Just two days ago, singer Hu Yanbin personally used AI to manually create a fan community app, making Vibe Coding a huge hit across the internet.
To be fair, without the keywords "personally," "AI," and "handmade," this matter would probably only generate some buzz in the domestic entertainment news section at best, and it would be quite difficult to get on Weibo's trending topics for free.
After all, fan communities are not a new concept: back in the early 2000s, relatively young singers like Jay Chou and Wang Leehom established fan club websites, aiming to create private traffic pools for fan management in the Web 1.0 era. Now, a quarter of a century later, numerous celebrities from various fields, both domestic and international, have tried their hand at this. Therefore, the buzz surrounding this topic will likely remain at a moderate level.
But just as anything enhanced by AI in recent years seems to have a performance boost, the rather unusual combination of "celebrity + AI + software development" instantly made this a hot topic in both directions. People on both sides were stunned: Huh? Music arrangers are all doing (or doing) programming now?
As we know, where there's traffic, there's usually conspiracy theories: these generally revolve around the core idea of "celebrities riding the AI wave," with typical examples being "it's probably done by an outsourced team, and they're just using their names." However, to be fair, this way of thinking, which immediately assumes the worst, isn't entirely without logical basis: according to common understanding, celebrities in the entertainment industry usually have computer skills focused on product usage rather than development.
However, there are always exceptions, and this logic doesn't seem to apply to Professor Hu Yanbin.
According to a video of Hu Yanbin being interviewed released by Bilibili UP master "Yu Ge Chu Xi Pao" in June 2024, he had already begun to learn to code through online videos at that time, and his tone was casual with a hint of pride.
It's important to understand that two years ago, while large language models already possessed certain software development assistance capabilities, there was still a significant gap between them and today's Vibe Coding experience, which allows "natural language to directly construct complete applications."
Therefore, from this perspective, Mr. Hu Yanbin was indeed systematically learning programming at that time.
Interestingly, the timing of this interview coincides with an important milestone for AI Coding.
On June 20, 2024, Anthropic released Claude 3.5 Sonnet model version, which brought significant improvements in coding capabilities. Some fellow programmers commented on this version: "Oh no, oh no, this might really mean I'm out of a job..."“
Of course, following normal logic, it's unlikely that Mr. Hu Yanbin had already ventured into AI programming back then. He most likely started with "Hello World," otherwise, when he appears on variety shows today, his title would probably include "Contemporary Harness Master" in addition to "singer and songwriter." :)
Another piece of evidence that Hu Yanbin doesn't seem to be approaching this with a casual, amateur mindset is that this fan community isn't just on iOS, but across Web, Android, and iOS—a very complete community system. It effectively ensures that whether you're resting after get off work or slacking off at work, you can immerse yourself in the feeling of interacting with your idol anytime, anywhere.
According to the product details page on the Apple App Store, the developer of the "Yanhuo" app is Shanghai Yunbin Culture Media Co., Ltd. Further investigation using the name of this company (99%), which is controlled by Hu Yanbin himself, through the Ministry of Industry and Information Technology's information filing management system reveals that before the app's launch, this company had already completed the website filing review process as early as April 15th of this year. Note the following:(This woman's name is Xiaomei)The registered domain name is hybfansclub.com, and Hmmm is undoubtedly the web version of Hu Yanbin's fan community.
Uh... we've gotten a bit off-topic. Let's get back to the product. After all, in the internet industry, the product speaks for itself.
As mentioned earlier, fan communities have become a very mature product in terms of both form and function. In terms of functional modules, it basically includes a homepage feed, account system, growth system, voting, user community, in-site messaging, AI assistant, and relatively simple video-on-demand.
These are essentially the most mature and common functions in internet products. From mature modules to open-source code, they can all be found easily. Moreover, these modules do not have very complex backend logic, nor do they include order and payment systems that require strong risk control.
Therefore, the fan community can be said to be the best place for Vibe Coding to be implemented, and it also becomes an excellent sample for observing the boundaries of Vibe Coding's capabilities.
Through initial observation of these three products, we have identified some issues that are worth raising for reference by more cross-industry entrepreneurs interested in personal software development.
1. Lack of hidden function
After trying out the iOS version of the "Yanhuo" App, some netizens raised two issues: the lack of risk control in the user's mobile phone number registration process and the lack of review in the content publishing function.
Simply put, if there are no risk control measures such as user verification during the mobile phone number registration process, it is easy for malicious people to use it to carry out SMS attacks or maliciously deplete your SMS service deposit.
If users don't use sensitive word filtering and undergo automated or human review processes when publishing content, it could pose legal risks to the software operator (who is actually Mr. Hu Yanbin himself). According to relevant Chinese regulations, the operator is responsible for reviewing and managing the platform's content.
So, why does this problem occur?
The reason is actually quite simple. When Hu Yanbin was planning the product's feature design, he probably thought more about the user experience needs and the platform's operational needs: how users register, how the community interacts, how content is displayed, and how the fan base grows.
These are features that users can directly perceive, and they are also the first priority for creators at the feature level.
However, capabilities such as risk control, compliance, auditing, and performance optimization are considered "hidden functions" in internet products.
These features may seem unnoticed by users, but their absence immediately exposes problems. The necessity of these features often requires a comprehensive understanding gained through the entire process of building a complete internet project from scratch, from planning to operation – what we commonly refer to as "Know How."
