Cooking with Codex — Charlie Guo & Gabriel Chua, OpenAI

This post details the practical application of AI agents, specifically Codex, in transforming individual requests into repeatable software development processes. It showcases building a macOS translation app from a Slack conversation, demonstrating the integration of context, plugins, and supervision. The talk highlights the creation of a shareable site from repository history, IKEA-style repository visualization, dashboard data download, and a Google feedback form builder via a Chrome extension. Longer projects include a live Q&A site, an internal form builder, and a Python package with a Rust backend. Key technical contributions include the use of goals for verifiable completion criteria, subagents, visible threads, audits, and progress dashboards for work division and blocker exposure. Remote threads for browser testing and hooks for deterministic checks are presented. Automations for server provisioning, Slack feedback review, and bug addressing are demonstrated. The custom interface Retrodex showcases App Server integration, emphasizing the transition from individual requests to a software factory model.