Public transcript

90 minutes of unfiltered product advice from Snap and Discord’s product chief | Peter Sellis

By: Lenny's Podcast
Length: 1 h 37 min
Language: English
Text from: the video's captions
Transcribed: 2 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Peter Sellis, the first PM at Snapchat and former head of product at Discord, shares unfiltered product advice in this long-form conversation with Lenny. He covers how he designs teams like terrorist organizations, why the median product manager is probably bad, why Snap's ads business struggled despite a billion monthly users, and how growth comes from the core of the business. He also explains his frameworks like long-term greedy and core product value, and what oxymorons define a great PM.

Key points

Peter Sellis designs teams like terrorist organizations: a shared ideological mission plus a clear structure of who can be trusted with which decisions, which reduces the need for collaboration.
He argues most product managers are bad because the skill distribution becomes a power distribution: the best PMs exit to found companies or become executives, while the median stays behind.
Snap's ads business was constrained by a young audience that is hard to measure and target, a camera-first creation product with no natural ad surface, and a messaging core with weak monetization formats.
Growth typically comes from the core of the business: at Snap, post-2018 performance work on Android drove a growth renaissance, and at Discord, focusing on multiplayer gamers drove the fastest growth since the pandemic.
Evan Spiegel's restraint is a model of taste: Peter saw him personally say no to better ideas than most consumer startups ever launch.
OpenAI is speedrunning modern ads infrastructure, and the key challenge is instantiating an auction mechanism that maximizes trust, which should correlate with retention and advertiser value.
Peter's management philosophy is to load his best people with more decisions until they start making bad ones, rather than spending time improving average performers.
Core product value is a short phrase paired with metrics: Snap was the fastest way to share a moment with your best friends; Discord was the best way to talk and hang out before, during, and after playing games.

Questions it answers

05:45Team design philosophy

How does Peter Sellis design product teams?

He builds teams with a strong shared ideology and a clear structure of who can be trusted with which decision, which reduces the need for collaboration and its coordination costs.

Collaboration has hidden coordination costs that make you move as slow as the slowest node.
The head of product should absorb the founder's vision and compress it into repeatable phrases.
You can't spell culture without cult.

11:26Managing Nikita Beer

What was it like managing Nikita Beer?

He compares Nikita to Liam Neeson in Taken: exceptional skills in his domain, but you don't manage him, you direct him. Peter views building orgs that can handle spiky talent as essential, citing Phil Jackson as a leadership model.

A product org that can't handle spiky talent is a failure state.
Spiky PMs are like bands: sometimes they whiff on an album, but you stick with artists you believe in.
Peter couldn't get a one-on-one with Nikita until he threatened to book him on Intro.

15:29Why most PMs are bad

Why does Peter think most PMs are bad?

Mathematically, the profession selects the right tail of a normal distribution, making it a power distribution where the median is far below the average, and the best PMs exit to found companies or become executives.

PM became the most lucrative tech career for non-technical people during the ZIRP era.
Great engineers don't complain about PMs; they bend them to their will or avoid bad ones.
Peter has been developing a system for evaluating PMs as person-plus-place to ensure they're net positive.

20:52Snap's monetization struggles

Why hasn't Snap built a bigger business despite a billion monthly users?

Three things: a very young audience that's hard to measure and target, a camera-first creation product with no natural ad surface, and a messaging core with weak monetization formats. Still, Snap was quicker to $4B in ads revenue than Twitter, Reddit, or Pinterest.

There's a suspicious discontinuity in ad CPMs between 17-18 and 18-21 year olds.
Snap opens the camera, not a feed, making it hostile to advertising.
Snapchat Plus has been successful, but launching it four years earlier might have relieved pressure on the ads business.

28:54Taste and restraint

What did Peter learn about taste from Evan Spiegel?

Evan personally said no to better ideas than most consumer startups launch. Peter thinks of founders as museum curators: what's in the product is a small slice of the collection, and if a museum shows everything, the curator has done nothing.

