One Person Instead of a Company. Did AI Make Business Easier — or Just Let Us Take the Risk Alone?

Until very recently, a good idea had one rather inconvenient characteristic: someone had to build it.

Suppose you’re a marketer. You understand the market, you know where to find your first customers, and one day you notice a problem people seem willing to pay to solve. Great. Now you need a developer. Then a designer. Then you discover you need another developer because the first one is working on the backend, while the app, inconveniently, also needs a frontend. Someone has to test the whole thing, connect payments, write the copy, set up analytics, and figure out why half the users can no longer log in after the latest update.

And suddenly you’re no longer a person with an idea. You’re an employer. Or someone who urgently needs money in order to become one.

A large part of startup culture over the past two decades grew out of this basic reality: co-founders, angel investors, seed rounds, accelerators, venture funds. Sometimes, just to find out whether your business could ever make a million dollars, you first had to spend several hundred thousand.

AI has started to break that sequence.

In 2026, a designer can write software, a marketer can assemble an app, and someone who doesn’t know a programming language can release a commercial game on Steam. An experienced engineer with several AI agents can, in some tasks, already perform the work of a small team. You can now start a startup without first building a company around it.

It sounds like the beginning of the solopreneur era.

There is just one problem: being able to build a product by yourself and being able to build a good business by yourself have never been the same thing.

So I wanted to understand what AI has actually changed. Has it made entrepreneurial success more likely? Or has it simply made a failed attempt cheaper?

The Wall Really Did Come Down

Let’s start with the good news: the technological part of this story is real.

One of the clearest recent examples is Payout, an app built for a hackathon. The first version was created in ten days. According to the project’s own Devpost page, the app attracted more than 17,000 users, 1,750 paying subscribers, and generated just over $30,000 in revenue. This is no longer a prototype a founder proudly showed to their mother and three friends on Telegram. People took out their credit cards.

Even more interesting is the Chinese app 小猫补光灯, essentially a small on-screen fill light for video calls and selfies. Its creator was not a professional developer. He built the first version with Cursor in roughly an hour, solving a very specific and rather ordinary problem. The app later reached hundreds of thousands of downloads, and according to figures published by the developer, the paid version generated hundreds of thousands of yuan.

One hour.

Of course, everything between “a first version in an hour” and “a working business” still exists: iteration, distribution, support, payments. But only a few years ago, even the first step usually required someone who knew how to code. Now, sometimes, it requires someone who can explain clearly enough to a machine what they want.

In corporate software development, the effect looks different. In a 2026 case study, one staff engineer working with four AI agents completed an initiative originally planned for four developers in roughly half the scheduled time. The authors estimated a reduction of more than 85% in direct staffing cost.

There is, however, an important detail that tends to disappear on the way to a headline like “AI replaced four programmers”: the person at the center of the experiment was an experienced staff engineer. The authors explicitly highlighted the importance of good specifications and deep knowledge of the system.

The machine turned out to be an excellent multiplier. But it was not multiplying zero.

When Everyone Can Open a Factory

When producing software suddenly becomes much cheaper, something entirely predictable happens: people produce more software.

And this is where the story of democratized entrepreneurship starts to change its expression.

According to RevenueCat, the number of new subscription apps in its dataset grew from roughly 2,000 per month in early 2022 to more than 14,700 per month by early 2026 — more than sevenfold. You might expect that this would trigger a great migration of revenue toward new developers.

Not quite.

In RevenueCat’s data, 69% of all subscription revenue still comes from apps launched before 2020. Apps launched in 2025 and later account for roughly 3%. Among new subscription apps, around 17.3% reach $1,000 in monthly revenue within their first two years. Around 4.6% make it to $10,000.

These are not solopreneur success rates, and they are not startup survival probabilities. RevenueCat’s dataset measures something else. But it does reveal another problem very clearly: software is being created much faster than new human eyeballs, wallets, and hours in the day are appearing.

Games tell a similar story. According to Chris Zukowski of How To Market A Game, 20,282 games were released on Steam in 2025. Only 608 of them reached 1,000 or more reviews — around 3%. A thousand reviews is not an official threshold for commercial success, but on Steam it is already a very serious level of traction, often associated with revenue in at least the hundreds of thousands of dollars.

