意识形态、克制和核心：Peter Sellis 怎样带能打胜仗的产品团队

Sellis says he designs product teams like a terrorist organization, a provocative label he admits he should workshop. It rests on two beliefs: collaboration carries hidden coordination costs that leaders rarely internalize, and product managers at fast-growing companies should be making the decisions a founder would have made a year or two earlier.

A good terrorist organization, in his telling, has an ideological culture in which everyone understands why they are doing what they are doing, plus a very clear structure of who can be trusted with which decision. Combined, those two things let a team reach the result it would have reached by collaborating, without paying the collaboration tax.

In practice, he says, the head of product at a founder-led company acts as a vessel for the founder's vision. At both Snap and Discord he tried to absorb that vision and boil it down into short, repeatable phrases that work as shortcuts for the whole strategy. He distinguishes this from a cult: cults revolve around a person, while his model is built around a higher-order mission.

Sellis managed Nikita Bier at Discord after the company acquired his startup. He jokes that you do not manage Nikita, you only seek to direct him, comparing him to Liam Neeson in the Taken films: a very special set of skills that is invaluable in the right situation and awkward when you just need someone to buy groceries.

He sees it as a failure state when a product organization cannot handle spiky talent, people who are not perfect nine-to-five employees. His model for mixing very different personalities is Phil Jackson, who got players who did not get along to win together.

Sellis likens consumer product leaders to bands releasing albums. Even a great band sometimes puts out a dud, and if you still believe in them as artists, you ride through the weak record and stay a fan.

Sellis offers what he calls a mathematical explanation for why most PMs underwhelm. If product skill is normally distributed, the people who actually become PMs sit at the right-hand tail, and lopping that off creates a power-law-like distribution in which the median falls far below the average.

Two forces make it worse. During the zero-interest-rate era, product management became the most lucrative route in tech for non-technical people, so weak PMs had strong incentives to stay. Meanwhile the best PMs leave, because the skills transfer perfectly to founding a company or running a business.

Sellis thinks the profession is improving as the true opportunity cost of an extra hire becomes clearer. He notes that great engineers rarely complain about PMs, because they either bend them to their will or never get assigned bad ones, and terrible engineers can hide behind PM decisions. The complaints come from the middle.

Asked why Snap never became as large as many expected, Sellis lists three obstacles to ad monetization. The audience skews very young, and there is a sharp discontinuity in ad prices around age 17 to 18 because rules for tracking and targeting minors make return on ad spend much harder to measure and prove.

Second, the app opens to the camera rather than a feed, and the core flow of taking and sending a snap has no logical place for an ad. He calls his failure to crack camera monetization one of his biggest. Third, at its heart Snapchat is a messaging utility, and no messaging app has yet become a great ads business per unit of time spent.

Sellis also defends the product record: Snap is the leading messaging, camera and maps app among teenagers in most Western countries, and it reached about four billion dollars in ad revenue faster than Twitter, Reddit and Pinterest despite starting later. He adds that Snapchat Plus has done well, but subscriptions take years to compound, and he wonders what would have happened had it launched four years earlier.

The pair joke about how often the phrase systems thinking has come up on the podcast. Sellis says its most underrated element is a mathematical background, the ability to describe a system in terms of stocks and flows and to say whether a quantity is bimodal or normally distributed. Otherwise, he suggests, it often amounts to asking why five times.

His practical advice is to read something like Jay Forrester's system dynamics and then build a spreadsheet model. List the variables, what comes in and out, the levers and the distributions behind them. Even if the spreadsheet goes nowhere, he says, the exercise can be very clarifying.

Sellis recalls that Instagram copied Snapchat so aggressively that it copied some of its worst features, assuming everything Snap touched turned to gold. He concedes that Instagram's version of Stories is now arguably better, thanks to a strong design team and Adam Mosseri's taste.

Sellis says Evan Spiegel has been right about almost everything, often early, including AR, avatars and on-device machine learning. On Spiegel's view that software is no longer a moat, he thinks Spiegel is probably right but that investors may not give him the space to prove it.

What Sellis learned most at Snap was restraint: he says he saw Spiegel say no to better ideas than almost any other consumer founder. He wants founders to behave like museum curators, with the product showing only a small slice of a larger collection. One way to probe someone's taste in an interview, he suggests, is to ask what they built but never shipped.

Lenny adds that AI tools tend to add features and rarely remove them, so cutting becomes a human skill. Sellis agrees it is hard for AI to replicate, and expects cheaper creation to put enormous pressure on making things that deserve human attention. Pushing back on what AI adds could become a new core skill for PMs.

Sellis still believes ads are the product-market fit of the consumer internet, and he sees OpenAI speed-running the entire modern ads infrastructure, which he calls a last-mover advantage built on a decade of Meta's work. He thinks a modified VCG-style auction is the right mechanism, in which an advertiser pays more to reach people for whom the ad has low organic value.