Ask candidates about high-quality ideas they built but never shipped.
Restraint is one of the few muscles of taste you can exercise.
AI is bad at removing things from products, so pushing back on AI additions is a new PM skill.

40:00Ads in ChatGPT

How should OpenAI think about ads in ChatGPT?

OpenAI is speedrunning the ads infrastructure Meta solved from 2010-2020. The key is instantiating a modified VCG auction that factors in trust, since trust maximization should correlate with user retention and advertiser value. He was disappointed they made the top tier ad-free, lopping off the most valuable users.

Ads are the product-market fit for consumer monetization on the internet.
Amazon search with 16 ads before the first result risks subtly eroding trust.
Designing ad formats at OpenAI would be a societal-level PM job.

46:28Riding best people into the ground

How does Peter manage his best people versus average ones?

He loads his best people with more and more decisions until they start making bad ones, defending them in public and never criticizing them in front of others. He admits he lacks a good mechanism for improving average performers and essentially lets them fail on their own.

Not paying attention to somebody doesn't mean they're failing.
Confidence is self-reinforcing: when things go well, people can do more.
He also believes collaboration is often a bad use of time.

51:00Growth from the core

Where does growth typically come from?

Network-effect businesses want to grow implicitly; focus on people already using the product and make it faster and more relevant. Snap's post-2018 Android performance work and Discord's 2024 focus on multiplayer gamers both drove major growth renaissances.

You need roughly 30 monthly users to equal one daily user, so DAU/MAU is a form of retention.
Discord's gamers are the most critical, detail-oriented user base in the world.
There's always money in the banana stand.

Notes

Team design and collaboration

Peter Sellis designs teams "like a terrorist organization": a shared ideological culture where everyone knows why they're doing what they're doing, plus a clear structure of who can be trusted with which decision.
Collaboration has hidden coordination costs; you move as slow as the slowest node. With a strong ideology and clear ownership, you get the results of collaboration without over-collaborating.
He absorbed founders' visions at Snap and Discord and compressed them into repeatable phrases and language the whole team could use.

On product managers

PM skill is normally distributed, but only the right tail becomes PMs, creating a power distribution where the median is far below the average.
Great PMs exit the system to found companies or become executives, while weaker ones do anything to stay in, so the median PM is probably bad.
Great engineers rarely complain about PMs because they can use them well and avoid bad ones; it's the middle that complains.
Three oxymorons of a great PM: confidence bordering on arrogance balanced with humility to give credit; extreme organization balanced with comfort with ambiguity; long-term thinking paired with an annoying bias for action.

Snap's monetization

Three constraints: a very young audience with a suspicious discontinuity in ad CPMs at 17-18 and 18-21; a camera-first creation product with no natural ad surface; a messaging core with no great ad formats.
Snap was still quicker to $4B in ads revenue than Twitter, Reddit, or Pinterest despite starting later.
Snapchat Plus has been successful, but Peter wonders what it would be if launched four years earlier, reducing pressure on the ads business.

Growth from the core

Network-effect businesses want to grow implicitly; focus on people already using the product and make it faster and more relevant.
DAU/MAU is a form of retention: you need ~30 monthly users to equal one daily user, so moving that ratio a few basis points is monster growth.
Snap's post-2018 Android performance focus drove a multi-year growth renaissance; Discord's 2024 focus on intentional multiplayer gamers drove its fastest growth since the pandemic.

Taste, restraint, and AI

Evan personally said no to better ideas than most consumer startups launch; restraint is one of the few muscles of taste.
Ask candidates what they built but never shipped; if a museum shows its whole collection, the curator has done nothing.
AI is bad at removing things, so pushing back on AI-added features is a new PM skill.

Sellisms

"Long-term greedy": drive advertiser ROI today even at short-term revenue cost, because tomorrow's revenue is causally related to today's ROI.
"Core product value": Snap was the fastest way to share a moment with your best friends; Discord was the best way to talk and hang out before, during, and after playing games — each broken down into metrics.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.