Again, this does not mean “97% of developers failed.” The denominator includes free games, experiments, hobby projects, shovelware, and products with entirely different goals. But the direction is clear enough.

AI made production cheaper. Human attention did not get cheaper.

If anything, the opposite may be true.

And Then One Guy Makes a Hundred Million Dollars

This is exactly where careful discussions about medians begin to sit irritatingly badly beside reality. Because solo developers really do win.

Sometimes, obscenely so.

Schedule I, built largely by Tyler on his own, became one of the biggest indie hits of recent years. Estimates of sales and gross revenue have long since moved into the tens of millions of dollars, and the game itself became a Steam phenomenon.

Megabonk sold a million copies in roughly two weeks. Its story is especially useful. At first, it looked like the kind of fairy tale the internet loves: an unknown developer releases his first game and wakes up famous. Then the creator voluntarily withdrew from a Best Debut Indie Game nomination at The Game Awards, explaining that it was not actually his first game — he had released projects under other names before.

A very modern version of overnight success.

The night lasted several years.

Tangy TD followed another path. Solo developer Cakez worked on the game for four years, streaming development along the way and gradually building a community around it. In the first week after release, Steam showed roughly $245,000 in gross revenue. A month later, the figure was already around $788,000.

Then there is Paddle Paddle Paddle: several months of development, more than 270,000 copies sold, and roughly $826,000 gross according to the creator. But this is where useful accounting enters the picture. After refunds, taxes, platform fees, the publisher’s share, and other deductions, the amount the developer described as actually ending up with him was closer to $250,000.

Still an excellent result. Just not $826,000.

That is why solopreneur stories tend to look particularly good in headlines and slightly worse in Excel.

How Much Does a Business Cost When It Has No Payroll?

There is a much less famous example I like precisely because it is more ordinary.

In August 2026, a developer released HEXSTORM: Tears of Arcadia on Steam. He had spent many years working in the game industry on the visual side of projects, but he could not program. According to him, the code for the new game was written with AI.

The project took roughly a year and around 1,000 documented hours of work. In its first week, the game generated a little over $11,000 gross. After refunds, taxes, and Steam’s 30% cut, the developer expected to receive about $6,000. Sales then stabilized at roughly 20–30 copies per day. The creator himself estimated that, if the pace held, the first year might generate around $25,000.

For a debut game, that is a good result. Technologically, it is remarkable: someone who cannot program created a commercial software product, released it on three operating systems, found customers, and started making real money.

The experiment worked.

But now change the question. Not “Can one person build a game with AI?” Clearly, yes.

The better question is: “Was 1,000 hours of his time a good entrepreneurial investment?”

And suddenly the answer is much less obvious. If the game really earns $25,000 in its first year, that is $25 in revenue for each of the first thousand hours invested — before future support, updates, and the opportunity cost of that time.

Meanwhile, nearby, there is a Chinese app whose first version took roughly an hour and later generated tens of thousands of dollars. There is Tangy TD: four years of work and hundreds of thousands of dollars in the first month. There is Paddle Paddle Paddle: several months and serious six-figure money. There is Schedule I: years of work and an outcome on an entirely different scale.

All of them are described with the same phrase:

solo success.

Economically, they are completely different animals.

The Metric Missing From Solopreneur Stories

Startup media love MRR. ARR is even better. Copies sold, valuation, acquisition price — perfect.

For solopreneurs, I would add one more metric:

Return on Founder Time.

Because the absence of salaries does not mean the absence of production costs. If a founder spends 2,000 hours of their life on a product, those 2,000 hours do not disappear from the books simply because no one issued an invoice for them. The cost has simply gone invisible.

And this makes it easier to see what AI has actually changed. Every entrepreneurial attempt has at least four prices: the money you have to spend before meeting your first customer; the skills you once had to buy through employees and contractors; the founder’s own time; and finally the market risk — the chance that, after all of this, nobody actually wants the product.

AI has already collapsed the first two.

The third is much more interesting. The fourth, even more so.

AI may let one person perform the work of five specialists. But the work of five specialists does not necessarily disappear. Sometimes, one very tired person simply ends up doing all of it.

