Five Hard Lessons from Spending RMB 200,000 on a One-Person AI Startup

AI lowers the barrier to building software. It does not lower the barrier to building a business.

That is not the most inspiring way to open an essay about AI entrepreneurship. But it is the conclusion I paid for with two months of work and at least RMB 200,000.

I’m Tymon. Since early 2023, I have worked on generative AI products at Z.ai and Meituan’s travel business. In May, I left to build a company of my own. We are a two-person team, just over two months in, and our combined costs have already crossed RMB 200,000.

Before I started, I believed that sufficiently capable AI would make it realistic for one person to build a product and close the business loop. I no longer see it that way. AI can help one person write code, produce a demo, draft copy, and automate workflows. It cannot automatically find customers who will pay. Nor can it tell you whether a need is genuinely painful or merely elegant in your own head.

If you are considering a full-time one-person company—often called an OPC in China—or a team of just one or two people, these are the lessons I would think through first.

01 / AI Accelerates Production. Distribution Is Still the Hard Part.

I knew customer acquisition would be difficult. I was still too optimistic about it.

My reasoning was straightforward: AI can write code and copy, generate images, and edit video. I also had more than three years of experience building AI applications. Surely I could use AI to run content on Xiaohongshu (RED), X, and other platforms, then turn that content into leads.

The logic was sound in theory. Reality exposed the gaps almost immediately.

First, I did not yet understand the fundamentals of audience building. AI can generate a post. It cannot decide why someone will stop scrolling, trust you, send a message, or pay for your service.

Second, AI-generated content is often correct but useless. It can be coherent and polished while containing no lived situation, no personal judgment, no market friction, and none of the texture that tells a reader, “This person has actually done the work.”

Third, content-led acquisition compounds slowly. We still have not fully cracked the system: how to create breakout pieces, publish consistently, and turn attention into qualified conversations.

When consumer content failed to produce clear results, we considered taking on B2B projects to generate cash flow. That market was no easier. Large companies are hard to reach without existing access. Smaller businesses may believe AI has potential and happily take a meeting, but many are reluctant to pay. They would rather “co-create” for free while you fund the experiment with your own time and runway.

That raises an uncomfortable question: If a customer believes an idea has potential but will not risk any of their own money, is the project really an opportunity?

Most of the time, our answer is no.

An idea offered at zero cost is often not a requirement. It is simply a thought. For an early-stage team, code is not the scarcest resource. Time, attention, and runway are. You cannot spend them validating someone else’s vague curiosity when that person has no willingness to pay.

We eventually adopted a filter: when we do not already have deep domain knowledge, any joint project must be paid. The initial fee can be negotiated. The amount matters less than the commitment. Payment is evidence that the problem hurts enough to act on.

02 / The Hardest Part Is the Silence

AI entrepreneurship looks exciting from the outside. One person can build the product, publish the content, and automate operations—one founder performing like an entire team.

What wears you down is not simply the workload. It is working hard while receiving no signal that any of it matters.

Over these two months, we explored two directions: multi-agent collaboration and AI-generated animated shorts. Neither worked for us. Neither turned out to be our opportunity.

Intellectually, I knew what the problem was. We did not need to build more products. We needed to establish a distribution loop, talk to the market, and collect real feedback. But my body did not keep pace with my mind. Every morning I knew what action to take, yet I felt as if I were underwater. Everything looked normal from the outside. Internally, I was deeply anxious.

That period taught me something important: entrepreneurship is not a continuous sequence of correct decisions. It is the ability to keep moving while the market remains silent.

Solo founders should confront that scenario in advance. What will you do after several weeks without revenue, customers, or positive feedback? When you suspect the direction is wrong but have no one to test your judgment against, how will you keep doubt from consuming all your energy?

AI can help you reason through a decision. It cannot carry the emotional weight of making it.

At a critical moment, one sentence from my co-founder while we were walking down the street mattered more than a thousand AI conversations:

“We’re okay, brother. We’ll keep trying together.”

That is why I no longer recommend that everyone quit immediately and build alone. A solo start can work at the very beginning. But if you are serious about staying in the arena, a partner with aligned judgment and complementary strengths can be invaluable.

03 / A Demo Is Not a Product

There is no shortage of content claiming that “building” is becoming trivial—that AI lets one person operate like a team, perhaps even create a unicorn alone. I do not deny the possibility. I now ask a second question: What is the probability of reaching that outcome?

I have worked on AI applications since 2023 across chat, workflow, and agent-based product models. I have also used AI to build MVPs for several commercial products. The simplest was a multi-agent analytics product with roughly 60,000 lines of code.

I am not skeptical of AI coding. I am one of its believers.

What these two months clarified is that a demo that runs and a commercial product that delivers reliably in the real world are at least an order of magnitude apart in complexity.

A demo asks, “Can this appear to work?” A product must answer a much longer list of questions:

• How do users register, and where is their data stored?
• How are permissions handled, and how does payment work?
• What happens when the system fails or produces inconsistent results?
• Who is accountable to the customer?
• What is the recovery plan when delivery goes wrong?
• Who owns ongoing support?

