This afternoon, a young AI entrepreneur visited Mankun Law Firm for a conversation. His company currently consists solely of himself.

We initially intended to discuss the overseas expansion of AI products. However, our conversation evolved from search traffic to product selection, from one-person companies to AI Agents, and from cross-border payment collection to legal service products. After more than two hours, my primary takeaway is this: while it is indeed far easier to develop a product using AI than in the past, entrepreneurship has not thereby become simpler.

Over the past two years, this individual has developed numerous AI products and is now essentially operating as a solo entrepreneur in overseas markets. He first assesses whether a specific niche demand has sufficient search volume and whether the average transaction value is high enough before deciding whether to proceed. Once the demand is confirmed, much of the coding is handled by AI, and generic modules such as payment systems and user management have long been standardized into templates. A new product can have a relatively complete website built within seven to fourteen days; smaller features can have a demo produced within a few days.

This prompted me to reconsider my understanding of Vibe Coding.

The fact that AI can perform tasks does not mean you know how to execute them

Vibe Coding is generally understood as a process where humans describe their ideas in natural language, allowing AI to complete the majority of coding, debugging, and iteration.

Recently, I have been employing similar methods to revamp the Mankun Law Firm website and our firm’s workflows. Previously, developing a product idea required engaging product managers, designers, and front-end and back-end developers, followed by waiting for scheduling availability. Now, I can directly articulate business scenarios to AI, have it produce an initial version, and then continuously refine it along actual usage paths.

Since the Spring Festival, my perception of AI has undergone a noticeable shift. Previously, I primarily used it for chatting, information retrieval, and auxiliary support. Now, it can genuinely deliver results. Consequently, my expectations have changed: AI should not merely be tasked with creating visually impressive but economically valueless outputs. It must either help us reduce real costs, improve client experience, or generate new revenue opportunities.

This realization is exhilarating. Business professionals who were previously hindered by technical barriers suddenly possess the ability to build products themselves. However, after today’s discussion, I see more clearly that proficiency in Vibe Coding merely grants entry to the field.

This individual shared that he undertook numerous attempts before arriving at his current product direction. Some products had excessively high customer acquisition costs; others had affordable traffic, but users were unwilling to pay. The seemingly clear product selection methodology he now employs was gradually formed only after a series of actual failures.

He studies where competing websites obtain their traffic and observes whether users are willing to proceed to payment. Low traffic volume is not necessarily detrimental; if demand is growing and users’ willingness to pay is strong, it may be more suitable for small teams to enter. Conversely, in a seemingly bustling sector where everyone competes for the same user base and advertising costs escalate, moving faster may result in quicker losses.

This point profoundly resonated with me. Vibe Coding addresses “how to build something,” but it does not automatically answer “what should be built.” If the underlying demand is incorrect, AI will only help you more efficiently create a product that no one needs.

In my recent work on the Mankun Law Firm website, I have had similar reflections. Getting the pages to function represents only the completion of the first step. Issues such as what queries clients will search for, which articles will be indexed by search engines, how Chinese and English pages correspond, where old links should redirect, and whether the consultation entry points are seamless cannot be resolved merely by creating an aesthetically pleasing interface. These issues stem from real business operations and must be validated within actual user journeys.

The Technical Bottleneck Is Disappearing

In the past, the most common obstacle for professionals with product ideas was the lack of a technical team. You had to first document requirements that you yourself had not yet fully clarified, and then strive to persuade programmers to understand them. Each revision of the requirements increased costs. Many ideas were exhausted through internal friction within the team before they ever reached users.

AI has significantly shortened this distance.

Today, entrepreneurs can first use AI to build a minimum viable product (MVP), allow users to engage with it in practice, and then decide whether to continue investing. In our discussions on scenarios combining AI and legal services, developing such products previously often led to premature discussions on organizational structure, fundraising, and development budgets. A more pragmatic approach now is to first develop the one or two core features, review a demo within seven days, and obtain initial feedback within twenty-one days.

I appreciate this pace. The professional services industry has historically tended to conceive of products as heavy undertakings, as if one could not begin without a complete system. In reality, many needs originate from small starting points: a client is unsure how to proceed at a specific moment, and existing professional services are too expensive to cover their needs. By resolving this specific segment first, the product gains the potential for further growth.

Of course, speed comes with its costs. The fact that AI-generated code can run does not mean it is secure, stable, or maintainable. When customer data, funds, payments, access permissions, and professional judgments are involved, testing, review, and the definition of liability boundaries cannot be skipped. While Vibe Coding reduces development costs, it does not absolve entrepreneurs of the operational consequences.