The creator of HEXSTORM describes the launch week well: more than ten patches in the first three days, balancing against live users, dealing with refunds and negative reviews, sleeping three hours a night.

There is no team. But the team’s work is still there.

Maybe We Are Asking the Wrong Question

Most conversations about AI entrepreneurship eventually reduce to some version of this: Will AI increase the probability that a startup succeeds?

So far, I have not found convincing data that lets us confidently answer yes. If anything, the explosion in supply suggests that competing for attention may be getting harder.

But suppose, for the sake of argument, that the probability of success for each individual product does not change at all. AI could still radically transform entrepreneurship.

Imagine a very hypothetical founder ten years ago. They have $100,000. Each serious MVP costs $50,000. They have two attempts.

Today, that same person may be able to build ten prototypes for a few thousand dollars, put them in front of real users, and only invest months of work in the one where someone finally reaches for a credit card.

AI does not have to improve the odds of a good bet. It lets you make more bets.

And this is where it becomes especially interesting to look not at people who spent four years building one masterpiece, but at serial makers. Marc Lou ships products one after another. Pieter Levels has worked this way for years. Jordan Morris, the creator of Rusty’s Retirement, has described how he builds several small prototypes, shows them to people, and continues with the one that gets the strongest response.

Perhaps the true AI-native solopreneur does not look like someone capable of spending three years building a huge company alone. Perhaps the real advantage is something else.

They can be wrong very cheaply.

Twenty hours. Doesn’t work. Kill it.

Another forty. Nobody cares. Kill it.

A hundred hours. The first ten paying customers appear.

Now you can work.

If that is true, AI really can increase an entrepreneur’s chances of success — just not in the way it is usually sold to us. Not by making every idea better, but by allowing a founder to survive more bad ideas.

So Who Got the Biggest Advantage?

Only now does it make sense to return to the question I originally started this research with: who benefits most — programmers, marketers, product managers, or people with no technical background at all?

After looking at dozens of cases, I like the question less and less. Professional experience did not disappear. Something else changed: some missing capabilities can now be rented from machines relatively cheaply.

For programmers, this is an enormous gift. They could already build the product, but they still needed design, copy, research, testing, localization, support, and sometimes marketing. Now part of a small company fits inside their AI subscriptions. There is experimental evidence for this too: in one Google study, professional developers using AI completed a complex task roughly 21% faster, while the case of one staff engineer working with several agents shows how far that model can go under the right conditions.

But the research also produces the opposite result. In a METR experiment, experienced open-source developers working on repositories they already knew completed tasks about 19% more slowly with AI.

The best part of the study is that the developers still believed AI had made them faster.

I think half of modern office life fits inside that one number.

Later METR data already suggests that newer tools may be starting to speed up even this class of developer. The technology is moving too quickly to carve one number into stone. But the broader pattern is interesting: AI appears very good at multiplying existing competence. There is much less evidence that it reliably creates competence from nothing.

That is why product managers look like particularly strong candidates for the new solopreneur archetype. They are used to turning a vague “I want something like this” into requirements, breaking work into parts, prioritizing, and reviewing the result. In a sense, a good PM already knows how to manage a team. It is just that now part of the team does not ask for a salary and never argues in the daily stand-up.

Marketers receive a different advantage. Their problem used to sound like this: “I know exactly who to sell this to. But I can’t build it.” AI is beginning to remove the second half of that sentence. And as software production gets cheaper, the ability to find customers becomes relatively more valuable.

Domain experts may be even more interesting. An accountant can spend twenty years seeing the same stupid problem that ten thousand accountants would happily pay $20 a month to solve. A programmer may never even notice the problem. The accountant does — but historically, there was a development team standing between the insight and the product. That wall is now much lower.

And creators already own the thing everyone else desperately needs: attention. When building an app costs less and less, a few million followers may be more valuable as startup capital than the ability to write your own backend.

So the winner, at least for now, does not look like a particular profession. It looks more like someone who already has at least one scarce advantage — technical depth, an audience, market knowledge, or strong product taste — while AI temporarily fills in the rest without requiring five hires.

There Is Another Uncomfortable Effect

Before modern AI tools, someone without technical experience faced a useful obstacle: they could not build the product.

