Context

Peter Sellis was the first product manager at Snapchat, where he spent seven years, and later head of product at Discord. Speaking with no employer, comms team, or book to promote, he gives an unusually blunt account of how he builds product teams, why he thinks the median PM is bad, why Snap's ad business never matched its user base, and why real growth almost always comes from a product's core rather than new bets. The episode matters to PMs and product leaders because it repeatedly argues against comfortable defaults: more collaboration, more attention to struggling teammates, more expansion into new markets, when Sellis's experience points the other way.

The Big Idea

Great product leadership works against several comfortable instincts at once: build teams for autonomy instead of collaboration, invest in your strongest people instead of your weakest, and find growth by going deeper into your core product instead of chasing adjacent ones.

Sellis backs this with real numbers from Snap and Discord: Snap's post-2018-redesign recovery came from making the existing app faster for existing users, not new features, and Discord's fastest growth phase came from doubling down on the exact niche (friends playing games together) where the company already had near-total penetration, not from riding the mid-journey or AI wave that was visible from the outside.

Key Insights

Teams should run on ideology and clear decision rights, not collaboration

Sellis deliberately designs teams the way he thinks a "good terrorist organization" works: a shared, almost religious-level clarity about why the team is doing what it's doing, plus a very clear structure of who is trusted to make which decision, so the org can keep functioning without every decision routing through a leader. His reasoning is that collaboration has a real, under-discussed cost: it guarantees a team moves only as fast as its slowest node, and no executive ever tells people to collaborate less, which he treats as a warning sign that the cost never gets weighed. At high-growth companies, he says the goal is for product managers to make the calls founders themselves would have made a year or two earlier, which only works if the ideology and decision rights are unusually explicit, phrases and shorthand repeated so often they act as shortcuts for the whole strategy.

Spiky, one-of-one talent needs a system built to hold it, not manage it

Sellis managed Nikita Bier (later head of product at X) at Discord and describes it as closer to directing than managing: exceptional in a narrow, high-value lane, and nearly impossible to reach through normal channels (he once threatened to book a paid one-on-one through Bier's own scheduling startup just to get a meeting). His model, borrowed from NBA coach Phil Jackson's ability to blend wildly different personalities into winning teams, is that an organization should treat handling "spiky" talent as a design requirement, not a tolerance. He compares evaluating a spiky product leader to following a band: you can love their work while accepting an occasional "dud album," and the decision to stick with them through it comes down to whether you still believe in them as an artist.

The median PM is bad because of how the talent pool sorts itself

Sellis argues this follows mathematically from a power-law distribution: if the underlying skill of product management is normally distributed across everyone, but only the right tail of that distribution actually becomes PMs, the resulting subgroup is itself a power-law distribution, where the median sits well below the average. Two forces make this worse in practice: during the zero-interest-rate (Zerp) era, product management became one of the most lucrative careers available to non-technical people, so mediocre PMs have a strong incentive to stay in the role rather than leave; meanwhile the genuinely great PMs keep exiting the profession entirely, into founder or executive roles where the same skills pay off even more. Sellis adds a pattern he's observed downstream of this: strong engineers rarely complain about PMs, either because they bend good PMs to their will or simply refuse to work with bad ones, so the loudest complaints about "useless PMs" tend to come from people stuck in the middle of the distribution.

Snap's ad business hit three structural walls, not one

Sellis lists three distinct reasons Snap's advertising business never matched its billion-plus user base, deliberately separate from the common explanation that it's "just because the users are young." First, there's a real, measurable step-change in advertising CPMs between under-17, 18-to-21, and 21-plus users in the US, because tracking and targeting minors is far more restricted, making ROI genuinely harder to prove for Snap's core Gen Z audience regardless of purchasing influence. Second, Snap's core flow opens straight to the camera, a creation-oriented, non-feed interface with no natural place to insert an ad, unlike every major ad-supported product, which opens to a screen built around consumption. Third, Snap is fundamentally a messaging app, and no messaging app anywhere, including the well-cited Asian "super apps," monetizes anywhere near as well per unit of time spent as a feed does. Sellis's summary: Snap tried to build a multi-billion-dollar ads business quickly on a lower-value audience, in a hostile ad format, on top of a mostly utilitarian product, and still got to four billion dollars in ad revenue faster than Twitter, Reddit, or Pinterest despite starting later than all three.

