Abstraction and building languages with LLMs

How AI will change software engineering – with Martin Fowler

Topic 5 of 23

Published Nov 19, 2025 · Clip 21:05–24:57 (3:52)

Martin argues the abstraction jump with LLMs is smaller than the determinism/non-determinism jump, and stresses that a key property of high-level languages is the ability to build your own abstractions, referencing the Lisp adage of creating your own language then solving the problem in it. He discusses his colleague Unmesh Joshi's idea of co-building abstractions with the LLM to talk to it more effectively, and the chess example showing LLMs understand chess matches better in rigorous notation than plain English, connecting this to domain-driven design's ubiquitous languages and domain-specific languages.