"True product taste combined with technical depth is 'very rare' — rare enough that when he encountered it in Joanne Jang at OpenAI, he had to invent a new role around her rather than fit her into an existing box."
"Product taste and conviction drive great products - not user feature requests"
Evidence from the Archive
Unknown
Instagram's filters, square format, and simplicity as craft decisions that created organic sharing behavior.
Granola (AI meeting notes) winning users despite Google Meet, Microsoft Teams, and Zoom all having built-in AI meeting features with native distribution.
Product leader who helped build Instagram in its craft-obsessed early days, scaled Uber's rider experience, and now builds product at OpenAI — giving him a rare vantage point across consumer, marketplace, and AI product craft. Their core argument: There is a threshold level of product craft that makes users willing to overcome massive distribution advantages and switch products.
The evidence is specific: Granola (AI meeting notes) winning users despite Google Meet, Microsoft Teams, and Zoom all having built-in AI meeting features with native distribution.. Furthermore, instagram's filters, square format, and simplicity as craft decisions that created organic sharing behavior.. Cursor, Windsurf, Lovable, and Bolt breaking through against Microsoft Copilot despite Copilot's massive first-mover and distribution advantages..
In Peter Deng's own words: "I do believe there is a level of product craft that will make it so that it's just worth it to switch or try something else. There are a few products out there that I see with this. I think Granola is one of them." (Arguing craft can overcome distribution advantages.)
OpenAI, Instagram, Uber
ChatGPT: product craft on top of foundation models created defensible user experience
Granola: AI note-taker competing against Google Meet, Microsoft Teams, and Zoom through pure craft and delight
Has been the product architect behind ChatGPT, Instagram, and Uber -- three of the most used products in history -- giving him rare firsthand evidence of how craft-driven products can overcome and create structural advantages. Their core argument: Product craft and delight CAN be a moat when execution quality is hard to replicate.
The evidence is specific: Granola: AI note-taker competing against Google Meet, Microsoft Teams, and Zoom through pure craft and delight. Furthermore, facebook: started as a database of human connections, not a technological breakthrough -- moat came from product iteration. Uber: combined existing technologies (GPS, cars, human need) and built tech to optimize, rather than starting with a tech moat.
In Peter Deng's own words: "I do believe there is a level of product craft that will make it so that it's just worth it to switch or try something else." (Arguing product craft can overcome distribution advantages.)
Instagram: simple idea (visual sharing) made iconic through taste -- filters, square format, social mechanics
VP of Product at OpenAI (ChatGPT), first Head of Product at Instagram, Head of Rider Product at Uber, 4th-ever PM at Facebook (built News Feed, Messenger, Groups); now GP at Felicis Their core argument: Product taste and conviction are what separate transformative products from incremental ones -- no amount of user research replaces the craft of getting the product just right.
The evidence is specific: Instagram: simple idea (visual sharing) made iconic through taste -- filters, square format, social mechanics. Furthermore, uber: no technological invention, but connected GPS + cars + human need; operations moat rather than taste moat. ChatGPT: product experience decisions (memory, voice, desktop) shaped adoption beyond the underlying LLM.
In Peter Deng's own words: "It wasn't anything that any other company couldn't have done, but it was that product taste that Kevin and Mike had and conviction that there's a certain sort of vibe, if you will, that people wanted, and building that and iterating." (Using Instagram as the canonical example of taste-driven product development.)
OpenAI
Peter Deng invented a whole new 'model designer' role at OpenAI around Joanne Jang because the combination of technical depth and taste was too rare to train
After managing PMs at Instagram, Uber, Facebook, and OpenAI, Deng's population-level claim is that the taste+technical combo is empirically scarce enough to require structural accommodation
Deng is the most instructive voice on the innatist side precisely because he argues from sample size, not mysticism. His strongest signal comes from Joanne Jang at OpenAI. After managing PMs at Instagram, Uber, Facebook, and OpenAI, he hit one person with Jang's specific combination of taste and technical depth — and his move was not to train it in others.
Instead he codified it: he had Jang write her own job description and built the 'model designer' role at OpenAI around her. That decision is the deepest tell in the archive. If taste were straightforwardly teachable, the right move would have been to turn Jang's approach into a curriculum. Instead Deng treated her as a unique resource to be structurally protected. He makes a similar move about Instagram's origin: the breakthrough, he says, was not technology any other company couldn't have built, but the product taste Kevin and Mike had.
In Peter's own words: "She is the only person that I've worked with that has as much technical depth as she does have product taste. And I just want to pause there. It's just truly special... She has this taste and those two things are very rare to find together." (On Joanne Jang — the population-level statement about how rare the taste+technical combination actually is.)