Why doesn't OpenAI just run 100 companies? The answers are telling
An Ask HN posed the obvious question: if the models are so good, why not use them to start companies? The top answers: they are doing it, moving into the app layer. Others said selling tokens is the only risk-free money, running one business is easier than running 100, and the gold rush rule says sell shovels. A commenter linked a piece titled 'Why Even Anthropic Can't Vibe-Code Away Salesforce.' Another compared it to Apple not building every app in the App Store.
The pattern: model labs make money from usage, not outcomes, and the last mile of business (sales, licensing, trust, operations) is not code. That is the actual moat for founders who build on top of the models.
The contrarian read is that labs will keep climbing the stack, and anything that is a thin wrapper gets absorbed.
So what?
Your defensible territory is distribution, domain knowledge and operations the labs won't touch. Avoid products that are one model release away from a first-party feature. Vertical specialization is the safer bet.