Kent Beck interviews Dan Blumenau, co-founder of OpenProse, a framework that lets developers write programs in structured English and execute them using AI agents like Claude Code. Each component uses requires/ensures blocks (similar to preconditions/postconditions) and an inversion-of-control container wires components together by matching these blocks — all in prose, run by an LLM. Sub-agents manage context by passing file paths instead of raw data. Beck reflects on how this approach represents a broader industry shift toward declaring outcomes rather than specifying implementation steps, with parallels to Codex's /goal feature and Anthropic's Outcomes API.
239 Impressions