Revisiting Steve Yegge’s 8 stages of AI-assisted coding from earlier this year, I’m struck by how what once seemed controversial is now mundane (e.g. coding via CLI instead of an IDE).
What feels outdated is measuring progress by how many agents you control instead of how you verify correctness.
The continuum of approaches to testing and reviewing AI-generated code is more interesting than progressing from hand-coding in an IDE to orchestrating 30 agents.
https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16dd04
