Descriptions:
Alexander Embiricos, product lead for OpenAI’s Codex, joins Claire Vo on How I AI for a zero-to-one walkthrough of Codex — OpenAI’s coding agent integrated into VS Code and compatible forks like Cursor. Embiricos covers everything from initial setup and account requirements (Codex is included in ChatGPT Plus, Pro, Business, Team, and EDU plans with generous limits) to advanced techniques that even experienced engineers rarely use.
The demo moves through Codex’s core strengths: answering natural language questions about unfamiliar codebases, generating structured plans before touching code, and running tasks in parallel using a “best-of-N” approach — submitting the same task four times simultaneously in the cloud and selecting the best result. Embiricos also explains how Codex runs on its own cloud compute, enabling fully asynchronous execution: developers can dispatch tasks and move on, returning when the agent surfaces results. He draws a parallel to the shift from a maker’s schedule to a manager’s schedule that many engineers are experiencing as these tools mature.
A standout data point: OpenAI used Codex internally to build a soar app for Android in 28 days, which immediately reached number one in its App Store category. Embiricos also references a same-day model update — GPT-5.2 — visible during the live demo. For software engineers and product leaders evaluating agentic coding tools in 2026, this episode delivers a rare, detailed first-party perspective on how Codex is designed to be used at its highest capability level.
📺 Source: How I AI · Published January 12, 2026
🏷️ Format: Tutorial Demo
