The Silent Revolution Rewriting Software Itself
How machines learned to program and why the transformation is far larger than AI writing lines of code - arxiv.org/abs/2511.18538v4
A New Epoch in the History of Software
Software used to be humanity’s most carefully handcrafted creation. Lines of logic flowed from human minds into text editors, shaped by discipline, intuition, and endless debugging cycles. But as we read through the sweeping 300 page survey on code large language models, a deeper truth emerges. In the span of only a few years, the craft of programming has begun to shift from human authored to machine reasoned. The transformation is not the polite automation of simple tasks. It is a revolution occurring quietly inside GPUs, where models digest entire repositories and build a latent understanding of algorithms, patterns, and the physics of software itself.
The evolution is shown in the figure below, which traces six eras of software development. It begins with manual coding in the 1960s, inches through the slow adoption of tools and frameworks by the early 2000s, and then suddenly accelerates in 2020 when large language models entered the scene. The figure’s final stage, labeled AI Autonomous, looks less like a continuation of the past and more like a rupture. Instead of humans teaching machines how to think, machines begin teaching themselves, guided by sprawling corpora and intricate training recipes.
