Inside OpenAI’s Codex With Andrew Ambrosino
Summary
Andrew, a founding member of OpenAI’s Codex team, discusses the evolution of the Codex app from a terminal-based agent to a unified macOS application integrated with ChatGPT. He explains the strategic shift toward a GUI-first approach, the implementation of "computer use" capabilities, and the design philosophy behind building a cross-platform "super app." The conversation also covers OpenAI’s bottom-up product development culture and the balance between rapid innovation and user trust.
Topics discussed
Introduction and guest Andrew Amrosino from OpenAI
History of Codex: From CLI agent to standalone app
Designing a fluid GUI surface for coding agents
Why not build an IDE? The 'middle ground' thesis
Evolution towards Computer Use and agent capabilities
Integrating Computer Use and History into the product
Trust, elicitation, and the psychology of agent cursors
Native Mac app vs. Web technologies and cross-platform
The 'Super App' concept and unifying Chat, Work, and Codex
OpenAI's internal culture: Everyone builds everything
Product restraint and avoiding feature bloat
Automated testing loops and benchmarking agent performance
Future of macOS integrations and local capabilities
Bonus: Andrew's history with Mac development and design
Bonus: Current Mac setup and favorite apps
Listen ad-free on Castria