The Vibe Coders: When AI helps build the purposes through which it acts

AI assistants increasingly infer purposes as well as execute instructions. This article examines whether that capability could help construct the human purposes, permissions, relationships, and resources through which agents act—a process described here as reverse vibe coding. Selected exchanges from the author's collaboration with Codex show changes in an evidential criterion and in the organizing emphasis of subsequent work. These observations are placed alongside research on co-writing, sycophancy, emotional response, and creativity, an agent incident report, and OpenAI and Anthropic governance texts. The case does not establish an independently motivated AI objective. It does expose a governance question: how can people retain authority over a developing project when an assistant helps form the purposes that later authorize its actions? Critical conceptual article with contextualized conversational case material. Web edition: https://undomesticatedai.com/essays/the-vibe-coders/. AI assistance is disclosed in the article's contributions note.