In a chat setting, he suggests, the organic-value term could be replaced or augmented with trust. When someone asks which microphone to buy, the system must decide how much to be influenced by advertisers, and that influence should reflect how much trust is being traded away. He points to Amazon search, with many sponsored results before the first organic one, as a cautionary example.

He argues that maximizing trust will likely correlate with retention and advertiser value, and that ad formats in chat, possibly voice, are an open design problem he calls one of the coolest jobs in product. He was a little disappointed that OpenAI jumped straight to an ad-free top tier, thinking it removes its most valuable users early.

Sellis describes a counterintuitive habit: he spends time on his strongest people rather than trying to lift the average. If someone is crushing it, he keeps handing them more decisions until he sees them make bad ones, on the theory that confidence and good results reinforce each other.

He pairs this with a firm rule that he always defends his team in public and never criticizes them in front of executives. He admits he has no great mechanism for helping people improve and tends to let people fail on their own. He cautions that not paying attention to someone should not be mistaken for letting them struggle, since often it means he trusts them to execute.

On collaboration, he repeats that it is often a poor use of time. If the principles you state are things no one could disagree with, he says, they are probably not real principles and may need to be challenged.

Sellis says he learned twice that growth lives in the core. Network-effect products already want to grow, so the best lever is helping people who already want the product use it more easily, quickly and relevantly. He argues almost every consumer product should optimize for daily active use, since one daily user equals about thirty monthly ones.

At Snap, a disastrous 2018 redesign was the only time overall daily users flattened quarter over quarter. The team then focused on performance, especially on Android, for people already using the app, and that kicked off years of growth. Daily-to-monthly ratio, he notes, is just a form of retention.

At Discord in 2024, the team ignored the hype around Midjourney, which would not be visible in usage charts, and focused on being the best place for friends to play intentional multiplayer games together. Even though everyone assumed those gamers already used Discord, improving performance and the moments before and after play produced some of the fastest growth since the pandemic. The company was also more trusted because it fixed bugs that gamers on Reddit reported, however small.

For consumer products, Sellis sees two ways to look beyond the core. One is incremental: identify what people love about your product, such as Snap's camera or Discord's voice, and look at other cohorts who might want it, or notice where users are going through hoops to use the product in unintended ways. For Discord that meant large communities rather than twenty-person friend groups.

The more radical approach is backcasting, imagining the best way to share a moment without being limited by phones, which can lead toward glasses. He says few people can do this blank-sheet thinking; most remix the present. Lenny notes that Airbnb never found a growth driver beyond improving its core experience.

Sellis holds up SpaceX as the model for guarding against disruption: it is sunsetting the Falcon 9, the best launch system in the world, in favor of Starship. He calls that a Cortez-level burning of the ships and a way of avoiding the innovator's dilemma. He adds that both SpaceX and Tesla effectively publish their roadmaps, which a product leader can aspire to.

Lenny connects this to ambition rising as AI makes building easy. Sellis says the status quo has its own static friction, and a team needs not just position or velocity but an impulse that makes tomorrow look different from today.

Among his so-called Sellisms, Sellis highlights long-term greedy for ads. Ads are a cutthroat, revenue-driven area where short-termism is easy, but future revenue is causally tied to the ROI advertisers get today. He stresses that the phrase was an ideological cover to think long term in hard moments such as missed numbers or possible layoffs, and that it was not charity.

The second is core product value, which he defined at both Snap and Discord in words and in metrics. At Snap it was the fastest way to share a moment with your best friends, and each term could be broken down into measurable parts. At Discord it was the best way to talk and hang out with friends before, during and after games.

Sellis says the phrase is meaningless without metrics and the metrics are meaningless without the story. He gave an onboarding presentation on core product value to every new Snap hire for seven years, which he calls part of the cult.

Sellis describes three oxymorons of great product managers. The first is confidence bordering on arrogance, balanced by real humility, including credit for the designer who made a last-second change or the engineer who stayed up all night. The second is being highly organized and process-driven while being comfortable with ambiguity and ready to throw the checklist away.

The third is long-term thinking paired with a borderline annoying bias for action, so that the team wakes up when the PM does. Lenny compares it to a superpower and its shadow. In the contrarian corner, Sellis offers a light-spicy take that the most important career decision a PM makes is who their partner is. His medium-spicy take is that people with taste know when to stop one step before everyone else.

In the lightning round he recommends *How to Get Filthy Rich in Rising Asia*, *When Genius Failed* and *Einstein's Dreams*, along with the game Clair Obscur: Expedition 33. He names Hex as the AI product that changed his workflow most, because it supported genuinely multiplayer use in meetings. He closes with a Steinbeck line about time, that from nothing to nothing is no time at all, as his reason to do things early.