KnowledgeSketches|← All EpisodesLenny’s PodcastResearch papers
From Lenny's Podcast
Knowledge Sketch

OpenAI Codex Lead on the
New Shape of Product Work

Andrew Ambrosino
Product and engineering lead for the Codex app at OpenAI
JUN 28 2026
The Inversion

The Process Flipped:
Building Got Cheap

BUILDINGWAS THE COSTTASTEIS THE COST NOW
"The implementation is actually not the expensive part anymore. It's, dare I say, taste."
  • Starting from scratch, you can talk to the models and stand up whatever feature you want.
  • Give people unlimited tokens and everybody builds everything. He is sure one feature they badly need has 90 uncoordinated teams trying it.
  • The old process de-risked implementation up front with documents, research and prototypes, because building was expensive. That has changed.
  • The work now is curation. Of those 90 attempts, what is good about them, what to fold in, how to frame it.
Framework

Taste Matters Most. Pick the Medium.

THE POINT?DOCUMENTPROTOTYPEFOR CLARITYTO STRESS-TEST
  • Taste has an aesthetic part. It is also how a thing fits the system, where the product is going, and how you present it.
  • He thinks the real taste question is this. If we can build anything, what is the goal here and how do we get there.
  • Pick the format for the point. Product clarity in a vague area might be a document. Stress testing an interaction is a prototype.
  • The medium used to signal where something was in the process. If it felt like the real app, it was late. Now those are sort of divorced.
Andrew's take"The taste to know what to work on, how to present that information, how to achieve the goals, what medium to use is emerging as the most important thing to do."
How They Work

Inside the Codex Team

  • More role collapse than elsewhere, but roles remain. Designers speak engineer. Product managers speak technical language and write code.
  • People are the average of where they work. Add up what a designer here does and there is plenty of code and product work. People are defined less by the fence between roles.
  • Product runs as zone defense. Spread out, find the gaps, cover the whole company. Two product people working too closely is often not a good signal.
  • The app is shaped by a dog fooding loop. The team tries to do as much as possible inside the app, even when it is not the best tool, so it can become the best tool.
The team size answer

Asked how many people are on Codex, Andrew says "somewhere between 10 and a few thousand." The team itself: double digits of engineers, probably half that in design, a few product people.

Everyone manages now

"If you are an IC, you're not typing code out character by character. You are managing something. You're managing agents."

Playbook

You Might Release It Six Times

  • At his last company, whether a feature was good came down to whether the model was smart enough, not its shape.
  • When something is not working, do not be stubborn and call it a bad feature. It might not be ready yet.
  • There, they prototyped every idea, decided which were ready now, and let the rest sit. At every leap in models, they tried them again with the new one swapped in.
  • Writing the code does not mean it should go out. It is an artifact to test against future models.
The Codex lesson"I am very confident that the Codex app that we released in February, if that had been ready in November, it would've absolutely failed in the market. And the only difference was the models between November and February."
Contrarian

What Andrew Pushes Back On

✗PRDs are dead, prototypes are in.INSTEAD →✓ Andrew does not buy it. When building is abundant, what matters is picking the right format for the point. Product clarity in a vague area might be a document.
✗Kill the product role, everyone's a builder.INSTEAD →✓ He calls it a terrible idea. A whole discipline with real best practices gets abandoned because people say, "Oh, I wrote some code."
✗A polished prototype is ready to ship.INSTEAD →✓ Looking production ready no longer means de-risked. Say which stage you are in, or people anchor on what was meant to be an exploration.
✗A feature that failed in market is a bad feature.INSTEAD →✓ It might not be ready yet. You might need to release it six different times before it works, and the shape might not change at all.
Based on Andrew Ambrosino's episode on Lenny's Podcast. All ideas on this page are from the episode.Watch on YouTube
Every sketch here began as a transcript.

Send us yours and we’ll draw it. $1 a sketch, or $49 for all 11 formats.

Sketch mine →
𝕏︎ X / Twitterin LinkedIn📸 Instagram🔗 Copy link
0:00