What a brick teaches us about AI-driven product development

What does a brick have in common with AI-driven product development?

Louis Kahn, one of the greatest architects of the 20th century, once said: You say to a brick, “What do you want, brick?” And brick says to you, “I like an arch”.

He didn’t impose a design on the material. He listened to what the material could do, then designed with that as foundation.

Most AI development workflows today? They’re writing detailed blueprints for bricks they’ve never touched.

Here’s the uncomfortable truth about spec-first AI tools such as spec-kit:
They’re optimizing for the wrong thing.

When you’re in unfamiliar territory, new domain, unclear requirements, uncertain constraints, you don’t know enough to spec accurately.

You’re not clarifying requirements. You’re hallucinating them, alongside with your AI. The spec becomes a confidence trick you play on yourself:

“Look at all this structure! We must understand the problem now.”

But you don’t.
You just have a very well-formatted description of your current confusion.

The architect knows better

Before designing the arch, Kahn understood:

• How brick bears load (compression strong, tension weak)
• How brick fails (predictable fracture patterns)
• What brick enables (modular, thermal mass, specific geometries)

The arch emerged from understanding the material’s properties, not from imposing a preconceived form.

In software, this means, before generating a detailed spec, you should:

• Build a small piece
• Touch the actual constraints
• See what breaks
• Learn what the system “wants to be”
• Then document what you discovered

The difference really lies on two very different mental models - one I’m guilty of not being able to switch from myself:

Architect thinking: “I understand the material, so here’s what it can become”
Spec-first thinking: “I wrote a detailed document, so the material will comply”

One discovers structure. The other hallucinates it.
Try all you want, a brick can never become a beam without external support.

The question that changes everything:
When you’re using AI to build software, are you asking it to generate plans for materials (assumptions) you haven’t tested? Or are you using it to maintain memory of what you’ve learned from the materials you’re actually building with?

That distinction determines whether you’re doing architecture or writing fiction.

Next in this series: Why “unknown territory” is exactly when you should spec less, not more.