Lore

product management

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

Peter Sellis (first PM at Snapchat, later Head of Product at Discord) argues that most product managers are structurally mediocre because the best ones self-select out into founder/executive roles, and that the fix — at both the individual and organizational level — is deliberately "terrorist-organization"-style design (shared ideology plus unambiguous decision rights), disproportionate investment in top performers, relentless focus on a company's core product value, and a "long-term greedy" approach to monetization.

Lenny's Podcast · 2026-09-20 · English

Key ideas

  1. The PM profession's quality distribution is structurally bad: a normally-distributed skill, once restricted to full-time practitioners, becomes power-law distributed so the median sits far below the mean, and this is worsened by ZIRP-era incentives that made PM lucrative for nontechnical people and gave them technical-ladder pay regardless of skill, while genuinely great PMs' skills transfer so well to founder/executive roles that they exit the profession entirely.

  2. Collaboration carries a hidden, unacknowledged coordination cost (a team moves at the speed of its slowest node); his remedy is designing teams like a "terrorist organization" — near-religious clarity of mission plus explicit single-threaded decision ownership — so the org gets collaboration-quality outcomes without paying the collaboration tax.

  3. He deliberately invests disproportionate time in his best people rather than rescuing struggling ones, piling responsibility on strong performers until they start making mistakes, arguing this runs counter to human instinct but not to business logic.

  4. Snap's perceived underperformance relative to the hype isn't due to poor execution but structural realities — a heavily regulated, hard-to-target under-18/21 audience (producing 'suspicious' CPM discontinuities at those age boundaries), a camera-first rather than feed-first product with no natural ad slot, and a fundamentally utilitarian messaging format that lacks great ad formats — while by relative/cohort metrics (share of under-21 messaging, maps, camera, astrology usage) Snap is actually a dominant multi-product portfolio.

  5. Durable growth comes from deepening the core product for its most engaged users (DAU/MAU, frequency-ladder growth) rather than chasing new markets, illustrated by Snap's post-2018-redesign recovery and Discord's 2024 refocus on "friends before/during/after games," with backcasting from founding philosophy offered as the rare legitimate path to genuine reinvention rather than incremental roadmap work.

  6. Sustainable ad monetization (at Snap, Discord, and by extension OpenAI) should be "long-term greedy": optimized for advertiser ROI and user trust today — even at short-term revenue cost — because that compounds into bigger long-term ad budgets and retention, with price treated as an auction output rather than a manual lever.

  7. Great PMs are defined by holding contradictory traits simultaneously — confidence bordering on arrogance paired with real humility, heavy process-orientation paired with comfort in ambiguity, and long-term vision paired with day-to-day urgency — and this "two wolves" pattern is what he screens for in hiring via concrete behavioral stories rather than self-description.

  8. His closing "spiciest" claim: building truly great consumer products for a specific cohort of humans requires PMs/designers who share certain unlearnable, unacquirable traits with that cohort — explicitly naming gender, race, ethnicity, and socioeconomic background — because lived/biological background shapes intuition for that cohort's needs (mapped to Maslow's hierarchy), while cautioning this isn't a basis for explicit hiring criteria or judging people by appearance.

  9. "Terrorist organization" team design — Combines near-religious ideological clarity of mission with explicit, single-threaded decision rights ("throat to choke") so teams move fast without over-collaborating. Apply: Articulate the team's mission as something close to dogma, then explicitly assign one directly-responsible owner per decision so people don't need to loop in the whole group.

  10. Cost-of-collaboration lens — Treats collaboration as carrying a real, usually unacknowledged coordination tax, since a team moves only as fast as its slowest node. Apply: Before adding stakeholders to a decision, weigh the coordination-speed cost against the marginal value each added person brings.

  11. "Give your best people more rope" resource allocation — A leadership philosophy of investing disproportionate attention and responsibility in top performers rather than shoring up struggling average performers. Apply: When allocating scarce leadership attention, default to expanding a strong performer's scope rather than coaching a struggling one back to baseline.

  12. "Operate without a PM" baseline test — A bar for justifying a PM role: the product should be operable without one, and the PM must clearly add value over nobody. Apply: Before staffing or keeping a PM on a team, ask what specifically breaks if the role were removed.

