Inside OpenAI’s Codex With Andrew Ambrosino

August 31, 2026

AI Summary

5 min read

OpenAI’s Andrew Ambrosino, a member of technical staff who helped found the Codex team, recounts the moment the team realized a coding agent could be more than a developer tool. "The entire comm team was just using the codex app to do all of their stuff. And they would just like hide the diff panel and ignore the code," he says. That internal adoption by non-technical people—people who didn't care about the code itself, only what the agent could do for them—was the moment he understood that coding agents were "for everything."

The App in the Middle

The first version of Codex, launched about a year before this conversation, ran on the web as a harness where you'd fire off a task and get a pull request back. But the models weren't quite ready for one-shot coding, so the industry moved toward interactive terminal-based CLI agents. Then came IDE extensions that integrated with VS Code and Cursor. When Ambrosino's small team of four looked at the landscape, they saw a crowded space between terminals and full IDEs—and they decided to build an app that sat right in the middle.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

Never miss an episode of AppStories

Get every new episode summarized in your inbox — free, ~5 minutes to read.

No spam. Unsubscribe anytime.

What you'll learn

1 (00:11) **Introduction and Guest Background** - Hosts John Vorhez and Federico Vitichi welcome Andrew Ambrosino from OpenAI, who led the team that created Codex.
2 (01:42) **The Evolution of Codex** - Andrew traces the product's history from a web-based coding agent to a CLI, then an IDE extension, and finally a standalone Mac app.
3 (04:27) **The Core Thesis: Codex as a Universal Agent** - Andrew explains how the team now separates Codex the product from Codex the coding agent harness.
4 (05:36) **Betting on a GUI App vs. Terminal** - Andrew discusses the counterintuitive decision to build a native app when everyone was using terminal-based coding agents.
5 (08:49) **Why Not Build an IDE?** - Andrew explains the deliberate choice to avoid creating a bad IDE by removing file editing from the first release.
6 (10:32) **The Model Quality Bet** - Andrew reveals that the app's design was shaped by anticipating future model capabilities.
7 (12:47) **Non-Developers Using Codex** - Andrew discusses how non-technical users began adopting Codex even though the app was "hostile to non-techno people."

+ Full timestamped outline available in the app

Guests on this episode

Andrew Ambrosino

Show Notes

This week on AppStories, Federico and John are joined by Andrew Ambrosino, the Codex team lead at OpenAI, to talk about the evolution of Codex, the unification of ChatGPT and Codex, Computer Use and Computer History, designing for user trust, and more.

On AppStories+, Andrew shares some of his favorite Mac apps and his Mac origin story.

Also available on YouTube here.

Subscribe here.

We deliver AppStories+ to subscribers with bonus content, ad-free, and at a high bitrate early every week.

To learn more about an AppStories+ subscription, visit our Plans page, or read the AppStories+ FAQ.

This episode is sponsored by:
Finalist – A paper-inspired daily planner with notes for iPhone, iPad, Mac, and now Apple Watch. A day you design your way. For 50% off your first year use this link.

Links and Show Notes