Why One‑Person Startups Are Harder Than They Look

The "solopreneur" or one-person company is getting a lot of attention. Earlier I discussed how AI has slashed the cost of product development. That reduces the need for outside investment and could push more teams to be their own boss, build something small and beautiful, and bypass venture capital entirely.

Seasoned players in Silicon Valley VC circles point out that venture capital provides more than money. It handles invisible work: outreach and publicity, judgment on where the industry is heading, and legal safeguards and protection. Those are real benefits of bringing capital into a startup. Go it alone, and you have to handle all of that yourself.

So the first requirement: to get a one-person company off the ground, you need a specific set of abilities from the start.

First, you need a clear-eyed view of the market and your product’s direction. This means you can’t just write code. How exactly should the product be designed? Where should it head? Do you have good taste? Can you grasp the texture of a product at the level of a product designer from Apple or Airbnb? That sense of "feel" and the ability to judge details — that’s a hard skill.

Second, when dealing with other matters like legal, finance, compliance, or even media publicity and PR, do you have the relevant capabilities? Sure, AI can help, but in actual execution, you can’t let it do everything for you. You need to personally, as a human, interact with the outside world. For social media promotion, you can ask ChatGPT to polish an article occasionally, but it’s very hard to let it auto-generate and publish posts in bulk; a great deal of human judgment is still required. And what about communicating with lawyers and accountants on platforms like Stripe Atlas or Clerky? Reporting and filing permits with government agencies — that’s not something AI can do on your behalf, right? These issues recur frequently and have very low tolerance for error. If you had a team backed by Andreessen Horowitz or Y Combinator, someone would be taking care of this for you. If it’s your own company, you shoulder it yourself.

Returning to the business of running the company, here’s a very practical issue: the experience AI gains based on its training models and data, or the content it scrapes from the web, might only reach an "average" level. It won’t necessarily give you a perfectly correct answer. Not every sentence it says is right or makes sense. Even if you pick a very powerful model, like GPT-4 or Gemini, it might lift you slightly above average. But if you want to truly stand out in your industry, to be more competitive than others who are also running companies with AI, you must have something unique of your own. So, do you have real experience operating a business? For instance, have you worked at Google or Meta and seen, firsthand, how a product goes from zero to one? At the very least, you know the names of the processes; even if you haven’t done it yourself, you have a clear idea of how things roughly work.

What you learn in school often cannot compare to real-world practice. If you’ve never been exposed to certain workflows on the job, you’ll simply have no idea how they work. You can’t even search for them effectively, because that internal knowledge might not be included in the training data. The ideal scenario is that the product you want to build is something you’ve worked on before in a small or mid-sized team. You’ve seen how that specific process runs, what the pitfalls are, and what you need to prepare. Only then, when you assign tasks to the AI, will you be in control. It won’t be the other way around — you, knowing nothing, starting by asking AI "What should I do?" or "What should I prepare?" If you do that, the most you can make is whatever existed in the AI’s training data, which is highly uncontrollable because you have no idea where its answer came from, whether the entire chain is a closed loop, or whether it’s an end-to-end executable plan. If you can’t even judge the plan and you rush to execute it, the risks are high and it probably won’t work out.

So that’s part two. Operating a product — from the broad architecture down to the nitty-gritty execution, making everything from product to market work smoothly — requires a certain amount of real business knowledge. Only when you have that can you effectively command the AI to help you, rather than being led around by it.

And to get more granular, even down to the code level, AI is currently best at text and data-related things. This goes back to the problem I mentioned before: generating things with AI isn’t the hard part right now. There are too many good tools out there. If ChatGPT doesn’t feel right, you can try Claude or Gemini; the cost is limited, and a few hundred dollars can definitely build something. So what’s the real problem? It’s whether you have the ability to judge if what’s been built is good or bad. Can it actually meet your needs? Forget the idea of "developing a product to serve a certain type of user." When it comes time to make the call — "Okay, this is the solution" — do you have the confidence to do that? What happens more often now is that you ask the AI to generate ten versions. You’re not satisfied with the first, you try a second, then the second doesn’t work, so you switch models and try again. In the end, you may end up with a pile of versions. Which one should you actually use? How do you choose? How do you determine the conditions for "satisfactory"? And even if you are satisfied, how do you know the customers will be? These are all very real problems.

So I’ve outlined three layers: First, can you handle the various aspects of running a business? Second, regarding the business framework and processes, do you have the ability to first figure things out on your own, without relying on AI guidance, and only then enlist its help? Third, regarding what the AI produces, do you have the taste to verify and distinguish the good from the bad? These three layers are the three core difficulties of the solopreneur. Only when you’ve truly internalized all three do you have the real ability to have AI work as your assistant. Ultimately, AI isn’t here to be your boss or to design for you; otherwise, you might as well just let AI run a fully automated, human-free company. But right now, AI doesn’t have that capability. I believe that by placing a human in the loop, the more experience a person has, the more they can use AI to accomplish missions that were previously impossible for one person alone.

This is especially true for someone who, perhaps from working at big companies like Apple or Google, or in a very specialized role, has seen the bottlenecks there — the product needs that, due to large-company syndrome, have remained unmet for a long time. This is the perfect moment to use AI to unblock a demand that was previously blocked. Because after being worn down in the big company, they’ve already thought everything through: what the exact pain point of the current product is, why the big company can’t satisfy it in the short term, exactly how to do it, what the path looks like. They know where the product can be improved and where the incremental value can be captured. Moreover, from a market perspective, the big company might not even want to chase that incremental gain, or their cost structure and efficiency prevent them from doing so in the short term. When you step in to provide that supplementary increment, you become very competitive. In addition, they likely already know certain channels and have mapped out the entire top-to-bottom process, then they use AI to accelerate it. That, to me, is the ideal state for a one-person company.