You can postpone all of that while building a demo. The moment you charge money, you cannot.

For a non-technical founder, AI coding is an excellent way to produce a demo and test the problem, the workflow, and whether users care about the core capability. Once the direction enters real commercialization, however, a strong technical partner should own quality. Otherwise, you can end up in a dangerous position: believing you have built a product, only to watch delivery fall apart.

04 / Validate Demand Before You Start Coding

AI coding is seductive because the feedback is immediate. Describe a requirement and code appears. Move a button and the change is implemented. Ask for another feature and it is wired in. We went through our own phase of coding continuously because the progress felt so satisfying.

But a startup is not a coding competition. Everything eventually returns to one question: Have you solved a problem that someone is willing to pay for?

If you keep your head down and code, it is easy to mistake motion for progress. You may think you are advancing the company when you are really avoiding market feedback.

I now put demand validation ahead of development. Before we build, I ask four questions:

1. Is the customer already spending money on this problem?
2. Can we reliably reach this group of customers?
3. Do we understand the domain better than an ordinary AI user?
4. Will someone pay for the outcome while the product is still imperfect?

If all four answers are vague, do not rush into development.

Validation does not always require an MVP. A front-end concept page may be enough. So might a promotional video, a conversation at a Demo Day, or a direct pitch in the environment where the problem occurs.

If no one asks a follow-up question, sends a message, requests a consultation, or asks about price, be careful. The idea may look sophisticated to you without being painful to anyone else. Of course, the positioning may also be wrong.

Technical and product-minded founders often explain how impressive the technology is. Customers ask different questions: Will this solve my problem? Will it be cheaper, faster, or more reliable? That is the language of customer value.

05 / Early-Stage Strategy Is Subtraction

Powerful AI can make founders overconfident. Suddenly everything seems buildable: AI entertainment, AI lead generation, AI fitness, AI analytics. Each market appears full of opportunity.

The early-stage trap is confusing “opportunities are everywhere” with “all of these opportunities belong to me.”

I now judge founder–market fit through four questions:

1. Do we have deep or distinctive insight into the domain?
2. Do we have a credible channel to reach its users?
3. Have we accumulated relevant data, cases, relationships, or other assets?
4. Does the direction use the team’s strongest capabilities?

If the answer is no across the board, proceed carefully. It does not mean the idea is impossible. It means you need to know whether the missing insight can be acquired through fast, disciplined experiments.

One of our mistakes was starting with too many ideas. Every concept seemed valuable, and every direction was difficult to abandon. But a one- or two-person team has very limited attention. Testing several markets at once can look like diversification while actually reducing the odds of success in each one. You do not spend enough time inside the workflow, meet enough customers, or stay with the feedback long enough to understand it.

We eventually set a rule: validate one direction at a time. If it fails, move to the next.

The goal of an early-stage startup is not to preserve the maximum number of possibilities. It is to disprove weak ideas quickly enough to find one worth pursuing.

Which Opportunities Fit a One- or Two-Person Team?

Today, I would lean toward content, tools, consulting, and services. They may not sound as glamorous as a unicorn story or an IPO, but they are more compatible with the resources of a tiny team.

I would prioritize opportunities with four characteristics:

1. A short buying path. Can the customer decide independently, or does the purchase require several departments, budget cycles, and security reviews? Long enterprise approval chains are heavy for a new team.

2. Lightweight delivery. Are you delivering a tool, a solution, or a defined result—or committing to a long, complex project? Heavier delivery demands more process, stability, and after-sales support.

3. Existing proof that the market pays. You do not need to invent something no one on Earth has attempted. At this stage, evidence that others already make money in the category is useful. Your job is to find a meaningfully better way to serve it.

4. Proximity to money. Does the problem connect directly to revenue, cost, productivity, acquisition, or conversion? The closer the value is to money, the easier it is for a customer to evaluate and pay for. The farther away it is, the more likely the response becomes, “Interesting—but not now.”

The Takeaway

My conclusion is not that AI has failed. Quite the opposite. I still believe it will transform individual productivity and create opportunities for one- and two-person teams that did not exist before.

What I no longer believe is that a business loop will materialize naturally once the product exists.

AI can help you make things faster. It cannot replace basic commercial judgment:

• Who feels the pain?
• Who has budget?
• Whom can you actually reach?
• Who is willing to pay now?
• Why are you better positioned than anyone else to solve this problem?

Without answers, faster coding may simply mean faster experiments, faster spending, and faster anxiety.

If you want to build an OPC, I am not trying to talk you out of it. I am suggesting that you lower the temperature before you jump in. Do not begin with, “Can AI help me build this product?” Begin with: Will anyone pay for this need? Do I have an insight advantage? Can I reach the customer? Is the opportunity close enough to money?

Answer those questions first. Then build.

AI has made one-person entrepreneurship more possible. It has not made entrepreneurship lighter.