OPENAI’s Reflection: PRD Didn’t Die, PM Don’t Cut

Human progress always swings left and right — like learning to ride a bike, moving forward through constant wobbling.

This year, AI-native organizational reform has become an industry trend, but OpenAI’s Codex lead shared a lot of reflections on a podcast.

After reading so many radical articles, this reflection serves as a useful correction.

Andrew Ambrosino, product and engineering lead for the Codex app. Codex is probably the most aggressively AI-driven product team on the planet — nearly 100% of OpenAI employees use it weekly, not just engineers but finance, legal, marketing. Five million weekly active users and growing.

He recently appeared on Lenny’s podcast. Everyone expected him to talk about how AI will upend everything. Instead, he said a bunch of counter-trend things.

PRD didn’t die.

He put it bluntly: “I actually don’t believe this at all.” There’s a weird phenomenon now. Non-engineers, previously locked out by code, can finally build prototypes and excitedly declare documentation useless. Engineers, meanwhile, have started writing tons of documentation that nobody reads. Both sides are cutting corners.

The issue isn’t documentation versus prototypes. It’s whether you’ve chosen the right medium for what you need to express. When direction is still fuzzy, documentation is more useful than a prototype. When you need to test interaction feel, a prototype is more useful. Blindly betting on either side is just calling laziness “innovation.”

“Don’t kill the product role — that’s a terrible idea.”

He said some companies are already doing it. Remove all PMs, everyone is a builder. His take: “And then what happened? They threw away all the best practices accumulated by an entire discipline.”

The PM role wasn’t invented out of thin air. It carries a mountain of hard-won experience from real attempts and failures. Cutting the product role wholesale doesn’t make product problems disappear. It just means no one is responsible for product problems anymore.

He said it’s a good thing that phrases like “this isn’t your lane” are disappearing, and boundaries are thinning. But thinner boundaries don’t mean roles vanish.

Should designers write code? Yes. Should PMs write code? Yes. But knowing Excel doesn’t qualify you to work in finance.

The design process isn’t completely dead.

Jenny, design lead for Claude Code, said on the same podcast that the design process is dead. Andrew’s response: both true and not.

True part: if you’re tied to specific tools and daily rituals, that process is indeed dead. Now you can drop a prototype straight into the product for A/B testing — who still follows that academic workflow? Wrong part: the awareness of “what stage of the process you’re in” is more important than ever.

The problem now is that prototypes look too much like finished products. An early exploration, because AI generated it so completely, everyone assumes it’s ready to ship. It’s not that the process is dead. It’s that the signals from the process have decoupled from the medium.

He offered a more honest framework for role dissolution.

Your role is the average of what you actually do every day.

Designers write code, PMs write code, but on average, each person’s center of gravity is still different. If you write more code today, you lean more toward engineering. If you’re triaging and coordinating, you lean more toward product. Roles haven’t disappeared — they’re just no longer defined by titles, but by what you actually deliver.

This framing is harder than “everyone is a builder.” It demands that you know what you’re actually good at.

The cleverest part of the entire conversation is Andrew’s position.

He doesn’t need to defend traditional roles. He’s using AI to eat everyone’s workflows. Codex went from a developer tool to the entire company’s home base. He’s the lion — no need to speak for the sheep.

Yet he chose to pull back.

He said it took him ten to fifteen years of continuous failure to get here. He said this is the first time in his life he doesn’t feel like a failure. A man who failed for fifteen straight years, standing at the most radical vantage point and looking back, doesn’t say “the old world is over.” He says “don’t throw away the good stuff.”

His last line was recorded after the show ended.

Don’t marry your process. Marry what you can uniquely deliver.

The reverse is the real point: you have to first know what makes you unique.

The faster the tool, the more expensive judgment becomes. AI has driven implementation costs to zero, which in turn makes the things that cannot be automated more valuable. A PM’s judgment, a designer’s taste, an engineer’s architectural intuition — these aren’t acquired through prompts, but through the accumulation of mistakes within a discipline.

Andrew shared a detail. If Codex had launched three months earlier — same product, same form — the market reaction would have been completely different. It wasn’t the product that changed; it was the underlying model. Whether a feature is good sometimes has nothing to do with the feature itself, and everything to do with whether you can wait for the right moment.

The people who can build something and wait for the model to catch up — they don’t rely on prompts. They rely on judgment.

Podcast link: https://www.youtube.com/watch?v=P3KDebPTUrw

Podcast transcript: Podwise

Writing and editing: Cola + Lite model

Writing skills used: writing-style-guard · orange-writing-style

Cover and illustration skills used: orange-line-illustration

Article guidance and hook: 橘子