I recently came across a striking take.
Boris, the father of CC, recently said: programming has been "pretty much solved."
It sounds absolute. But if you've been using LLMs to write code these past two years, you know it's true — not fully solved, but we've crossed the threshold where you no longer "have to write it yourself."
Which raises the question: If writing code is no longer scarce, what is?
The knee-jerk reaction: is the software industry about to be flattened? Is SaaS doomed?
But look closer, and you'll find the opposite in places. Some guardrails AI still can't touch.
AI is rapidly dismantling moats we once took for granted.
Take switching costs.
You used to get locked into a system: data won't migrate, APIs don't match, your team doesn't know the new tool. Now, an agent can migrate your data, write adapters, even "learn" the new system for you. Switching platforms went from an engineering project to a task.
Or take process barriers.
Many companies' edge wasn't in the product — it was in the process: a complex, internal-only way of doing things that outsiders couldn't replicate.
Today, you throw a goal at a model, let it iterate, and it can decompose processes, optimize them, even execute them. "We know how to do this" — far less valuable now.
So here's the surface picture: Barriers are falling. Capabilities are diffusing. Small teams can do more than ever.
But here's the line most people missed — Boris's real punch:
Network effects, economies of scale, scarce resources — AI hasn't changed any of these moats.
This is the crux.
Because it's saying something uncomfortable but deeply true:
AI changed the cost of doing things, but not the nature of competition.
You can use AI to build a product fast, but you can't use AI to conjure a user network out of thin air.
You can use AI to rewrite a system, but you can't use AI to build a global supply chain.
You can use AI to boost efficiency, but you can't use AI to create exclusive data, channels, and brand.
A clearer structure starts to emerge:
The ability to write code — depreciating. The ability to ship products — depreciating. Even "getting things built" itself — depreciating.
But at the same time,
The ability to aggregate users — unchanged. Cost advantages from scale — unchanged. Control over critical resources — more important than ever.
In this sense, AI hasn't flattened the world. It's just re-sorted it.
Many people think this is an era where "anyone can build a product." But the more accurate version is:
This is an era where anyone can build a product, but not everyone can build a business.
From this angle, a harsher, more realistic trend emerges:
AI will make bad companies die faster, but it won't automatically create great ones.
Because "writing code" is no longer scarce. "Ideas" are no longer scarce. Even "products" are no longer scarce.
What's truly scarce are other things:
People. Data. Distribution. Scale. And the ability to organize all of them together.
If the last decade's core question was "can you build it," the next decade's question becomes:
Why should you own the users? Why should you own the data? Why should you own the distribution?
Code is becoming infrastructure. And business is becoming business again.
