In September 2026, CoderLegion invited us to a conversation about AI. This is the complete written record of that conversation — seven questions. We also turned it into a two-episode video series.
Part 1 · How I Got Here
This section corresponds to video "Episode 1 · How I Got Here", which covers Q2, Q5 and Q6.
Q1 From medical consultation operations to going independent, I traded goods for services
Q：What was it like transitioning from your earlier career into development and building AI-centered products and companies?
Most of my career was in product operations. Interestingly, my last job as an employee was at an online medical consultation app in China, where I was responsible for the entire consultation business.
Back in 2023, we were already experimenting with AI-driven improvements to the pre-consultation process. The project was called “Medical Assistant Integration.”
GPT had just appeared. The developers were extremely excited about it. I honestly didn’t really know what GPT was at the time.
But I saw our own AI-powered autocomplete feature working in the product.
I remember telling one of my colleagues:
“This is probably the thing that can actually make the consultation process more efficient in the future. You just need to make this work really well.”
Eventually, I left the company for reasons related to the internal organization.
After that, I continued building my own businesses, mainly in e-commerce.
But that period of entrepreneurship convinced me that I wanted to build around services rather than physical goods.
Compared with physical products, services make it easier to get away from inventory and heavy assets, and they allow you to adapt and iterate much faster. With physical goods, you’re still fundamentally dealing with buying and selling, and a lot of the business comes down to the spread between cost and price.
I increasingly see the market as a dark forest.
The best entrepreneur may not be the T. rex.
Maybe it’s the little mouse that keeps adapting and evolving.
Pull quote：I increasingly see the market as a dark forest. The best entrepreneur may not be the T. rex. Maybe it’s the little mouse that keeps adapting and evolving.
Q2 Products that just wrap an API will eventually be eaten alive
Q：You’ve also written about building AI resume tools and choosing platforms. What have you learned from actually building and experimenting with AI products, rather than simply chasing the AI hype?
I think AI products may have a much shorter lifespan than other kinds of software products.
I’ve built an AI resume builder and an AI resume screening tool, and I’ll continue exploring products around enterprise AI transformation.
But I’ve increasingly come to believe that products whose core value is simply wrapping an API will struggle to maintain their differentiation over the long term.
If your product is essentially just packaging the capabilities of a foundation model into a user interface, then as those models become more capable, your own differentiation gets squeezed.
It’s like drinking poisoned wine in the desert. You may have no choice but to drink it, but eventually, it can still kill you.
So if you’re building an AI product yourself, and there’s no better option in the early stage, API-based products can be useful for validating the market and keeping costs low.
But once you have a meaningful user base and enough accumulated data, I believe you must start moving toward local deployment, running models on your own servers, and doing post-training on them.
We don’t need to become GPT.
We don’t need to become GLM.
We just need to become the specialist in a niche.
I don’t think a truly durable AI application should build its core value forever on someone else’s API.
Pull quote：It’s like drinking poisoned wine in the desert. You may have no choice but to drink it, but eventually, it can still kill you.
Q3 I haven’t got the results I wanted yet, so I am in no position to say what success is missing
Q：What do you think is still missing from today’s AI startup ecosystem, particularly for founders trying to turn ideas into real products and businesses?
I think that’s a very broad question.
I’m also a founder who is still trying to turn ideas into products, and I haven’t achieved the results I ultimately want yet. So I don’t think I’m qualified to answer this from the perspective of someone who has already “made it.”
But from the perspective of an independent developer, I continue to see a lot of opportunity in the application layer.
For a small team like ours, hardware is basically out of reach. The supply chain behind it is enormous, and we simply don’t have the capital or resources to participate meaningfully.
The foundation-model layer is obviously a huge opportunity, but models such as GPT, Claude, GLM, and DeepSeek are already becoming extremely capable, while their capabilities and pricing are still changing rapidly.
At this stage, building your entire product around one specific foundation model creates a certain degree of instability.
So I’m particularly interested in what happens after foundation models become more stable: smaller, specialized models trained around specific industries, businesses, and use cases.
We don’t need to compete with the foundation-model companies.
We just need to become the specialist who truly understands a particular domain, its data, and its real-world use cases.
I also don’t think the massive explosion of the application layer has fully arrived yet.
But that may actually be where independent developers have the biggest opportunities.
As infrastructure costs gradually become less important, we should see an explosion of applications.
The areas I’m watching most closely include AI applications across different industries, AI-powered growth for businesses, and data compliance and security for AI-driven businesses.
Part 2 · AI and Where I'm Going
This section corresponds to video "Episode 2 · AI and Where I'm Going", which covers Q7, Q8, Q10 and Q11.
