity entrepreneur to the CTO of an Agent marketplace, from exiting through an acquisition to going all-in once again, much of Shu Junliang’s serial-entrepreneur character comes from his sensitivity to the tides, as well as a distinctly technical stubbornness—in his view, when a technology demonstrates potential unlike anything seen in the past, committing oneself to it and continuing to delve deeply into it is itself the inevitable choice.

Whether a platform like MuleRun will truly become the world’s largest AI Agent marketplace remains too early to say. But at least one thing is already clear:

It is trying to bridge the gap between individuals and technology, building out the trivial yet crucial infrastructure first;

It is using one concrete developer story after another to prove that one person + one platform + a little frontier awareness can indeed achieve good results in this era.

The real key lies with every individual:

Do you treat an Agent as a smarter tool, or do you see yourself as an operator capable of harnessing a whole group of Agents?

Will you continue waiting for a major tech company to offer you a position, or start seriously thinking about how you can use AI + platforms to build a value chain of your own?

People are already building the platforms, and the technology has indeed matured enough to give you leverage.

That leaves only one question: This era has already given individuals so much—how big are you prepared to make the word “individual”?

Selected Interview Q&A

Q1: Could you first briefly introduce yourself and MuleRun? What kind of company does it aspire to become?

Shu Junliang: I’m Shu Junliang, and you could say I’m a serial entrepreneur. I completed my PhD at Jiao Tong in 2019, then co-founded a cybersecurity company with some classmates and served as its CEO for 6 years. The company was acquired this year. After that, I chose to go all in on joining MuleRun, where I am now CTO.

MuleRun was founded in the first half of 2025 by a group of serial entrepreneurs who reunited to start a company. What we want to build this time is not a single app, but the world’s largest AI Agent marketplace—a connector linking Agent developers on one side with ordinary users on the other, using commercialization to drive the creation and adoption of more useful Agents.

Q2: What specific problem is MuleRun’s product solving? How is it different from the “Bot stores” of the past?

Shu Junliang: The opportunity we see is simple: in the future, the vast majority of AI applications will take the form of Agents, so there will inevitably be a need for a genuinely useful “two-sided platform.”

For developers—especially independent developers and small teams—the difficulty is not really that they cannot “build an Agent,” but that they have to deal with a pile of tedious, labor-intensive tasks:

How do you integrate different model APIs in a unified way? How do you deploy an Agent as a stable service accessible worldwide? How do you collect payments and settle accounts? How do you turn it into a product?

MuleRun’s value lies in taking care of this underlying infrastructure for everyone, lowering the barriers to development and commercialization, and giving more people the motivation to keep building Agents.

For users, we hope to put more useful, better-performing Agents in front of them, making “using Agents to solve problems” an everyday habit rather than a one-time novelty.

Q3: What were your personal original intentions and motivations for entering the AI field this time?

Shu Junliang: To put it rather directly, I think the potential of this wave of AI technology is simply enormous.

Large language models have demonstrated many capabilities that traditional technologies do not possess, creating opportunities to build entirely different kinds of products. For someone with a technical background, when a technological wave of this magnitude arrives, it is actually harder to bear not diving in.

So my original motivation was very simple: I wanted not merely to “participate,” but to truly accomplish something amid this wave.

Q4: In your view, where was generative AI’s biggest breakthrough in 2025? How is MuleRun integrating these cutting-edge developments?

Shu Junliang: If I had to sum up 2025 in one word, I would choose Agentic AI. It represents AI’s shift from “answering questions” to “handling complex tasks.”

The most notable recent advances have been in multimodality, particularly image- and video-generation models. A considerable number of Agents based on the latest multimodal models have already emerged on our platform, and image- and video-generation Agents have seen very high usage and attention.

MuleRun itself is a marketplace designed around the Agent model, so the faster and more powerful cutting-edge models become, the more beneficial that is for creators and users on our platform.

Q5: With technology evolving so rapidly, how do you balance “technological innovation” and “commercial implementation”?

Shu Junliang: For a platform product, these two things are actually closely intertwined.

Commercial implementation is the basic condition for getting a two-sided platform up and running;

Technological innovation directly determines what kinds of Agents developers can build, how stable those Agents are on the platform, and whether users can enjoy a smooth and seamless experience.