Growth almost always comes from making the core better for people already using it

At Snap, the only quarter where daily active users truly flattened was right after the widely criticized 2018 redesign; the recovery came not from new features but from a sustained focus on raw performance, especially on Android, for people who were already using the app. At Discord, despite outside observers assuming Midjourney and the broader AI wave were driving growth, Sellis says you can't spot Midjourney's impact in Discord's usage data at all: the real 2024 growth push came from focusing the entire team on being the best place for friends to play games together before, during, and after playing, a use case where the company already had effectively 100% penetration among PC multiplayer gamers. The insight he draws is straightforward math: a daily active user is worth roughly 30 times a monthly active user, so nudging existing heavy users from using the product 10 days a month to 12 or 13 is often far more valuable than chasing net-new users who may never retain.

Ride your strongest people harder instead of propping up the weak

Sellis says his instinct runs opposite to most managers': when someone on his team is consistently making good decisions, he keeps loading them with more responsibility and pushes them toward the edge of failure, rather than spending his attention shoring up people who are struggling. He frames this as self-reinforcing, confidence and momentum let strong performers take on more, while he admits he doesn't have a strong mechanism for improving underperformers and effectively lets them fail on their own. The one guardrail he names: not paying close attention to someone isn't the same as that person failing, this approach depends on already trusting that the strong performer knows what to do.

Mental Models & Frameworks

Core Product Value: a phrase plus metrics, not either alone

Both at Snap ("the fastest way to share a moment with the people you care about") and Discord ("the best way to talk and hang out with your friends before, during, and after playing games"), Sellis's teams distilled the product's daily value into a short phrase that breaks into concrete components (fast / share a moment / people you care about), then paired that phrase with hard metrics definitions tracked internally and referenced constantly in growth reviews and onboarding. Sellis argues you can't rely on either half alone: a phrase with no metrics never gets actioned day to day by engineers and data scientists, and metrics with no phrase feel meaningless to the team. Use it as a roadmap filter: when someone proposes a new feature, ask whether it clearly strengthens the phrase's own components, and treat a weak or indirect link as a signal the feature may not deserve priority.

Long-term greedy for ad and marketplace businesses

A phrase Sellis's teams used to justify prioritizing advertiser ROI over near-term ad revenue, grounded in a simple causal claim: tomorrow's ad revenue depends on the ROI advertisers get today, and the easiest lever to boost that ROI in the short term (cutting prices) actively works against long-term revenue growth through retention and rising advertiser budgets. He calls it "greedy" deliberately, framing it as self-interest rather than charity, which gave the team cover to hold the line on advertiser ROI even during hard moments like public-company earnings pressure or the threat of a layoff, when short-term thinking becomes most tempting.

Back-casting instead of forecasting for genuinely new bets

Sellis distinguishes two ways to think about expanding beyond a product's core. One is incremental: identify the core capability people already love (the camera, voice), then look for adjacent cohorts or use cases where that same capability could matter, essentially concentric circles radiating out from what already works. The other, rarer approach he associates with Snap's most creative bets, is back-casting: start from the company's founding philosophy and imagine the future totally unconstrained by current technology (not "how do we improve the camera app" but "what's the best way to share the moment my child just took their first steps"), then work backward to what needs to be built. He says very few people can genuinely think this way; most creative people instead remix the present rather than imagining a future untethered from it.

The three oxymorons of a great PM

Sellis's model for what makes a PM genuinely great is three tensions each person has to hold simultaneously, not resolve in either direction:

• Confidence bordering on arrogance, paired with real humility: enough conviction to drive a path forward and take the credit others assume comes with the job, paired with actively redirecting credit to the designer or engineer who made the real difference.
• Process discipline paired with comfort bordering on discomfort with authority: genuinely valuing good process (not too much, not too little) while being just as comfortable throwing a plan out and redoing it, almost allergic to rigid authority for its own sake.
• Long-term patience paired with a near-annoying bias for action: a clear picture of what the product should look like a year or two out and the patience to sequence the building blocks toward it, combined with a day-to-day urgency that visibly sets the pace for the whole team.

