
AI, Obsolete Skills, and New Leverage

The State of Software Engineering 🔨 — with Kent Beck
Topic 7 of 7
Published Feb 15, 2024 · Clip 42:44–59:08 (16:24)
The conversation turns to AI's impact on Kent's skills and workflows. Kent recounts an experiment using his Rentakent personalized model to write an executive briefing of Tidy First, discovering he now writes 'goal by goal' rather than word by word, and that his structuring skill became far more leveraged while word-level writing skill became less so. He describes prompts as a new form of literature between writing and programming, and envisions books becoming like vinyl releases drawn from a personal model of ideas. Asked how to TDD an LLM application, he describes an experiment with two chatbots writing tests and code for each other, and explains that there's a minimum amount of information you must specify, though models with rich context can eventually accept more concentrated prompts and produce reliable code.