Our team has a strong technical background, so we invest more deeply in technology—for example, we have a dedicated small team working on post-training models and exploring areas such as Agentic RL — reinforcement learning. In the future, breakthroughs in these technologies will also feed back into the product side as features and drivers of growth.

The overall judgment is: AI technology still has enormous room for growth. We will continue pouring resources into frontier exploration, but only on the condition that it genuinely serves developers and improves the user experience.

Q6: MuleRun has been global from the very beginning. What have been the biggest opportunities and challenges you have encountered?

Shu Junliang: From its very first day online, MuleRun has been a product aimed at the global market.

The opportunity is that demand for AI is highly globalized. Users and developers in countries such as the United States, Japan, India, and Singapore are all highly active, and our platform is naturally capable of “cross-border matchmaking.”

The challenges, however, are numerous:

Laws, regulations, and localization requirements differ from country to country, and users in different regions have different preferences regarding Agent types. As a platform, we also have to serve the platform, developers, and users at the same time, which makes our business and compliance challenges more complex than those of a single application.

At present, on the one hand, we are focusing on serving markets with particularly active demand for AI; on the other, we are building localized teams and working with creators around the world to use the platform’s infrastructure, tools, and traffic to support Agents that address pain points in different vertical fields.

Q7: When building an AI platform across countries, data and compliance are unavoidable issues. How do you handle them?

Shu Junliang: The first step is always to gain a clear understanding of your own product: its business model, data flows, and service architecture all need to be clearly defined.

AI products are more complex than traditional businesses because they all rely on large-model APIs at the underlying level. At present, there are only a few mainstream model APIs, and sooner or later, data on your platform will flow to the countries where these providers are based. Europe and the United States each have their own requirements for data compliance, so you cannot handle the matter carelessly or without a clear understanding.

What makes MuleRun more complex is that three parties are active on the platform: the platform operator, developers, and users.

We have therefore invested considerable effort in legal and data compliance. There are several data security experts on our team, and I myself started out in security, so I consider this one of our advantages.

Q8: How do you view the position and shortcomings of Chinese AI teams in the global market?

Shu Junliang: In terms of strengths, I think Chinese teams are highly competitive in the global AI market. The core reason is the engineer dividend: years of investment in education and the importance parents place on education have trained a large number of outstanding engineers for this industry—an advantage that is difficult for many countries to match.

On MuleRun, many exceptionally talented Agent creators come from China, and the products they build are highly competitive when presented to users around the world.

But the shortcomings are also obvious: over the past decade or more, our participation in global competition has been limited, so the abilities of this generation of engineers have not been fully “monetized” in global markets.

The areas where we need to catch up include the laws and regulations of different countries, cultural differences, how to build paid products, and the mindsets and business practices of users in global markets. Once we address these shortcomings, our strengths can truly unfold on the global stage.

Q9: How do you understand the theme “Pioneering Intelligence | The Individual Era”? What is AI changing about individuals?

Shu Junliang: I think this theme captures it with great accuracy. It points to the most far-reaching change brought about by this AI revolution: the upper limit of individual capability has been raised across the board.

On the one hand, large models have dramatically lowered the barrier to acquiring knowledge. Learning by conversing with a model is far more efficient than searching online or listening to lectures used to be. For people who are motivated to improve themselves, it is easy to become a “generalist”: as long as you have basic scientific literacy and logical thinking, you have the opportunity to quickly develop into a “near-expert” in multiple fields.

On the other hand, the rapid development of Agentic AI is giving individuals enormous leverage in “work done through computers.” Writing code, testing, drafting documentation, creating PowerPoint presentations (PPT), making images, and producing videos—tasks that once required a small team to spend a long time working through—may in the future all be delegated to Agents. Individuals will only need to make judgments, issue instructions, and review the results.

That is why I believe we will definitely enter an “era of the individual”: many things that once required a team and a great deal of time to accomplish will eventually be completed by one person together with a group of Agents.

---

Original publication: https://uniqueresearch.substack.com/p/unique-research-2025-11-07-01
On-site reading page: https://ffcap.cn/en/research/unique-research-2025-11-07-01