  13. Founder-vision-to-"syntax" distillation — Condensing a founder's vision into short, repeatable phrases that act as strategy shortcuts for the whole organization. Apply: Turn a founder's stated priorities into a handful of memorable, repeatable lines and use them consistently in reviews and roadmap discussions.

  14. Museum-curator/restraint framework — Taste is measured by what a leader deliberately leaves out of the product, not by everything they could ship. Apply: Audit a roadmap for how many good ideas were deliberately rejected, not just how many features shipped.

  15. Phil Jackson–style talent-meshing leadership — A management approach of blending very different, sometimes clashing high performers into one cohesive winning team rather than forcing conformity. Apply: When managing unconventional "spiky" talent, design roles and processes around their strengths instead of standardizing everyone to one working style.

  16. "Band/album" evaluation of spiky performers — Judging unconventional high performers by their body of work over time, like an artist's discography, rather than by any single weak project. Apply: When a strong-but-unconventional PM ships a weak project, review their track record before deciding whether to keep backing them.

  17. Power-law explanation of PM quality — A normally-distributed skill, once restricted to only the people who practice it full-time, becomes power-law distributed, so the median performer sits well below the average. Apply: When benchmarking a role's talent pool, expect the median in-role performer to underperform the mean and set hiring/promotion bars accordingly.

  18. "Person and place" PM evaluation system — An evaluation approach that assesses a PM as a fit between the individual and their specific context, not the individual in isolation. Apply: When a PM is underperforming, consider a role or team change before concluding the person themselves is the problem.

  19. Systems thinking via stocks, flows, and distributions — Reframes "systems thinking" away from a buzzword toward literally modeling a system mathematically, including stocks, flows, and the distributions behind key levers. Apply: Build an explicit spreadsheet model of a business system before claiming to "think in systems" about it.

  20. 5 Whys — A Toyota-associated root-cause technique of repeatedly asking "why" (traditionally five times) to trace a problem to its origin. Apply: When diagnosing a recurring product or process failure, keep asking "why" past the first surface-level answer until you hit a structural cause.

  21. Jay Forrester's system dynamics — A 1970s, manufacturing-originated formal discipline for modeling systems, recommended over pop-management "systems thinking" books. Apply: Use system-dynamics-style stock-and-flow modeling instead of informal "systems thinking" language when analyzing a business problem.

  22. Unshipped-prototype interview probe — A technique for assessing a candidate's taste by asking about things they built but deliberately chose never to ship, rather than things that simply tested poorly. Apply: In design or PM interviews, ask candidates to describe a product they built and held back, and why.

  23. VCG (Vickrey-Clarke-Groves) auction — The mechanism Sellis argues is the mathematically correct basis for ad auctions, including inside a chat interface like ChatGPT. Apply: When designing an ads marketplace, use a VCG-style mechanism rather than simple highest-bidder pricing.

  24. Unified organic+paid auction (Meta-style) — Factors an ad's organic value/relevance to the user into its bid so organic and paid content compete in one auction denominated in dollars. Apply: When building an ads product inside a content or chat surface, weight bids by predicted user relevance, not bid amount alone.

  25. Trust-as-organic-value in chat ads — In conversational (vs. feed) contexts, "organic value" should be reframed as how much user trust is being traded away for advertiser influence on an answer. Apply: When deciding how much an advertiser can influence a chat response, calibrate that influence against the trust cost to the user.

  26. Peter Principle — The idea that people are promoted until they reach their level of incompetence. Apply: Use it as a check when deciding whether a high performer's next promotion is actually the right fit, not just the natural next step.

  27. "No-one-could-disagree" test for team principles — A test that a stated team tenet is either vacuous or should be actively defended if literally nobody could object to it. Apply: Audit a team's values or principles doc and cut or rewrite any line that no one could plausibly disagree with.

  28. DAU/MAU as retention framework — Treats the DAU/MAU ratio as a direct measure of retention — the probability a user active today returns over a given period. Apply: Prioritize product work that nudges DAU/MAU over work aimed purely at growing MAU.

  29. "There's always money in the banana stand" core-focus heuristic — A heuristic for returning attention to perfecting the experience of existing core users before chasing new bets. Apply: Before greenlighting a new market or segment bet, check whether the core experience for current best users is already maximized.