What we discussed at greatest length was not code, but growth.

A remark by this friend left a deep impression on me: Unless one aims to build a world-changing product, the ultimate differentiation of most products lies in their users. Whether you can find users and whether users are willing to pay determine whether it constitutes a viable business.

This statement may sound simplistic, yet it is easily overlooked by product developers. After a feature is built, one must still address search, content, channels, payment collection, refunds, risk control, and customer service. This is particularly true for overseas AI products, where payment channels may suddenly tighten due to business type, content risks, or abnormal traffic patterns; having website traffic does not guarantee successful payment collection. The final step in a technical demonstration is often merely the first step in real-world operations.

This is also why I have always emphasized content and SEO. In my early years working on legal internet projects, I experienced large-scale paid user acquisition. Once traffic bidding enters a vicious cycle, customer acquisition costs rise continuously, eventually exceeding the customer lifetime value, which undermines the business model. Since founding Mankun Law Firm, we have largely avoided paid traffic advertising, focusing instead on consistently producing professional content. It does not matter if an article receives no views on the day it is published; six months or two years later, someone may still find us through that article when seeking solutions to a specific problem.

In the AI era, such content assets have become even more important. Users increasingly consult AI first, and AI then seeks out materials it deems credible. If professional institutions do not continuously publish their case experiences, business judgments, and proprietary information, even superior services may disappear from new search entry points.

Therefore, when developing products through Vibe Coding, one cannot focus solely on development. Code, content, search, payment, compliance, and delivery must be integrated into a complete end-to-end pathway. If any segment is missing, the product may remain stuck in a state of "appearing to be completed."

With AI augmentation, what lawyers need is product intuition.

This exchange also revealed another possibility for small-team entrepreneurship in the AI era.

An individual with industry expertise, paired with someone proficient in AI products and growth, supplemented by necessary professional services, can now accomplish tasks that previously required a team of more than ten people. Teams can form temporarily around specific needs to rapidly build demos, acquire users, and test willingness to pay; upon successful validation, a more stable company can be established, whereas if validation fails, the project is terminated and the team reconfigured. This approach is better suited to the current pace of technological change than first building a large organization and then seeking direction.

However, small teams do not mean an absence of division of labor, nor do they mean delegating everything to AI. It remains essential to clarify who defines requirements, who is responsible for the product, who reviews professional conclusions, who handles client data, and who bears ultimate responsibility. AI can write code, execute workflows, and generate materials, but it will not handle equity matters, intellectual property, data compliance, payment disputes, or client commitments on behalf of the team.

I increasingly believe that the most competitive small teams in the future may not have the largest number of full-time employees, but will certainly excel at combining several capabilities: some members understand business scenarios, others understand products and growth, and others safeguard professional and compliance boundaries. While AI reduces collaboration costs, human responsibilities must be defined even more clearly.

Recently, I have often discussed with young lawyers what they should learn in the AI era. Many people’s first reaction is to learn prompt engineering, programming, or a new tool. These are certainly useful, but they evolve too quickly to constitute a long-term advantage.

A more important capability is understanding a business process: why clients come, where payments originate, how materials flow, which steps are most time-consuming, which steps are most prone to error, and which judgments must be made by lawyers. Only after gaining this understanding should one decide how to integrate AI into the workflow.

This is also the experience I have accumulated while building my own website, operating an AI-enabled law firm, and researching legal technology products. Do not pursue a grandiose system from the outset. Instead, identify a real problem and develop it into a usable minimum viable product.

The most valuable aspect of Vibe Coding is that it returns product development capabilities to those closest to the requirements. Lawyers, accountants, consultants, and operations managers, who previously could only hand off requirements to technical teams, can now build initial versions themselves. This will create opportunities for product innovation across many niche industries.

However, whether a person can effectively leverage Vibe Coding ultimately depends on the depth of their business understanding.

The insight this friend provided me today was not about how impressive it is to "build a website in seven days," but rather that he already thinks about requirements, data, development, growth, and cash flow within a unified framework. AI has enabled him, as an individual, to possess the execution capacity that previously required a small team; and his trial-and-error efforts over the past year or more have shown him where to direct this execution capacity.

I am increasingly convinced that many promising products in the future will not emerge from the conference rooms of large corporations. They may instead originate from a client consultation, a flawed process, or a long-neglected niche need, and be rapidly developed by someone who truly understands the use case, leveraging AI.

As code becomes increasingly inexpensive, product judgment will become ever more valuable.