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

Peter Sellis was the first product manager at Snapchat, spent seven years there, and later ran product at Discord. The episode is framed around the fact that he is now between roles: no comms team, no PR team, and, as he points out, nothing to promote. That framing matters for how the rest of the conversation should be read, because the candor is partly a function of having no current employer to answer to.

I'm no longer beholden to any corporate comms team right now

The limit of that freedom is worth naming early. When the discussion turns to Snap, Sellis prefaces his answer by calling it "a little bit of my life's work" and conceding he may carry a chip on his shoulder, then later warns that some of his information is dated and that he has not really looked at the product recently. His Snap analysis is therefore a retrospective from a former insider, not a current measurement.

Lenny Rachitsky's introduction sets out the four arguments the episode actually delivers: teams organized like terrorist organizations, what it was like to manage Nikita Bier, why most PMs are bad, and why Snapchat struggled to build a massive business despite roughly a billion monthly active users — with the teaser that the reason is not the obvious one.

The metaphor arrives with disclaimers attached. Sellis says he should workshop the phrasing, that it is an example of something better said behind closed doors, and that the Chief People Officer at Discord was "simultaneously horrified and intrigued" by it. He also explicitly sets violence aside, and denies starting from the metaphor: he enumerated what he wanted from an organization first, then noticed that organizations already operating that way happened to be terrorist organizations. Lenny's follow-up about building a cult gets a partial yes — cults are effective at getting things done — but Sellis draws a line between cults organized around a person and organizations organized around an ideology.

The first of two beliefs behind the model is that collaboration has a cost nobody prices properly. His evidence is a piece of organizational silence: no executive ever tells a team to stop collaborating, and he treats that absence as a warning sign rather than reassurance. Coordination, in his account, does not just slow a decision down; it caps the whole system's speed at the speed of its slowest participant.

you'll never hear really any executive at any company be like, "Stop collaborating."

The second belief supplies the replacement for coordination: at a high-growth company, product managers should be making the calls the founders would have made six months ago, or a year or two ago. That is a demanding test, because it asks a PM to hold founder-level judgment without founder-level context. The organizational answer is to make the strategy so legible that judgment can be delegated.

The two ingredients Sellis says a good terrorist organization has, and which he wants in a product org: an ideological culture in which everyone understands why they are doing what they are doing, and a very clear structure of who can be trusted with which decision.

The mechanism is linguistic. Because a head of product at a founder-led company is, in his phrase, a vessel for the founder's vision, he tries to absorb that vision and compress it into small phrases that can be repeated across the team — shortcuts, or synecdoches, standing in for the whole strategy. A shared shorthand plus clear decision rights is supposed to produce the alignment that heavy collaboration would have produced, at a fraction of the coordination cost.

Lenny restates the keepable parts as autonomy for people further down the chain, in case leaders are unreachable, plus a clear mission that makes decisions fast. Sellis accepts that reading and adds the cult distinction: cults are about the person leading them, whereas the organizations he is borrowing from are about a higher power, a religion, an ideology. He then jokes that reading about cults up to the moment just before they become notorious would teach a lot.

You can't spell culture without cult.

The honest limitation is that this is a metaphor doing work that evidence would normally do. Sellis offers no data, only a reaction from one HR leader and his own experience; he admits the branding is what gets attention, and volunteers twice that the framing needs workshopping. What survives the caveats is narrower and more defensible than the phrase suggests: decision rights should be explicit, and coordination should be treated as expensive rather than free.

Nikita Bier came to Discord through an acquisition and stayed more than a year, bringing engineers with him. Sellis immediately qualifies the premise of the question: managing Nikita is a loose use of the word. The comparison he reaches for is Liam Neeson in Taken — a very particular set of skills, hard to come by and decisive in the right situation, and close to useless when the task is buying groceries. The implication is that the value of an extreme operator depends entirely on whether the problem matches the skill.

you don't manage Nikita, you only seek to control him or direct him

The org-design claim behind the anecdote is stronger than the anecdote: a product organization that cannot absorb spiky talent is, in his words, in a failure state. The supporting story is deliberately unglamorous. Bier reported up into him, and Sellis still could not get a one-on-one, eventually having to reach him across four different messaging apps, with a joke about booking him on Intro to force the meeting.

I view it as a failure state if you have product orgs that can't essentially handle really spiky talent.

His management model is Phil Jackson, chosen for the ability to mesh very different talents that do not otherwise get along, for having done it as both player and coach in two organizations, and for drawing on both Eastern and Western philosophies. The point is not harmony but synthesis: winning teams assembled out of people who would not naturally cohere.

The band analogy explains the tolerance policy. You can love a band and still watch it release a dud album; the real question is what you do as a fan — stay, keep listening, retreat to the old records, or leave. Applied to product managers, a miss is a bad album, not a verdict on the artist. It is a humane frame, and it is also the loosest part of his argument: it tells you to extend patience but gives no threshold for when patience becomes sunk cost.

The claim that most product managers are bad, and often net negative, is one Sellis says was contrarian during the zero-interest-rate era and is now close to mainstream. He notes that the chief product officer at Whatnot has argued something similar, and that several of Whatnot's most senior PMs came from his own team at Snap — which he reads as corroboration, while acknowledging the awkwardness of well-paid former PMs now declaring that PMs are bad.

His explanation is statistical rather than moral. If PM skill is normally distributed across the population, the people who actually become PMs sit in the right tail; truncating the distribution there turns it into a power distribution, and in a power distribution the median falls well below the mean. Most practitioners, on this account, are on the left side of a curve that was already selected for competence.