  30. Frequency-ladder growth metric ("illness"), from Meta's growth playbook — Moves users up a frequency ladder (e.g., 10-of-30 days to 7-of-7 days) rather than treating retention as a binary retained/not-retained metric. Apply: Track and target incremental frequency-of-use milestones per user instead of a single retained/churned number.

  31. MBA-style 2x2 for adjacent bets — Maps a product's core strength (e.g., camera, voice) against other niches or cohorts to find defensible expansion opportunities. Apply: List the product's core strength on one axis and candidate new audiences or use cases on the other to spot where that strength already transfers.

  32. "Voting with their feet" / concentric-circles expansion signal — Treats users going through extra hoops to use a product in unintended ways as a genuine signal for expansion, radiating outward from the core. Apply: Watch for unintended or awkward usage patterns in product data and consider building direct support for that behavior.

  33. Backcasting vs. forecasting — Instead of projecting forward from current technology, imagine an ideal future state unconstrained by present tech, then work backward to what to build now. Apply: When planning a bold reinvention rather than incremental roadmap work, describe the unconstrained ideal outcome first and work backward to today's constraints.

  34. Taleb's "Thanksgiving turkey" black-swan risk — A metaphor for how steady, core-focused improvement can mask vulnerability to a sudden, discontinuous 10x-better competitor. Apply: When a strategy leans heavily on doubling down on the core, explicitly flag that it has no defense against a sudden discontinuous competitor.

  35. Self-obsoleting product strategy (applied "innovator's dilemma") — Deliberately building the product that will make your own best current product irrelevant before a competitor does, as illustrated by SpaceX reportedly sunsetting Falcon 9 for Starship. Apply: Evaluate whether your organization would greenlight a project explicitly designed to make your current flagship product obsolete.

  36. Long-range roadmap publishing as ambition-forcing device — Publicly committing to a roughly 10-year roadmap (and, in Tesla's case, even releasing patents) as a way to force and signal high ambition. Apply: Consider publishing a long-range roadmap publicly to create internal and external pressure toward higher ambition.

  37. Ambition-as-physics metaphor (position/vector/velocity/acceleration/impulse) — Argues real ambition requires not just movement (velocity) but acceleration, which requires active "impulse" — someone repeatedly pushing. Apply: When assessing a team's ambition, check whether the rate of progress is increasing, not just whether progress is happening, and identify who is actively pushing for that.

  38. PM-as-"ambition elevator" (1000x question) — Reframes the PM's job as repeatedly asking the team "what would a thousand-x version of this idea be?" to raise the ceiling of ambition. Apply: In planning or review sessions, explicitly ask what a 1000x bigger version of the current idea would look like.

  39. "Long-term greedy" ad-monetization philosophy — Optimizes for advertiser ROI and user trust today, even at short-term revenue cost, because it compounds into bigger long-term ad budgets and retention; treats price as an auction output, not a manual lever. Apply: When under pressure to hit short-term ad revenue targets, resist cutting price to artificially inflate ROI and instead invest in genuine ad relevance and targeting quality.

  40. Core Product Value (CPV) framework — Defines a product's core value both as a qualitative mission phrase (e.g., "the fastest way to share a moment with your best friends") and as matching measurable metrics, used to prioritize roadmap and onboard new hires. Apply: Write the product's core value as a short phrase plus a matching metric, and test every roadmap item against whether it strengthens that phrase.

  41. "Rule of threes" pattern for core-value phrasing — An observed pattern that effective core-product-value statements naturally decompose into roughly three components (e.g., fast / share / people you care about). Apply: When drafting a mission or core-value statement, aim for about three concrete, testable components rather than one abstract line.

  42. "Two wolves"/oxymoron PM interview technique — Screens for great PMs by testing for three paired contradictory traits — confidence with humility, process-orientation with comfort in ambiguity, long-term vision with day-to-day urgency. Apply: In PM interviews, ask for specific stories that reveal both sides of each contradictory trait pair rather than relying on candidates' self-assessment.

  43. Maslow's hierarchy of needs (applied to cohort intuition) — Used as the framework for understanding which specific human need a consumer product cohort is trying to satisfy, underlying the claim that lived background helps a PM intuit those needs. Apply: Map a product's value proposition to a specific level of Maslow's hierarchy for the target cohort to clarify what job the product is really doing for them.

Insights

Ad CPMs show a 'suspicious discontinuity' at ages 17→18 and 18→21, in the US and abroad, because of strict rules on tracking/targeting minors — a structural, not execution, cap on Snap's ad yield from its core young audience.

Despite Midjourney being built entirely on Discord and widely assumed to have driven a Discord growth wave, Sellis says you can't actually isolate its effect on Discord's DAU chart — the real growth driver was friends connecting before/after playing games.

Great engineers rarely complain about PMs — they either bend PMs to their will or simply refuse to work with bad ones — and the worst engineers also don't complain because they can hide behind PM decisions; it's the average engineers in the middle who complain.

Instagram reportedly once copied Snap features so aggressively that it copied ones that didn't even work, apparently assuming everything Snap touched was 'turning to gold' — it later shipped a product literally called 'Instagram+' mirroring Snapchat+.

Sellis frames one monthly active user as worth roughly 1/30th of one daily active user, the arithmetic basis for prioritizing frequency (DAU/MAU) over raw reach.

SpaceX is reportedly sunsetting Falcon 9 — its most proven, cost-effective reusable rocket — in favor of Starship; Sellis reads this as a rare real-world case of a company deliberately executing Christensen's 'innovator's dilemma' against itself before a competitor could.

Amazon's sponsored-search experience (roughly 16 ads before the first organic result) is cited as a case where a genuinely strong core product (selection, fast shipping) is quietly eroding user trust through ad load.

Nikita Beer, who told Sellis's team he doesn't believe 'product management is real,' went on to become Head of Product at X for about a year — cited as ironic validation of the claim that the best PM talent exits the discipline.

«If you see somebody drowning, you want to save them. I did the absolute opposite. If somebody is strong and crushing it, I'm just handing them more and more rope to hang themselves with.»

— 00:36

«A good terrorist organization essentially has two things. One is it has this ideological culture that everyone understands why they are doing what they're doing. And then it has a very very clear organizational structure of who can be trusted with which decision.»

— 00:14

«If a museum has all of their collection on the walls then the curator hasn't done anything.»

— 01:29

«You can't spell culture without cult.»

— 10:44

«For any power distribution, the median is going to be far below the average. Most people are on the left side of it.»

— 16:29

«I really don't hear great engineers complain about product managers that much because great engineers a) they know how to bend product managers to their will... and b) they just like don't put up with bad ones.»

— 20:19

«Snap opens the camera... in no part of that flow... is there like a logical place for an advertisement. It is just not a feed-based product... it is a creation-oriented, a visual creation-oriented product.»

— 23:34

«Imagine if you're like hey there's a holding company... a Jenzie Berkshire Hathaway that has the number one camera app, the number one messaging app, the number one maps app, the number one astrology app... that's an insane consumer product track record.»

— 26:00

«I think I saw Evan personally say no to better ideas than I've seen almost any other consumer startup launch.»

— 01:14

«the reason people use Discord is to connect with their friends while playing before, during, and after playing games»

— 54:12

«There's always money in the banana stand.»

— 56:31

«A thousand and one days in the life of a Thanksgiving turkey.»

— 61:53

«This is like Cortez level burn the ships upon arriving.»

— 63:46

«Just because we have the thing that is the best, we're completely willing to build the thing that will compete against ourselves and make the thing that we built irrelevant.»

— 64:31

«Long-term greedy.»

— 71:40

«This is why people choose Snapchat every single day in the face of tons of alternatives and a bunch of savage competitors.»

— 76:04

«I think that to truly create great consumer products for certain cohorts of humans... there are certain inalienable traits about those product managers that cannot be learned, cannot be acquired... by that I mean literally gender, race, ethnicity, socioeconomic background.»

— 83:47

«Ustream was essentially a failure because we should have focused on gaming.»

— 92:14

A dense, quotable practitioner interview that genuinely teaches a wide set of reusable product-management and growth frameworks grounded in specific Snap/Discord experience, but its final 'spicy take' on demographic traits is presented with far less rigor than the rest of the episode and is the most contested part of its reception.

96:53

↳ Lenny's Podcast · YouTube

Watch original