The recent viral article claiming that most one-person companies (OPCs) are doomed to fail missed the point. Yes, 80% of solo ventures fizzle out — but that mirrors the failure rate of any startup. A restaurant’s first-year closure rate hits 60%. Independent games? 95% never break even. The real mistake is framing this as a unique crisis rather than the baseline cost of starting anything.
What deserves attention isn’t the failure rate itself, but two deeper shifts: how AI has lowered the cost of failure to near zero, and what separates those who eventually succeed from those who burn out.
The first shift is a structural change in risk economics.
Nassim Taleb’s concept of convexity captures it perfectly. When losses are capped but gains are open-ended, you want to run as many experiments as possible. Venture capitalists do this with portfolios — ten bets, nine fail, one covers the fund. Before AI, an individual couldn’t replicate that. A single failed restaurant could wipe out years of savings. Now, a failed AI-powered side project might cost a week of time and a few hundred dollars in API fees.
This is exactly how large language models improve: reinforcement learning from human feedback (RLHF). The model acts, receives feedback, updates weights, and acts again. The same loop now applies to solo entrepreneurs. Marc Lou ships twenty products knowing nineteen will fail; he hit $1M ARR in twelve months. His advantage isn’t a masterpiece — it’s the ability to ship one hundred times while others ship once.
AI compresses the feedback cycle from years to days.
Let’s make this concrete. Pieter Levels runs Nomad List and several other products solo, pulling in about $138K monthly from users across seventy countries. Before AI, achieving multilingual, multi-region reach required a translation team and years of localization. Now AI handles that pipeline, letting him focus on product judgment. Dan Koe earned $4.2M in 2024 with a single weekly newsletter repurposed across seven platforms — a task that used to require a content team. Tony Dinh built TypingMind to over 20,000 paid users working four hours a day, because AI removed the coding bottleneck.
But here’s the critical nuance: AI is a lever, not an engine.
It amplifies whatever you already bring to the table — taste, market insight, problem selection. If you have no sense of what people need, AI won’t invent it. The solo founders who succeed are those who use AI to multiply their judgment, not to replace it. They treat AI as a cheap trial-and-error platform, running dozens of micro-experiments, each costing nothing but attention.
The original article’s selection bias is dangerous precisely because it obscures this new convexity. By cherry-picking three failed stories and declaring the whole model toxic, it discourages people from engaging with the one tool that makes low-cost experimentation possible. Yes, most solo ventures will fail — but the ones that succeed now do so at a scale and speed impossible five years ago.
The question isn’t “should you start an OPC?” It’s “can you design a low-cost learning loop that lets you fail cheaply and often?” If the answer is yes, the odds are no longer stacked against you. They’ve never been better.