How to empower a "one-person company"? Chengmai Technology Chairman Wang Jiping: Deployment is difficult, use cases are limited, and high token costs urgently need to be addressed

Nominated by Jingjing News reporter Zheng Yu, Nanjing report

“The emergence of agents is the biggest technology opportunity after smartphones, but the challenges that need to be overcome in their development cannot be ignored. Deployability is difficult, use cases are narrow, and token spending is high—these three pain points are the urgent issues that need to be solved right now.” Wang Jiping, Chairman of Chengmai Technology (300598.SZ), said in a recent interview with reporters from the “China Business Journal.”

As OpenClaw has fueled a surge in OPC (one-person company) entrepreneurship, how to provide lower-threshold services to OPC has become the key. Public information shows that recently, multiple provinces and cities—including Shanghai, Anhui, and Jiangsu—have issued relevant policies to support the development of crayfish agents and OPC.

This has also prompted the market’s “water carriers” to step up product supply. Recently, Chengmai Technology launched an AI assistant for individual users, a locally deployable product for enterprise users, and a crayfish aggregation platform for the consumer market.

In response to the core pain point of high token spending, Wang Jiping proposed two cost-reduction paths: first, replacing cloud calls with local deployment to avoid continuously accumulating cloud call fees. Second, by aggregating on the platform to dilute the cost per single call, achieving scale benefits of “the platform bears the cost of one call, while multiple users share the results.”

Supermarket model for cost reduction

“China Business Journal”: You’ve mentioned multiple times in public that the high fees generated by crayfish cloud model calls are a user pain point. Can you explain how that bill is calculated? How much cost can your solution reduce?

Wang Jiping: Now, when using crayfish cloud LLMs, the billing unit is the number of calls. Light use may be on the order of tens of yuan per month for users; once usage increases, spending of several thousand yuan per month—or even tens of thousands of yuan—is not unusual.

The aggregation logic is a one-time investment. The LLMs and agents run locally, so there’s no need to keep paying call fees to the cloud. In the short term, it’s a hardware expense; but over the long term, the savings in call fees will exceed the hardware cost. Of course, the premise is that users’ usage frequency is high enough, and that’s also the issue we need to solve in market promotion.

The aggregated interface is like a supermarket. The supermarket places all suppliers’ goods in one location. Users buy different items without needing to run a bakery, a dairy plant, or a vegetable base separately. For an aggregated interface, what is placed on the shelves is not the products themselves, but the “usage rights” of each AI large model.

In the past, if users wanted to use OpenAI’s models, they had to register an account with OpenAI, top up funds, and obtain an API key. An aggregated interface is like opening a supermarket between these large model vendors and users: users only need to connect to this single entry point. What models are connected behind the scenes, how scheduling works, and how fees are settled are all handled uniformly by the “supermarket.”

From a product solution perspective, through an enterprise-local deployable product called “Dragon Box,” the large model can run locally, and a one-time hardware investment can avoid continuously accumulating cloud call fees; in the long term, the overall cost is lower than a pure cloud solution. Through “Dragon Palace,” by unifying and processing high-frequency public needs such as information aggregation and then distributing them to a large number of users, it can achieve the scale benefit where “the platform bears the cost of one call, while multiple users share the results.”

“China Business Journal”: This time, your new product is also intended to solve the cost issue, but during the platform cold-start stage, a paradox is common: without users, basic developers don’t come; without developer content, it’s hard for user value to be demonstrated. How is the developer ecosystem of “Dragon Palace” progressing right now?

Wang Jiping: This structural dilemma is real. “Dragon Palace” currently has several agents listed on the platform, mainly seed applications we developed in-house. One of them is an information aggregation agent. The number of trending news headlines in the market each day is basically limited. The platform processes it once in a unified way and distributes it to all users, which saves a large amount of expense compared with each user separately calling the large model to process it. The platform spends 50 yuan to process it and can meet the needs of 1 million users. But if each person calls independently, the total would be 50 million yuan. Behind this is the basic logic of “turning shared public needs into a shared resource.”

In addition, we have already partnered with a professional legal institution, integrating its capabilities for drafting and reviewing contracts and providing legal consultation into “Dragon Palace,” as a public agent aimed at the general public, and also offering paid services. Professional agents in vertical domains such as electricity, fire safety, and stock analysis are also being connected one after another.

The strategy for cold start is to first create persuasive benchmark case studies within vertical scenarios, using commercialization results to attract more professional institutions to enter. There is no shortcut in this process; it requires enough time.

Don’t be an entry point—be a pipeline

“China Business Journal”: Currently, leading well-known companies in the industry are all deeply investing in the agent direction. Compared with them, what is your differentiated route?

Wang Jiping: Directly competing head-on with big vendors is neither realistic nor necessary. The advantage of big vendors lies in the large-scale operations of general cloud platforms and foundational large models. We focus on local deployment and end-to-end integration of software and hardware—these are two different capability systems.

Our positioning is not to compete with big vendors for the “entry” to agents, but to provide the “pipeline” for users to use agents. No matter which vendor’s large model the user chooses, as long as they call and manage it through our product, we have room to exist with value. This logic is similar to the role of aggregated interfaces: we don’t produce content, but we provide a more convenient and more economical path to use it.

“China Business Journal”: Recently, relevant industry associations have issued warnings about agent risks, and some institutions have started to restrict employees from using agent tools. How do you view the current tests facing agents in terms of safety and data privacy protection?

Wang Jiping: At this stage, the condition of agents is similar to a child who hasn’t been sufficiently trained. If someone asks it for sensitive information, it may respond truthfully without any safeguards. The solution lies in training—so that agents have the ability to recognize sensitive information and awareness of how to protect it. This requires continuous data accumulation and iterative model development. The entire industry is exploring this, and there is no one-size-fits-all solution.

Looking from a longer time horizon, the regulator’s attention to data security precisely reinforces the demand logic for the local deployment model. The core concern of institutions is the uncontrollability after data leaves. This is exactly the problem the “Fusion Box” is trying to solve.

At the current stage, users have independent decision-making power over the data they put into the box. This means the corresponding risks are also judged by users themselves. This is an objective limitation in the product’s current maturity. We recommend that, at this stage, users should not put highly sensitive data into any agent system, including our product.

(Editor: Xu Lu; Reviewer: He Shasha; Proofreader: Yan Yuxia)

Massive information, precise interpretation—exclusively on the Sina Finance app