Peter Sellis, former first product manager at Snapchat (7 years) and Head of Product at Discord, shares unfiltered lessons on product leadership, team design, and consumer product strategy now that he’s between roles and free from corporate communications constraints.
Designing teams like terrorist organizations
The core insight: high-growth companies need teams that operate with minimal coordination overhead while maintaining perfect alignment, which terrorist organizations achieve through ideological clarity and explicit decision rights.
Collaboration has hidden coordination costs that force teams to move at the speed of the slowest node; aggressively internalizing this cost enables faster execution.
Product managers at founder-led companies should make decisions the founders would have made 6–12 months ago, requiring deep internalization of founder vision.
Two pillars create this dynamic: an ideological culture where strategy functions like a religion everyone understands, and a crystal-clear organizational structure defining who owns which decisions (single-threaded owners, clear “throats to choke”).
Sellis distills founder vision into repeatable phrases and language shortcuts that act as “synctices” for the entire strategy, enabling autonomous decision-making without constant collaboration.
The “terrorist organization” label is deliberately provocative to force attention on these principles; the actual model is about autonomy within a shared ideological framework.
Cult versus terrorist org: the key distinction
Cults center on the leader’s personality; terrorist organizations center on a higher ideology or religion, making them more durable and scalable.
The distinction matters because product organizations need mission alignment that survives leadership transitions, not personality dependency.
Managing Nikita Beer
Nikita Beer (acquired via his startup at Discord) is a “one of one” operator with a very special, spiky skill set — likened to Liam Neeson in Taken: unstoppable at targeted missions (growth, virality), but ill-suited for routine operational work.
Managing him meant directing rather than managing; Sellis once threatened to book time on Intro just to get a 1:1.
The ability to absorb and productively deploy spiky, non-conforming talent is essential; organizations that can’t handle it are in a failure state.
Sellis models this on Phil Jackson’s coaching: meshing wildly different, often conflicting personalities into championship teams by respecting their artistry and riding out “dud albums.”
Why the median PM is probably bad
Mathematical explanation: PM skill is normally distributed, but only the right tail enters the profession, creating a power-law distribution where the median falls far below the mean.
Two forces worsen this: (1) ZIRP made PM the most lucrative non-technical tech role, attracting people who stay for compensation despite low aptitude; (2) great PMs exit the system to become founders or executives, while weak PMs cling to the role.
Result: a self-reinforcing adverse selection where the median PM is net negative.
Great engineers rarely complain about PMs — they bend them to their will or avoid bad ones; bad engineers hide behind PM decisions; only mid-tier engineers feel the pain.
Is the PM profession getting better?
Yes, gradually: higher interest rates raise opportunity costs, reducing low-value hiring; the market is correcting.
Sellis has developed a system to evaluate PMs in context (person + role + time) to ensure net-positive teams.
Snap’s three ads monetization challenges
Young audience: Snap’s user base skews extremely young (Gen Z/Alpha); CPMs drop discontinuously at ages 17 and 18 due to tracking/measurement restrictions for minors, making ROI harder to prove.
Camera-first, not feed-first: Snap opens to the camera (creation-oriented), not a consumption feed; no natural ad slot exists in the core capture-and-send flow. Camera monetization proved intractable despite heavy experimentation.
Messaging utility: Core use case is messaging; no messaging app has cracked high $/time-spent ads monetization (Asian super-apps included). Messaging is a utility, not a consumption surface.
Despite these structural headwinds, Snap built a multi-billion dollar ads business faster than Twitter, Pinterest, or Reddit — a underappreciated achievement given the degree of difficulty.
Snap’s underrated product track record
Snap holds #1 position among Western teens for messaging, maps, camera, and astrology — an extraordinary consumer product portfolio largely built by the OG design team.
The narrative of “failed to live up to potential” overlooks this track record and the unique difficulty of monetizing a creation-first, youth-skewed, messaging-centric product.
Systems thinking
True systems thinking requires mathematical modeling: mapping stocks, flows, levers, and distributions in a spreadsheet to clarify causal logic, not just asking “why” five times.