It sounds strange to call that an advantage, but the wall forced them to go talk to other people. Find a technical co-founder. Speak to a developer. Show the concept to a publisher. Apply to an accelerator. Ask someone for money.

And all those people started asking deeply annoying questions. Who will pay for this? Why? How much? Who are the competitors? Why is the user not solving this problem already? Why you? Why does this need eighteen months of development?

Sometimes the founder walked away from those conversations offended.

Sometimes they walked away saved.

AI removes the need to ask permission. “I want to build an app.” — Great idea. “A game?” — Absolutely. “A blockchain marketplace for owners of aquarium shrimp?” Well, if you write the prompt well enough, the machine is unlikely to ruin your mood there either.

The ability to execute an idea alone means a founder can now spend a very long time avoiding the market. And that means AI can simultaneously reduce financial risk while increasing founder-time risk.

A founder used to be able to burn $200,000 of an investor’s money. Now they can much more efficiently burn 2,000 hours of their own life.

Maybe a Team Was Not Just a Cost

The cult of the one-person company tends to treat employees as a problem finally removed from the system: salaries, taxes, management, meetings. Someone gets sick. Someone quits. Someone inexplicably has their own opinion. Compared with this, an AI agent looks like an almost perfect colleague.

But teams perform another function. A designer tells the developer the interface makes no sense. A marketer tells the founder that their favorite feature is irrelevant. Sales brings back one sentence from a customer that forces half the product to be rebuilt. An investor takes equity, but every now and then asks why the sales graph is lying on the floor.

People create friction.

And sometimes friction is useful.

In the HEXSTORM story, AI even helped the developer choose the price. The model recommended around $14, reasoning that a game priced too cheaply might be perceived as lower quality. Real customers turned out to be less philosophical. After complaints, the price had to be reduced.

AI helped build the game. The market still set the price of the business.

China Has Already Found the Next Problem

The new paradox is especially visible in a 2026 Chinese study of one-person companies.

The researchers collected roughly 1,500 questionnaires and conducted more than a hundred hours of interviews and case analysis. About 75% of participants came from non-technical professional backgrounds. So the technological democratization is very real.

But more than half named customer acquisition as one of their biggest problems. The researchers described the situation with an almost perfect phrase:

“Excess capacity to build, insufficient capacity to acquire customers.”

That seems to be exactly where we have arrived.

For twenty years, the startup industry worked to make product creation cheaper: open source, cloud, no-code, app stores, Stripe, AWS, GitHub. Now AI has attacked the final major cost center — human intellectual labor.

We have finally learned how to produce almost as much software as we want.

And discovered that nobody promised us the same number of buyers.

So Has the Solopreneur Era Arrived?

I think it has.

It just looks slightly different from the YouTube thumbnails.

The most interesting thing AI may enable is not one person building a company alone. It is one person testing a business before they need a company at all.

That is a fundamentally different model. The old path often looked like this: idea → team → money → product → market. Increasingly, the new one can look like this: idea → AI → prototype → market. And only then: product → team? → investment? → scale?

The question marks are the most important part.

If the market stays silent — kill it. If users come but do not pay — change it or kill it. If the first twenty people pay — keep going. If a thousand pay — maybe it is time to hire.

Or maybe not.

This is where AI may eventually change the probability of entrepreneurial success in a meaningful way. Not because the machine has learned to predict which startup will become the next billion-dollar company. If it had, venture capital funds would already consist of a few servers and one very happy accountant.

But because the cost of being wrong is falling.

A founder can test five ideas with the capital that once funded one. Ten. In some cases, fifty. And save real money, years, and a team for the one idea that finally shows signs of life.

Perhaps this is where the line runs between two very similar solopreneurs of the future.

One will say:

“Now I can finally build everything myself.”

The other:

“Now I no longer have to build everything to find out whether it is worth building.”

I think the second one understands AI a little better.

Because AI has not made successful businesses easy. It has made the attempt to build one dramatically cheaper.

And perhaps the real solopreneur revolution is not that one person can now effortlessly create a successful company. It is that, for the first time, one person can afford to find out relatively cheaply whether there is a company worth creating at all.