OpenAI's Core Product Engineering lead Akshay Nathan discusses the launch of ChatGPT Work and how Codex evolved from a developer coding tool to a platform for general knowledge workers. Key topics include the shared agent harness underlying both Codex and ChatGPT Work, why OpenAI merged the experiences rather than keeping them separate, and how features like Sites, artifacts, agentic spreadsheets, memory, and sub-agents are changing what people can delegate to AI. The conversation covers the unexpected adoption of Codex by non-developers inside OpenAI, the product philosophy behind simplifying model selection, and OpenAI's roadmap to bring useful agents from software engineers to all knowledge workers and eventually everyone. Akshay also reflects on how AI is blurring role boundaries, why ideas and taste become bottlenecks when anyone can build, and how to measure meaningful productivity versus AI-generated motion.
