The other day I was doing some late-night research, and I clicked a broken link.
It led me to a strange PDF from 2012, buried deep in a forgotten Hacker News archive.
I was supposed to be doing client work… but I was deep down a rabbit hole, clicking through internet wayback machine snapshots.
And what I found made me stop breathing for a second.
It was an obscure essay, allegedly written by Paul Graham over a decade ago.
It never went mainstream.
It was never turned into a viral tweet thread by a tech bro with a marble statue profile picture.
But it predicted… with terrifying accuracy… exactly what is happening in the AI market today.
Right now, everyone is panicking.
Startups that raised fifty million dollars six months ago are quietly shutting down.
Founders who thought they were visionaries for wrapping the OpenAI API in a new UI are suddenly realizing they have ZERO competitive advantage.
The market is crashing, and the bubble is bursting.
And this 2012 essay explains exactly why.
Let me break the fourth wall for a moment. You probably clicked on this because you are worried. You have been building with AI, or maybe you are thinking about it, and you see the blood in the water. You want to know if you are wasting your time.
You are.
Unless you understand the fundamental shift that is happening right now.
The core argument of this forgotten essay was simple:
Whenever a massive technological shift occurs, the first wave of companies always builds the wrong thing.
They build “features,” not “platforms.”
And features eventually get swallowed by the underlying infrastructure.
Think about the early days of the App Store. There was a company that made an app that turned your phone screen into a flashlight. They made millions of dollars. They thought they were brilliant entrepreneurs.
Then… Apple released an iOS update that included a flashlight button natively in the control center.
Overnight, the flashlight app company went to ZERO.
Their entire business model was a feature that Apple decided to give away for free.
This is exactly what is happening right now in the AI space.
Look at the graveyard of AI startups over the last year. Companies built entire businesses around summarizing YouTube videos.
Then Google integrated summarization directly into YouTube. Zero.
Companies raised millions to build AI copywriters.
Then Microsoft built Copilot directly into Word and Notion added AI to every blank page. Zero.
The essay from 2012 warned about this specific trap. It called it the “Infrastructure Squeeze.”
When the base layer of technology is improving at an exponential rate, any business built directly on top of it — without a unique proprietary dataset or a deep workflow integration — will inevitably be crushed.
If your entire value proposition is “I send this text to an LLM and show you the response”… you are a flashlight app.
And OpenAI is coming for you.
Let's look at the latest trends. We are seeing models getting smaller, faster, and cheaper. Open-source models like Llama 3 and Mistral are now capable of running locally on a laptop.
What does this mean for your SaaS wrapper?
It means the cost of intelligence is trending towards zero.
If your only differentiator is that you know how to query an API… you are dead in the water.
So, how do you survive this crash? How do you actually build something that does not get vaporized by the next model release?
You have to stop building horizontally and start building vertically.
Horizontal AI tools try to be everything for everyone. “We write emails for marketers, code for developers, and recipes for moms!”
That is a death sentence.
You cannot compete with the foundational models on general intelligence.
Vertical AI tools solve ONE incredibly specific, painful problem… for a highly targeted group of people who have the budget to pay for it.
Do not build an AI that writes generic blog posts. Build an AI that specifically generates compliance reports for dental clinics in California.
Do not build a generic AI code assistant. Build an AI tool that specifically audits legacy Fortran code for banking mainframes and translates it to modern architecture.
The more boring and niche your product sounds, the safer it is from the Infrastructure Squeeze.
The giants like Google and OpenAI are not going to waste their engineering resources building a highly specialized tool for Californian dentists. The market is too small for them.
But it is perfectly sized for YOU to build a hyper-profitable, defensible business.
The founders who survive this crash will not be the ones with the flashiest demos on Twitter.
They will be the ones who dig deep into obscure industries, gather proprietary data that the big models cannot access, and embed themselves so deeply into their customers' workflows that replacing them becomes impossible.
The essay from 2012 ended with a warning that still rings true today.
It said that in every gold rush, the people who get rich are not the ones panning for gold. It is the people selling the pickaxes.
But the catch is… once everyone starts selling pickaxes, the only way to win is to own the mountain.
Stop trying to sell generic pickaxes.
Go find an ugly, unsexy mountain that nobody else wants to climb… and claim it as your own.
The crash is here. It is time to stop playing games and start building real assets.
Are you going to be a flashlight app… or are you going to own the mountain?