For a creator who has just transitioned to the role of product manager, expecting him to consider all these things meticulously when creating a product for the first time is simply mission impossible.
As tools that provide Vibe Coding functionality, whether it's Cursor or Claude Code, their core value lies in answering questions, rather than defining and expanding them.
They can fulfill your explicitly stated requirements very well, but they struggle to proactively identify issues that users have never raised. You ask them to develop a community, they develop a community; you ask them to implement check-in, they implement check-in. But they won't suddenly stop and ask you, "Does this community require content moderation?" or "Has your SMS verification code API considered anti-fraud measures?"“
From this perspective, the problems exposed by "Yanhuo" are less about development capabilities and more about product planning. And the latter is precisely the aspect that AI finds hardest to replace today, and also the one that Vibe Coding most easily overlooks.
2. The visual effects experience has not been fully validated.
Next, let's take a look at the web homepage of this fan community. You'll find that its visual effects and color scheme are very immersive, but at the same time, it also surprisingly has some small problems that are immediately visible to the naked eye.
These issues would generally fall under the level of importance below P3 during the testing phase, but they are prominently displayed on the homepage, as shown in the image below:
A quick glance reveals several minor layout issues on this page, including but not limited to: incorrect line breaks, the navigation bar not being aligned upwards, inconsistent spacing between English and Chinese text in the navigation bar, and the new message tab in the news module being pushed to the side to compress the title and body text.
These issues reveal another shortcoming of Vibe Coding: the lack of a forward development process can easily lead to missing essential steps.
Looking at the source code of the website's homepage, the code style does indeed have a distinct AI flavor.
For example, the styles contain many hard-coded dimensions, spacing, and positioning values, and the layout relies more on specific pixels rather than using more robust responsive constraints to adapt to different devices. The code looks like it's designed to make the current page "look right" rather than systematically handling boundary issues related to multiple resolutions, browsers, and content lengths.
The main reason is that AI tends to stick to the bottom line when it comes to writing code, focusing on a green, energy-saving, and resource-efficient token, and prefers to prioritize meeting current needs rather than considering various boundary conditions in advance.
Therefore, it is common for numbers to be hardcoded without considering resolution; this also means that the page may be normal in Mr. Hu Yanbin's own development and testing environment, but when switched to my main typewriter T480 laptop (I can't help it, I'm poor) Chrome @ 1080P resolution, the page is completely distorted.
The problems with AI can be addressed by adding more detailed requirements, but the lack of key workflows is the root cause of the problem persisting even after deployment.
In the traditional forward development model, after the product manager completes the feature definition, interaction designers need to design the layout and create mockups based on the requirements. After the developers develop according to the mockups, they first conduct self-testing; after passing the self-test, they submit it to the testing department; then the testers write test cases and conduct targeted testing.
However, in the process where Mr. Hu Yanbin completed the entire product definition and launch process alone, AI did not play a role in testing in conjunction with visual design requirements. The lack of work in visual acceptance and multi-platform testing led to the page being launched with defects.
3. Statically mount MP4 files directly on the page.
Actually, there was some disagreement with my colleagues when we discussed this point, so including it here may not necessarily be a weakness of Vibe Coding, but rather just an engineering trade-off.
Those who support the current technical approach argue that since these videos are free content and the initial user base is limited, it's not a big deal to not have a complete on-demand system right now; it can be replaced later once VV's data improves. This logic itself isn't flawed.
The other side focused on a new question: Was this choice a decision made after active consideration, or a question that has never been raised before?
Given Hu Yanbin's massive fan base, this module will likely be iterated upon quickly once it's fully rolled out across all three platforms. Currently, open-source or commercially available video-on-demand solutions are quite mature, and the development cost for integration isn't actually that high.
Since you have to do it sooner or later, and it's not difficult, why not do it now?
For a product aimed at commercial use, when launching version 1.0, it's better to cut some non-core features to ensure the maturity of existing modules. This priority judgment is precisely the part of the Vibe Coding process that is most easily overlooked.
Finally, please allow me to once again thank Professor Hu Yanbin for his outstanding contributions to making Vibe Coding reach a wider audience.
For a musician like Hu Yanbin to be able to deliver and launch a cross-platform internet product solely with the help of AI is already a real success.
If we go back to the moment two years ago when he gave an interview and frankly admitted that he was learning programming, even he himself probably wouldn't have believed that AI's capabilities could advance to this extent in such a short period of time.
But this case also clearly shows us the true limits of Vibe Coding's capabilities today:
AI can write code, generate pages, and build databases, but it can't take a holistic view of the project and ask all the necessary questions for you.
It is essentially a system that answers questions rather than defines them.
It is quite capable in local tasks; however, it is not yet its place to proactively identify and raise problems, and take responsibility for those judgments.
To put it bluntly: AI has not yet mastered the art of asking questions.
This means that in the era of AI Coding, the truly valuable skill is no longer writing good code, but knowing what questions to ask. Code is becoming cheaper, and good questions are becoming more valuable. The value of each role in the software development chain is also converging in this direction.
Finally, I'd like to take this opportunity to ask Professor Hu Yanbin:
Do you need a product manager?
My salary expectations aren't high; I just need concert tickets...
Back row, the back row is fine :)
