Martin Fowler on Matters of Non-Determinism (thenewstack.io)

Martin Fowler, the chief scientist at Thoughtworks, emphasizes the transformative impact of AI on programming, likening it to the industry's shift from assembly language to more abstract high-level languages. He argues that while high-level programming languages allowed for more straightforward coding, large language models (LLMs) represent a different paradigm altogether: nondeterministic computing. Unlike traditional deterministic computing, where outcomes are binary and debuggable, LLMs produce variable outputs based on statistical reasoning, which challenges conventional approaches to coding and necessitates a new mindset among developers. Fowler showcases several practical applications of AI in software development, particularly in modernizing legacy systems through generative AI. In his latest annual Radar report, Thoughtworks highlights using AI for semantically analyzing codebases, which aids significantly in understanding complex legacy applications. However, he warns that while AI can assist in code comprehension, its ability to modify code safely remains limited, demanding meticulous scrutiny of each change. He stresses the importance of establishing clearer communication with LLMs to enhance accuracy, suggesting that incorporating concepts from domain-driven design can lead to improved outcomes. By framing LLM use through a lens of precision and tolerances, akin to structural engineering, developers can better navigate the complexities of nondeterministic computing.