Use it as an interview lens: ask a candidate for a concrete story that shows both sides of one of these pairs, not just one side, since demonstrating only "confidence" or only "process" is the common, weaker failure mode.

Trade-offs & Nuance

Collaboration speeds understanding but slows execution

Sellis isn't arguing collaboration is worthless, he's arguing its cost is systematically under-priced: every added collaborator adds coordination overhead, and a team's effective speed caps out at its slowest node. The trade-off he accepts is that a highly autonomous, ideology-driven team will occasionally make a call a full group discussion might have caught earlier, in exchange for moving at the pace of its best individual decision-makers rather than its most cautious one.

Ads speed-running versus the mechanism that actually earns trust

Sellis is optimistic that OpenAI is "speed-running" the infrastructure of ad-supported products inside ChatGPT, since the core auction mechanics (a Vickrey-Clarke-Groves-style mechanism where advertisers effectively pay more to reach less-relevant users) are largely solved problems from a decade of Meta's work. But he flags a real tension specific to a chat interface: a feed can blend organic and paid content because both compete for attention in the same unified auction, while a chat answer to "what should I buy" is a direct trust transaction, so the real design problem isn't the auction math, it's how much an AI assistant should let advertiser influence shape an answer without eroding the trust that makes people keep asking it questions. He points to Amazon's sponsored-search results, sixteen ads before the first organic result, as an example of a company that seems to be trading trust for revenue at a rate its underlying product quality can't fully absorb.

Practical Application

Write your product's core value as a breakable phrase, then check every roadmap item against it

Following Sellis's Core Product Value model, write a short phrase for what your product delivers daily, broken into two or three named components (the way "fast way to share a moment with your best friends" breaks into fast, share a moment, and best friends). Before prioritizing a new roadmap item, explicitly check which component it strengthens; treat a weak or missing link as a real signal to deprioritize, not just a formality to note and move past.

Look for growth in your existing daily-active base before chasing new segments

Before investing in a new market or user segment, calculate how much moving your current heavy users up the frequency curve (say, from 10 days a month to 12 or 13) would be worth, using the roughly 30-to-1 value of a daily active user over a monthly one that Sellis cites. If that math looks better than acquiring and retaining net-new users, as it did for both Snap post-2018 and Discord in 2024, prioritize performance and depth work on the existing core over expansion.

Interview for the oxymoron pairs, not just one side of each

When hiring a PM, ask for a specific story that demonstrates both halves of one of Sellis's three oxymorons, for example a time they pushed hard for their own conviction and then gave credit away publicly for exactly who made the difference. A candidate who can only describe confidence, or only humility, or only process, or only comfort with ambiguity, is showing you half the picture.

Ask what a strong candidate held back from shipping

Sellis suggests one concrete way to probe for product taste in an interview: ask about ideas they built and held in their hands (a working prototype, not just a tested-and-rejected concept) but chose not to ship, because the timing or the world wasn't ready. A candidate with several strong ideas still sitting on the shelf is demonstrating the restraint muscle Sellis thinks matters more than most people realize, especially as AI makes it easier to add features and harder to practice saying no.

Notable Quotes

"If a museum has all of their collection on the walls, then the curator hasn't done anything." (Peter Sellis, on the discipline of restraint in product decisions)

"There's always money in the banana stand." (Peter Sellis, summarizing the lesson that durable growth usually comes from perfecting the core product, not chasing new ones)

Bottom Line

Peter Sellis's core argument, drawn from Snap and Discord, is that the comfortable defaults in product leadership (collaborate more, support your weakest performers, expand into new markets) are usually the wrong instinct: real speed comes from clear ideology and decision rights instead of consensus, real growth comes from deepening your core product instead of chasing adjacent ones, and real leadership means giving your strongest people more rope rather than spending your attention on the struggling middle.