I keep seeing a line from Rob Saker float across my feed: "Business logic is migrating from code written by specialized engineers to markdown files that anyone with domain expertise can write." I want to sit with that sentence for a second, because it is doing more work than people realize.
The two kinds of knowledge
Every software product sits on top of two kinds of knowledge. One: how the domain works. What a radiologist actually looks for. How a freight forwarder actually prices a container. What a tax auditor actually flags. Two: how to turn that into code that runs.
The founder historically sold the bridge between the two. They found someone with domain knowledge, added someone with coding skill, or were themselves a rare hybrid. The product was the bridge. The moat was "I can do both."
Here is the thing. Domain knowledge lives in the customer. It always has. The radiologist is the customer. The freight forwarder is the customer. The tax auditor is the customer. The founder was never the repository of that knowledge. They were a translator, hired by circumstance because the customer couldn't code.
The moat was the coding skill
Not the software. The coding skill. Software was the output. The moat was the thing that generated the output, and the moat had nothing to do with understanding the customer's problem. The moat was knowing Python, or React, or how to wire a CI pipeline.
When the moat is "I can do a thing the customer can't", and the thing is something AI agents are collapsing toward near-zero cost, the moat is not a moat. It is a delay. A delay ending on an unknown date, usually sooner than the holder hopes.
The customer never paid you for your code. They paid you for their code, that you happened to write.
What happens when the compiler moves
In two years, maybe three, the bridge stops needing a translator. An agent asks the radiologist what they need. The radiologist, who has been doing this for twenty years, explains in plain language. The agent builds the tool. The radiologist iterates. No intermediary. No product manager. No roadmap negotiation. No SaaS bill.
This sounds utopian. It is not. It is messy, and it will produce a lot of bad software, and a lot of customers will discover that they didn't actually want to build their own tools, they wanted someone else to build tools they would approve of. That was always the real value proposition of vertical software. Approval, not construction.
But a good chunk of customers will find out they do want to build their own. And when they do, the intermediary disappears.
The asymmetric exit
If you are a founder sitting inside this compression, there is a move that works. Stop being the software. Start being the standard. Or the data. Or the trust layer. Or the certification. Or the regulator's relationship. Become something the customer cannot generate, no matter how good their compiler gets.
The solo founders who survive the next cycle will be the ones who figure out, fast, that the code they wrote was never the asset. The asset was the relationship with the customer, the knowledge of the edge cases, the network of other customers who trust the brand. Everything else is going to become a text file.
There is a line from Jaron Lanier I think about whenever this subject comes up. "The technology is never the story. The story is always what we decide to do with the technology." We are deciding, right now, whether the next five years of AI make founders rich or make customers independent. It is not clear which one we are choosing. It is very clear which one the tools are choosing.