Programming as Theory Building: Peter Naur's Essay

Christian Ekrem wrote a great article revisiting Peter Naur's 1985 essay, Programming as Theory Building, and applying it to today's generative AI landscape.

The key ideas:

A program is not its source code. Code is just a lossy representation of the theory — the shared mental model of how a system works, why it works that way, and how it should evolve. When developers leave, the theory dies with them.

Today's "theory-less code" crisis is driven by three factors: reflexive AI usage (accepting code without understanding it), domain-blind LLM-generated code (statistics without business understanding), and the integration problem (importing foreign architectural decisions that contradict the domain model).

Senior developers matter more than ever as theory builders, architectural coherence guardians, intentional AI collaborators (using AI for mechanical tasks while preserving human judgment for what matters), and mentors who transfer deep understanding.

AI cannot build theory. LLMs generate syntactically correct code but cannot understand business context, make thoughtful trade-offs, or maintain conceptual integrity. The question isn't whether they can write code — it's whether we can maintain the human theoretical frameworks that turn scattered code into coherent programs.

Original source: Programming as Theory Building