Most people claiming “systems thinking” are just doing first-principles reasoning; the real leverage comes from quantitative system dynamics (e.g., Jay Forrester’s work).
Building a spreadsheet model — even a private one — forces precision on variables, feedback loops, and leverage points.
Instagram copying Snapchat’s worst features
Instagram aggressively copied Snap features, including ones that weren’t working, assuming everything Snap touched turned to gold.
Stories format ultimately fits Instagram better; Instagram’s version now surpasses Snap’s due to superior architecture for consumption and Adam Mosseri’s taste-driven iteration.
Software moats
Evan Spiegel’s view: software moats are eroding; focus must shift to hardware and ecosystem/network effects.
Sellis respects this but notes investors/public markets may not give SpaceX-style runway for hardware bets; Snap’s ads monetization ceiling constrains the capital needed for ambitious hardware.
Taste, museum curators, and the art of saying no
Evan Spiegel says no to better ideas than almost any consumer founder launches; this restraint is a trainable “muscle of taste.”
Analogy: a museum curator’s job is selecting what not to put on the walls; if everything is displayed, the curator has failed.
AI exacerbates this: it adds relentlessly but never suggests removal. PMs must become editors who cut AI-generated bloat.
Interview test for taste: ask what high-quality prototypes they built but never shipped because “the world wasn’t ready” or “it wasn’t ready.”
OpenAI, ads, and the trust auction mechanism
Ads remain the internet’s true product-market fit for monetization; OpenAI is “speedrunning” the modern ads stack (last-mover advantage).
The core mechanism: a modified VCG auction where organic value (or in chat, trust) is factored into bids, unifying organic and paid content in a single auction with dollars as denominator.
In chat, the tradeoff is trust: how much advertiser influence is acceptable before user trust erodes? Amazon search shows early signs of this tension (excessive sponsored slots).
OpenAI’s challenge: mathematically instantiate trust maximization in the auction to align long-term retention with advertiser value.
Format questions remain (text vs. voice); ad format PM/design roles at OpenAI are societal-level opportunities.
Disappointment: OpenAI gated ads behind the top tier too early; ads could be additive to free-tier chat if designed for trust.
Counterintuitive lesson: ride your best people hard
Human instinct: save the drowning person. Sellis does the opposite — load high performers with more rope (responsibility) until they approach breaking point.
Publicly defend them always; privately push them to their decision-making limit. Confidence compounds; people rise to the trust placed in them.
Risk: neglect ≠ failure. Some people need autonomy, not attention; distinguish between “I trust you” silence and “I’ve given up” silence.
Why collaboration is often overrated
If a team principle is something no one can disagree with (e.g., “we collaborate”), it’s either vacuous or needs challenging.
Collaboration’s coordination costs are systematically underestimated; autonomy with alignment beats consensus.
Growth lives in the core product
Network-effect businesses grow implicitly; the highest-leverage growth comes from deepening engagement of existing core users, not acquiring new monthly actives.
Daily active use (DAU/MAU) is the fundamental metric; moving the ratio by basis points compounds more than top-of-funnel acquisition.
Snap 2018: After the disastrous redesign flattened DAU, laser focus on Android performance for existing users sparked a multi-year growth renaissance.
Discord 2024: Despite Midjourney/crypto/AI hype, growth accelerated by obsessing over core gamers (voice, before/during/after play) — a segment assumed to have 100% penetration. Fixing edge-case friction (in-game voice, FaceTime switching) moved users from 10 to 13+ days/month.
Gamers are the most critical, high-IQ user base; delivering for them built unmatched product velocity and community trust.
“There’s always money in the banana stand” — return to core, perfect it.
When to bet on something genuinely new
Concentric circles: Identify core primitives (Snap: camera; Discord: voice) and find adjacent cohorts/niches where they apply.
User-led expansion: When users contort the product for unanticipated use cases (e.g., Discord for large communities), they’re voting with feet — reduce their friction.
Backcasting (blank-sheet): Envision the ideal future unconstrained by present tech (e.g., “best way to share a moment” → glasses/AR). Rare skill; most “creative” people just remix the present.