Q4 AI will not replace humans. It will start calling them
Q：Beyond AI, you’re also interested in marketplaces and remote work. As AI-native companies continue to develop, how do you see these areas converging?
Our basic view of the relationship between AI and humans is simple: I don’t believe humans will be replaced as a whole by AI, but I do believe that a huge number of individual jobs will be redefined.
Humans are extremely good at adapting to changing environments. We will continue to adjust the relationship between ourselves and AI as a form of production.
That’s my personal view.
I think we may go through three stages.
The first is AI to B.
AI enters businesses and society’s production systems and starts generating real economic and social value.
The second is AI to AI.
Once a large number of companies have become AI-enabled, AI systems will increasingly call on and provide services to other AI systems in order to push productivity further.
The third stage is what I call Human to AI.
As productivity gains from AI in digital production reach a point of diminishing returns, humans will continue to provide value that AI cannot fully replace.
At that point, AI won’t simply assist humans.
AI will start calling on humans.
We’re already seeing early examples of AI systems using humans for research, real-world information, and physical-world experiences.
But I think the mature version of this third stage will be more than a one-off “human task.”
It will be a long-term working relationship.
AI and humans will each contribute what they are uniquely good at, creating a new kind of production relationship.
Pull quote：AI will start calling on humans.
Q5 The biggest challenge so far is still growth
Q：As both a founder and a developer, what has been your biggest challenge so far, and how have you dealt with it?
So far, I think the biggest challenge has been growth.
And I suspect it will continue to be.
In the long term, AIOMNIU aims to become a global platform connecting AI talent, businesses, and services. But our ability to promote ourselves in global markets has always been limited.
Over time, I’ve also realized that difficulty with growth isn’t always a marketing problem.
At its core, it often comes down to whether the product is compelling enough.
That’s why we’ve continuously experimented with different products, including tools like AI resume screening, to see whether we could attract the users we want to serve.
But our focus is now evolving.
Rather than constantly building another tool just to attract users, we want to get closer to real businesses, understand how their operations actually work—their processes, systems, data, and people—and then identify where AI can create genuine value.
For us, China is currently the market where this is much easier to execute.
I think growth will remain a long-term challenge.
Even when we eventually have a large user base and reach a more mature stage, growth won’t disappear. It will simply become a different kind of pressure.
And if AI eventually reaches the AI-to-AI stage I mentioned earlier, the transition from a platform that primarily serves people to one that increasingly serves AI systems may create another period of growth and transformation.
Q6 We don’t want to just generate an AI report
Q：What are you currently building or exploring through AIOMNIU, and what are you most excited about next?
AIOMNIU is an AI talent and services platform.
We’re currently focused on two core areas: talent outsourcing and AI services.
The outsourcing side can be thought of as a small, AI-focused version of Upwork, connecting businesses with people who can provide AI-related talent and services.
On the services side, we’re increasingly exploring enterprise AI transformation.
What we want to do is not simply ask a company to upload some documents and then generate an AI report.
Real enterprise AI transformation requires understanding how a business actually works: how it makes money, how its operations run, how processes move through the organization, where its data lives, how its systems connect, and how people collaborate.
Only then can we identify the parts of the business where AI can genuinely create value.
That’s why we’re currently more interested in using content and real conversations with businesses to understand what Chinese companies actually need from AI, and then deciding which parts should eventually become products and which should be delivered through FDE and talent services.
Every improvement in foundation-model capabilities makes our products and services better.
That’s the technology flywheel.
But what I’m really excited about is the eventual explosion of the application layer.
That’s when I think we may start seeing AI applications create results at a completely different scale.
I believe that moment will come, and I’m excited to be here for it.
Q7 Don’t become arrogant because of scale
Q：You’ve been active in CoderLegion’s AI, Open Source, and WebDev communities. What has your experience with CoderLegion been like so far, and what would you like to see from the community in the future?
I have a very positive impression of CoderLegion.
They’ve been proactive about encouraging creators, and they’ve also been actively maintaining the community.
When creators feel that a community provides real value, they create more. In turn, the community gains more UGC and becomes more influential.
To me, that is part of the spirit of open source.
A product shouldn’t be treated purely as private property. It can become a shared asset built by everyone who participates in it.
I hope CoderLegion continues to maintain that openness.
No matter how large it becomes, I hope it never becomes arrogant.
A product becomes vulnerable when it starts believing that all of its success came from its own effort, while forgetting the timing, the people who helped build it, and the users who contributed along the way.
Pull quote：No matter how large it becomes, I hope it never becomes arrogant.