Thou Shalt Not Make a Machine in the Likeness of a Man's Mind

In 1965, Frank Herbert put a warning in a science fiction novel that took sixty years to become obvious.

“Once, men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them.”

This is from Dune. The backstory: humanity fought a war to destroy every thinking machine — not because the machines went rogue, but because humans had stopped thinking. The machines worked fine. The humans broke. So they created a commandment: thou shalt not make a machine in the likeness of a man’s mind.

Now look at how founders build software today.

The dependency, not the machine

Herbert’s warning was never about AI. It was about atrophy.

There is a type of founder emerging right now who can ship a landing page in an afternoon, generate a codebase from a prompt, and produce a pitch deck before lunch. They look fast. They look productive.

But ask them why they are building what they are building. Ask what happens when the model they depend on changes its pricing, behavior, or availability. They do not know. They never thought about it. They outsourced the thinking part.

These are not builders. They are prompters with deployment pipelines.

The ground truth argument

At GTC 2026, Jensen Huang was asked whether AI agents would “vapor-code away” legacy software companies. His answer was blunt: no.

Agents do not replace tools. Agents use tools. And because agents work at a scale humans cannot, the demand for those tools explodes. But here is the part most people missed: Jensen explained that agent outputs must be deterministic, verifiable, and controllable. “I can’t have a ‘probably’ work. I need to have it exactly work.”

Agents need ground truth. Structured systems underneath. Tools that keep outputs coherent. The probabilistic layer is powerful. But it means nothing without the deterministic layer beneath it.

This maps directly to how founders should think about building.

The builder’s test

Three layers are emerging in the new stack.

Thinkers. They understand what to build and why.

Tools. They enforce correctness and repeatability.

Agents. They execute within constraints set by the first two.

The people who disappear are the ones who were only ever operating as human agents — doing tasks without understanding the system. They were never thinkers. They were butts in seats. And agents do that work faster and cheaper.

One question separates the two types of founder:

If the AI tool you depend on disappeared tomorrow, could you still ship?

If yes, you are a thinker who uses tools. The AI made you faster, but you were already dangerous. You understand the problem, the architecture, and the reason each piece exists.

If no, you were never building. You were prompting. And you have no ground truth to fall back on.

The real enslavement

Go back to Herbert: “That only permitted other men with machines to enslave them.”

The men with machines today are the infrastructure owners. The compute providers. The model companies. If you cannot articulate why your product exists and how the pieces fit together independent of any specific tool, you are a customer, not a builder. You exist at the pleasure of whoever controls the machine you depend on.

Build with AI. Build aggressively with AI. But build as a thinker, not a prompter.

Because the machine in the likeness of a man’s mind is not the AI.

It is the man who forgot he had a mind in the first place.