Risk: local optimum trap (Taleb’s Thanksgiving turkey) — everything improves until a 10x disruption zeroes you out. SpaceX guards against this by design.
SpaceX and sunsetting the Falcon 9
SpaceX is the greatest company in the world; its culture solves the innovator’s dilemma by design.
Recent example: sunsetting Falcon 9 (best launch system in history) to force transition to Starship — “burn the ships” commitment to the Mars mission.
Every employee embraces this; the culture makes self-cannibalization feel like the only logical path.
Elon publishes 10-year roadmaps (SpaceX, Tesla) and executes them; the roadmap is the moat — “copy us if you can” swagger backed by delivery.
Ambition as a learnable muscle
AI lowers build costs; ambition must rise to match. Status quo persists due to static friction; breaking it requires impulse (third derivative: acceleration of acceleration).
Great PMs elevate team ambition credibly: “What would 1000x look like? What blocks it?” — not empty exhortations to “go faster.”
This is a rare, trainable skill: combining long-term vision with daily urgency.
Long-term greedy: the ads philosophy
Ads teams face intense short-term revenue pressure; “long-term greedy” reframes restraint as selfishness: maximizing advertiser ROI today (even at short-term revenue cost) compounds via retention and budget expansion tomorrow.
Mathematical basis: long-term revenue is causally driven by today’s advertiser ROI; lowering effective price (denominator) via relevance/quality grows the pie.
The phrase gives ideological cover to resist quarterly pressure, especially as a public company.
The Core Product Value framework
Define the core value proposition as a simple phrase that doubles as a strategic filter and metric scaffold.
Snap: “Fastest way to share a moment with the people you care about” → decomposes to: fast (load time), share (camera-first), people you care about (best friends focus).
Discord: “Best way to talk and hang out with friends before, during, and after playing games” → talk, hang out, games, before/during/after.
Each component maps to causal retention metrics; onboarding presentations drilled this into every hire for years.
Requires both poetic clarity (for humans) and metric rigor (for data scientists/engineers) — neither alone suffices.
The three oxymorons of a great PM
Confidence bordering on arrogance + deep humility: Conviction to drive decisions, but credit goes to the designer who made the late change, the engineer who pulled the all-nighter. Confidence enables humility because PMs get disproportionate credit anyway.
Extreme process orientation + comfort with ambiguity (near pathological demand avoidance): Checklists for everything, but throws them out instantly when context shifts. Not “process is useless” — it’s holding both.
Long-term vision (6–24 months) + borderline-annoying daily bias for action: Sets the company’s wake-up pace; patient on strategy, impatient on execution.
Contrarian corner
Light spicy: The most important career decision is your life partner — no podcast or framework compounds like a great intellectual/emotional partner.
Medium spicy: Taste = knowing when to stop. People with taste get passionate, then exit before it becomes cringe. The unshipped prototypes you loved but held back are the proof.
Heavy spicy: Certain inalienable traits (gender, race, ethnicity, socioeconomic background) make people inherently better at building for specific consumer needs because consumer products hit Maslow’s hierarchy in ways shaped by lived experience. Not hireable or judgeable by cover — but real, uncomfortable, and societally unacknowledged.
Lightning round highlights
Books: How to Get Filthy Rich in Rising Asia, When Genius Failed (LTCM), Einstein’s Dreams (Alan Lightman) — the last for consumer PMs.
Game recommendation: Clair Obscur: Expedition 33 (indie, France) — Louvre-level artistry proving games are an undervalued art form.
AI product: Hex — first multiplayer AI workflow that markedly changed his work (used live in meetings with Discord co-founder Stan).
Life motto: Steinbeck, East of Eden: “Time has no signposts upon which to drape duration. From nothing to nothing is no time at all.” — do things, meet people, experience life to slow subjective time.
First PM role: Ustream (2009), product lead for the Shiba Inu puppy cam — 24/7 stream of puppies that became a cultural phenomenon; mentored by Josh Elman and Matt Schlicht (later sold MultiOn to Meta).
