AI Made Starting a Company Nearly Free. That Just Made Starting Worth Nothing.

The numbers dropped this week and everyone is reading them wrong. New U.S. business applications hit 5.6 million in 2025, up 24 percent since ChatGPT launched, according to a Citadel Securities research note that made the rounds again Monday. Gen Z is leading the surge, and the detail buried in the coverage is the one that matters: a growing share of these new businesses do not plan to pay any wages, ever. AI cut the cost of starting a company so far that a 22 year old with a laptop and a $40 a month tool stack can incorporate, build, ship, and sell without hiring a single person. The press is framing this as a golden age of entrepreneurship. I think it is something more uncomfortable and more interesting: the moment starting a company stopped being an achievement at all.

When the barrier falls, the barrier was the moat

For most of my career, the sheer difficulty of getting a working product into the market was a filter. You needed capital, or you needed to be technical, or you needed to convince technical people to join you. That difficulty was ugly and exclusionary and I am glad it is dying. But it was also, quietly, the first moat every startup had. If it took a team of five a year to build what you built, you had a year and five salaries of head start on anyone who wanted to copy you. That head start is now measured in weekends. When 5.6 million businesses can spin up a year and a meaningful chunk of them can produce competent software with no employees, existence is no longer evidence of anything. The fact that your product works is now table stakes so low they are practically underground.

This is the same logic I keep applying to models, playing out one layer up. Models commoditized, so the value moved to workflow, data, distribution, and trust. Now software production itself is commoditizing, and the value is moving to exactly the same places. It was never really about who could build. It was always about who could get used, and who could get believed.

The zero employee company is a real thing, not a stunt

I want to be clear that I am not sneering at the solo wave. The zero employee company is legitimate now in a way it was not even three years ago. One founder with good taste, an open model running on rented GPUs, and an agent stack for support, bookkeeping, and outbound can run what would have been a 12 person company in 2021. Bank of America's numbers show planned hiring per new business collapsing, and I do not read that as weakness. I read it as founders correctly noticing that the first ten hires at most startups existed to compensate for expensive software production, and that compensation is no longer needed.

What I am saying is that if you are one of these founders, you need to be honest about what you actually own. You did not build a moat by shipping. You bought commodity production at market price, same as everyone else. The 5.6 million applications are 5.6 million people with access to the same models, the same agent frameworks, the same templates. Your codebase is not an asset. It is a receipt.

Where the leverage actually sits now

So where does a solo, AI leveraged founder actually accumulate advantage? Three places, and none of them can be generated by a model. Distribution: an audience, a channel, a wedge into a buyer's existing workflow. The founder who spent two years in an industry and knows the 40 people who feel a specific pain has something no weekend cloner can reproduce. Data: not scraped data, earned data. Usage patterns, outcomes, corrections, the exhaust of real customers doing real work through your product. Trust: the slowest asset in the economy and now the scarcest, because when anyone can ship software, buyers stop evaluating software and start evaluating people. Reputation compounds precisely because it cannot be prompted into existence.

There is a stack lesson hiding in here too. A zero employee company has no migration team, no platform group, nobody to unwind a bad dependency later. Every vendor choice a solo founder makes is close to permanent, which means lock in is more dangerous for the smallest companies, not less. If your entire business runs through one closed model's API and that vendor changes pricing or policy, you personally are the incident response team. This is why I keep telling small founders, even more than big ones, to build model agnostic and keep an open weight fallback they can serve themselves. Cheap to start must not decay into expensive to stay.

The uncomfortable math of 5.6 million

Here is the part nobody puts in the celebration posts. If starting is nearly free, most of these companies exist because the cost of trying collapsed, not because 5.6 million viable opportunities appeared. The failure rate is going to be spectacular, and that is fine, that is what cheap experiments are for. But it changes the game for anyone raising money or selling into this wave. Investors are already discounting shipped product to zero and underwriting distribution and founder specific knowledge instead. Customers are drowning in functional tools from unknown sellers and defaulting to whoever they already trust. The bar did not disappear. It moved to the one place AI cannot yet clear it for you.

So here is the test I would run on your own company this week, whether you are solo or fifty people. Delete your product from the equation and ask what is left. A list of customers who would follow you to a new product? Data no competitor can replicate? A reputation in a niche where buyers know your name? If the honest answer is nothing, then your company is one of the 5.6 million, indistinguishable from the crowd the moment someone with better distribution notices your market. Building has never been easier. Mattering has never been harder. Spend accordingly.