Matias Castello, Product Lead at Alchemy, is using OpenAI's Codex in production for code review and shipping side projects. The conversation, hosted by Romain Huet, Head of Developer Experience at OpenAI, runs through concrete workflows: how Codex integrates into Alchemy's product and engineering process, and how Castello uses the Codex App Server to build outside of work hours.
The most technically specific section starts around the 18-minute mark, covering Codex App Server projects and computer use. Castello also maintains a recurring personal benchmark: he rebuilds Snapcat, a cat-themed Snapchat clone, every time a new model ships. That makes it a live eval, not a toy. It is a simple, repeatable signal for model capability that any builder can replicate with their own project.
The closing advice segment targets people who are not yet building with these tools. The episode is worth reading in full if you want to understand how a product leader at a crypto infrastructure company operationalizes AI coding tools across both professional and personal work, and why a deliberately silly side project might be the most honest way to track model progress.
[WATCH ON YOUTUBE →]