What You Should Know Before Starting an AI Business

AI Entrepreneurship: What You Must Know

AI entrepreneurship is all the rage now, and it feels as if not following the trend means being left behind by the times.

These are some thoughts I have come up with from my limited perspective, and with a student's outlook and way of thinking:

First, as an entrepreneur, you essentially face two questions: Can I build a good product? And can I make a sustained profit from it?

So what makes a product a good one?

In my view, good products are driven by needs. Addressing real needs naturally leads to good products.

At a hackathon, I used AI to make a browser extension that imports one's university timetable into a calendar in just three or four hours (it was something I genuinely needed). I posted it on Xiaohongshu and it immediately took off, earning likes, shares, and praise from many fellow students, because they needed it too and my product worked well.

Then what does a bad product look like?

The first kind is an impractical one. Take campus message walls, campus surveys, or lost-and-found platforms: such products exist, but scaling them and earning steady profits is extremely difficult. Even setting aside the many layers of regulatory approval, how would you get customers to use your product in the first place?

I also built a 'my AI' once. In reality it was a DeepSeek wrapper with a system prompt made from highlights of my public-account articles. It was a failure, of course, with no real value. My system prompt was easily broken, and a hacker even used its API to run up 40 yuan in charges in just a few hours.

That was only a small experiment after I had learned a little about using AI, but it also taught me how difficult it is to build a mature product.

To sum up what I mean: you cannot make money by simply wrapping a model. Even if you can, the model will eat up most of your margin.

On the other hand, what does sustained profitability require? A moat.

You cannot live on tips. I have written for my public account for seven years and have received a tip from a complete stranger only once (five yuan). (I have received plenty of compliments, though—thank you all.)

You have to understand that, in China, people do not have much of a habit of paying for knowledge. Yet a good product inevitably needs to charge money, so you need an impregnable moat.

All kinds of AI jailbreaks are prevalent now. If your AI product consists of nothing more than a packaged client interface, it is easy to crack and use for free. Your core assets must be protected on the server side for you to reap compounding returns.

You also need to consider whether your product is distinctive; otherwise, a large company might just make the same thing in passing...

For example, the wildly popular Jev was casually added by OpenAI as a new feature. The Agent identity authentication that we have been laboring over also has a product such as Qoni, with visibility and partnerships. How can ordinary people starting AI businesses compete with that...

(That also shows why, if you have a good idea, you should build it quickly; otherwise someone else will.)

So what kind of product can succeed?

For example, ComindX, built by senior students at our university, has attracted a lot of investment. They have formal technical training and work on technical problems—training models, infrastructure, and the like—and are still in demand.

Then what can ordinary people like us do?

Whatever we do, we cannot outperform the giants. After discussing this with classmates, we all felt rather pessimistic. Aren't the big companies, like those market-manipulating 'whales,' monopolizing innovation?

Still, opportunities exist. At this moment of excitement, hackathons are being held everywhere. In the past you needed technical expertise to take part; now they even encourage you to build with AI. If you are quick-witted and have ideas, earning a little extra money is not bad. We can also work with big companies and join them. My ultimate goal is a partnership, or perhaps an acquisition.

Of course, there are alternative routes too. Building a personal brand is popular now, but most people are too embarrassed to put themselves in front of a camera and keep producing content. Traffic is money; opportunities will come, and the path to monetization is quite clear.

Build a wildly popular little tool (a 'lobster,' a skill), then get recruited by a big company like Peter Steinberger did. 🤣