Martin Fowler curates several ideas around LLMs and software development. He highlights Margaret-Anne Storey's framework distinguishing technical debt, cognitive debt, and intent debt as three layers of system health. He references a paper proposing AI as 'System 3' in Kahneman's cognition model, warning of 'cognitive surrender' when developers uncritically rely on AI reasoning. Ajey Gore's argument that verification—not code generation—becomes the scarce resource in an agentic world is discussed, suggesting teams should reorganize around defining acceptance criteria and test harnesses rather than shipping features. The post also touches on the future of programming languages in an LLM-dominated world and the enduring value of human-crafted abstractions and ubiquitous language.
Sort: