More Collaboration Makes You Slower: Counterintuitive Management from the Product Chiefs at Snap and Discord
Collaboration has hidden costs nobody has internalized, and coordination costs guarantee you can only move as fast as the slowest node in the system. So Peter Sellis deliberately and aggressively internalizes the cost of collaboration, rather than defaulting to more collaboration being better.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The more you collaborate, the slower the system gets
Peter says you will never hear an executive say ‘stop collaborating’, and to him that is precisely a warning sign. Collaboration brings enormous coordination costs, and coordination costs guarantee you can only move as fast as the slowest node in the system. So he deliberately and aggressively internalizes the cost of collaboration, rather than defaulting to more collaboration being better. This is also where his ‘design teams like a terrorist organization’ comes from: one ideological culture that makes everyone understand why they are doing this thing, plus a very clear organizational structure that specifies who can be trusted to make which decision. With those two things, you don't need to over-collaborate and you still get results that look ‘as if you had collaborated’.
— Peter SellisThe median PM gap is a mathematical structure problem
Peter offers a distributional explanation: product management ability is normally distributed in the population, and the people who actually become PMs are cut from the right tail; once you cut it, it becomes a power-law distribution, and the median of a power-law distribution is far below the mean. Layer on two more mechanisms: in the ZERP era it became the most lucrative career in tech for people without a technical background, and the definition is subjective, so the people who get in have an extremely strong incentive to stay; while the genuinely strong ones have skills that fit almost perfectly into other executive roles or into founding their own companies, so ‘great product managers are exiting the system’. Squeezed from both ends, the median is naturally bad. He also points out that the people actually complaining about PMs are the middle cohort of engineers — great engineers know how to bend a PM into the shape they want, and especially bad engineers can hide behind the PM's decisions.
— Peter SellisThere are three mathematical reasons Snap's ads can't scale
First, users are extremely young, and between 17 and 18, and between 18 and 21, there is a ‘suspicious discontinuous jump’ in ad CPMs in and outside the US; the rules for tracking, measuring and targeting under-17s are extremely strict, and what is hard to measure is hard to build performance products around. Second, almost all non-search display advertising runs on a screen that exists to show ads, whereas Snap opens to a camera — in the core flow of taking a snap and sending it to a friend, there is no logical place to put an ad; it is a creation-oriented product, not a consumption-oriented one. Third, it is fundamentally a messaging app, and there is currently no good messaging ad product; Asian chat apps get cited over and over, but break it down by dollars spent per unit of time and none of them does especially well.
— Peter SellisTaste is the nerve to leave good things on the shelf
Peter says one thing he learned at Snap: he watched Evan say no to better ideas than almost anything any other consumer startup has ever shipped. He describes the product leader's goal as turning the founder into the world's richest museum curator — what is currently on display in the product is only a small part of the whole collection, the pieces the curator judges suitable for this time and place. He offers a way to probe taste in interviews: ask the person about something they made but never shipped — not something killed because the data tested badly, but something they held as a prototype and then judged ‘the world isn't ready’ or ‘this thing isn't ready’. If a museum hung its entire collection on the walls, the curator would have done nothing.
— Peter SellisAdvertising is the internet's product-market fit
Peter comes down squarely on Eric Seufert's side: advertising is the internet's product-market fit, and subscription consumer products are overrated. He judges OpenAI to have ‘last mover advantage’ in ads — Meta already solved the modern ad auction, this ‘solved problem’, between 2010 and 2020, and OpenAI is just running through each component quickly. Mechanically he names the modified VCG (Vickrey-Clarke-Groves) auction: it isn't the highest bidder who wins, but rather the advertiser's natural value for that user is folded into the bid, so whoever bids high has to pay more to reach people who don't want to be reached. This mechanism lets organic content and ad content go into the same auction, with dollars as the common denominator.
— Peter SellisIn chat, the denominator for ads is trust, not relevance
Peter replaces ‘natural value’ in VCG with ‘trust’ in the chat context. A user asks ‘which microphone should I buy’, and the product has to decide whether to give an objective answer, a set of answers, or an advertiser-influenced answer — and the degree of influence should equal how much trust you are willing to trade away. He uses Amazon search as an example: search for something and you scroll past 16 ads before the first real result, which subtly erodes users' trust in the system. So OpenAI's real deciding factor isn't whether to do ads, but how to mathematically maximize long-term trust inside the auction mechanism — trust maximization is probably correlated with user retention, and may even be causal.
— Peter SellisGrowth comes from the core, not from new markets
Snap and Discord both confirmed it twice: network-effect businesses want to grow on their own, and the most effective move is to serve the people already using the product, or wanting to use it for its actual job, better — faster, more relevant. He explains it with eighth-grade math: it takes 30 monthly actives to equal one daily active, so consumer products should optimize the DAU/MAU ratio, and moving that ratio a few basis points is far more worthwhile than pulling in a pile of new monthlies and gambling on retention. Snap's lesson came from the disastrous 2018 redesign, the only quarter in which Snap's overall DAU was flat quarter over quarter; after that the team ground away at Android performance, producing a growth revival that lasted several years. Discord's version is more counterintuitive: at the time Midjourney was built on Discord and everyone outside thought this was Discord's big opportunity, but Peter says if you pull up the Discord usage curve, you simply cannot pick out the Midjourney segment.
— Peter SellisSpaceX is burning its own ships by hand
Peter considers SpaceX the greatest company right now, and the most worth learning from is how it fights the future: Falcon 9 is the world's best reusable launch system, other countries would trade GDP for it, and yet SpaceX announced it will stop selling Falcon 9 seats in a few years and pivot to Starship. He calls it a Cortez-level burning of the ships, a way to avoid what Christensen called the innovator's dilemma — because even holding the best thing in hand, you are still willing to build the thing that makes your current product irrelevant. He also mentions that SpaceX and Tesla publish ten-year plans openly on their websites, and Tesla even opens its patents; that confidence of ‘I'll tell you what to do, and if you think you can do it better, come copy it’ is what he aspires to as a product leader.
— Peter SellisLong-term greedy: using ideology as cover for long-termism
The phrase Peter repeatedly used on the ad teams at Snap and Discord is long-term greedy. Behind it is math: tomorrow's ad revenue is causally linked to the ROI you create for advertisers today, so the focus should be on advertiser ROI. The hard part is that the easiest way to improve ROI is to lower the denominator — just cut prices — and in a modern auction price is an output, not an input. What he really means is this: in any career there will be moments when long-termism is extremely painful, like a public company missing its numbers and possibly having to lay people off, and at those moments you need a set of ideology as ‘cover’ that gives you room to make long-term decisions — we're not doing charity, we're giving up margin today because we know they'll come back tomorrow with a bigger budget.
— Peter SellisCore product value takes seven years of repeating it weekly
Peter says Snap's core product value was not a slogan on the wall, but an onboarding presentation he personally gave to every new employee, every week or every other week, for seven straight years. His logic: with only metrics and no story, data scientists and engineers won't act on it every day; with only a story and no metrics, nobody knows what to optimize. So you must have both, and repeat them — part of what he calls the ‘cult’: this is how we deliver core product value, this is why people still choose Snapchat every day in the face of a mass of substitutes and a pack of ferocious competitors.
— Peter SellisA great PM is three pairs of wolves fighting each other
Peter thinks great PM is itself an oxymoron, specifically three contradictions: first, extremely self-confident to the point of arrogance, alongside genuinely serious humility, able to accurately give credit to the designer who changed the design at the last second, the engineer who pulled an all-nighter for a partner event; second, extremely organized and respectful of process, while also extremely tolerant of ambiguity, even pathologically averse to authority — ‘I have a checklist for everything, but I can also throw it out and redo it at any moment’; third, long-term thinking coexisting with an almost annoying bias to action, a good PM can set the cadence for the whole company, ‘he wakes up in the morning and the team is awake’.
— Peter SellisSome consumer product ability can't be learned
Peter's spiciest take: to build truly great consumer products for certain specific groups, there are some inalienable traits on the PMs and designers working on those products that cannot be learned or acquired after the fact — he explicitly lists gender, race, ethnicity, socioeconomic background. He stresses this is not the simple correspondence of ‘building a product for young Black men means hiring young Black men’, but rather that most consumer products are essentially playing to certain parts of Maslow's hierarchy, and people with certain upbringings and physiological conditions understand those needs better. He also admits this take is in many cases not even executable, which is exactly why it's spicy.
— Peter SellisIn their own words · checked verbatim
there's actually like a a hidden cost to collaborating that no one really internalizes uh extremely well
Peter Sellis6:07
the great product managers are exiting the system. Um they're becoming founders, they're becoming executives etc.
Peter Sellis17:13
If if a museum has all of their collection on the walls, then the curator hasn't done anything.
Peter Sellis36:26
And the reason that I say that this is good for chat is not necessarily that organic value is what's important here, like how much you like an ad or not in a content feed, but in chat it probably feels like this could be replaced or augmented with essentially trust.
Peter Sellis42:30
they're sunsetting the Falcon 9 and this is like Cortez level burn the ships upon arriving
Peter Sellis1:03:46
you're going to encounter times in your career where thinking long term is actually like really really [ __ ] tough
Peter Sellis1:12:50
I have a checklist for everything, but I'm also comfortable if I have to throw it out and redo it all the time
Peter Sellis1:17:57
Figures
| Snapchat penetration among under-21s in most Western countries | 70% to 100% | 25:19 |
| Number of ads before Amazon search results | 16 | 43:32 |
| Snap's DAU performance after the 2018 redesign | The only quarter-over-quarter flat quarter | 52:36 |
| Discord's target for increasing days of user usage | From 10 days out of 30 to 11, 12, 13 days | 55:38 |
| Duration of the Snap core product value onboarding talk | 7 years, weekly or every other week, for every new employee | 1:15:56 |
| Number of good-PM contradiction pairs Peter lists | 3 pairs | 1:16:57 |
| Number of simultaneous 24/7 livestream cameras during the Ustream era | 3 | 1:33:15 |
Glossary
- modified VCG
- An ad auction mechanism that folds the advertiser's natural value for the user into the bid, rather than simply letting the highest bidder win.
- long-term greedy
- Making present decisions by the logic of maximizing long-term payoff, using ideology as cover for long-termism when necessary.
- DAU/MAU
- The core metric for user stickiness; consumer products get more from optimizing it than from pulling in new monthlies.
- innovator's dilemma
- Christensen's theory: incumbents miss disruptive innovation because their existing business is too good.
How to listen
Product leaders, startup founders, engineering and design managers currently wrestling with collaboration and decision-making efficiency.
The opening introduction and the Ustream reminiscences can be skipped; the ad mechanics and management principles in the middle and back half are the most valuable.