Principle
Taste is restraint: a curator who hangs the whole collection has done nothing
Measure product taste by the quality of what was built and deliberately not shipped.
Sellis aimed to turn founders into the richest museum curators in the world, where the live product is a small slice of the raw collection chosen for a time and place. He extends it in contrarian corner: people with taste stop one step before everyone else and let a trend go when it turns cringe. With AI adding features by default, pushing back and cutting becomes a core PM skill.
Keep an explicit shelf of built-but-unshipped work and review it; if the shelf is empty, your curation is weak.
Principle
Be long-term greedy in ads: tomorrow revenue is caused by advertiser ROI today
Optimize ads for advertiser ROI even at the cost of short-term revenue, because that is what compounds.
Ads is hand-to-hand, revenue-goal-driven combat, so teams drift short-term. The phrase long-term greedy (we are not a charity; we give value today because they come back bigger tomorrow) gave ideological cover for long-term thinking at Snap, Discord and companies Sellis advised. In modern auctions price is an output, so the ROI lever is the value delivered, not discounting.
Put advertiser ROI next to revenue in every ads review, and require any revenue lever to show its ROI impact.
Principle
If no one can disagree with your team principles, they are not principles
A useful team tenet must have a credible opposite that a reasonable team could choose.
Sellis puts collaboration under this umbrella: it is the canonical unobjectionable tenet that hides real cost.
For each tenet, write the reasonable opposite; delete tenets whose opposite is absurd.
Principle
A product org that cannot absorb spiky talent is in a failure state
Design the org to deploy extreme specialists on the problems they are uniquely good at, and tolerate the rest.
Sellis managed Nikita Bier at Discord after acquiring his startup, likening him to Liam Neeson in Taken: a very special set of skills that is devastating when the daughter is kidnapped and useless for buying groceries. He could not get a one-on-one and messaged him on four apps. His model is Phil Jackson meshing talent that does not otherwise get along, and he treats spiky PMs like bands: stick with an artist through a dud album if you still believe in them.
Identify your spikiest operator and redesign their scope around their special skill rather than their weaknesses.
Principle
A PM must provide value over nobody: the baseline is operating the product with no PM
Judge every PM against a team that has no PM; many will not clear that bar.
Snap famously did not hire PMs for a long time. Sellis links it to a VORP-style mindset and describes building a system for evaluating PMs not just as individuals but as the person and the place they are at that time, to make sure the team is net positive.
Before opening a PM req, write down what the team would lose with no PM; if you cannot name it, do not hire.
Principle
The median PM is bad by the math: a truncated distribution plus exits at the top and clinging at the bottom
Expect the median PM to be net negative, because the profession structurally exports its best and retains its weakest.
During ZIRP, PM became the most lucrative tech career for non-technical people with a subjective definition, amplifying the adverse selection. Lenny adds the second-order effect: most people have never worked with a great PM, so the profession looks bad. Sellis notes great engineers rarely complain about PMs (they bend PMs to their will and refuse bad ones) and bad engineers do not either (they hide behind PM decisions); complaints come from the middle.
Raise the PM hiring bar and assume the default candidate is below average; staff PMs only where they beat the no-PM baseline.
Principle
Burn the ships: be willing to build what makes your best product irrelevant
An ambitious enough mission lets a company sunset a world-leading product to build its successor.
SpaceX telegraphing the sunset of Falcon 9, the best reusable launch system ever, for Starship is Sellis example of Cortez-level ship burning that avoids the innovator dilemma. He also admires Elon companies publishing 10-year plans (Tesla Roadster funds S funds 3) as roadmaps so ambitious they can be announced openly. He contrasts this with the Thanksgiving-turkey risk facing companies that only refine the core.
Name the product that would make yours irrelevant; if your mission would not justify building it, your mission is too small.
Principle
Ride your best people into the ground instead of rescuing the average
Leader time and decision load should flow to the strongest performers until they start to break.
Sellis says the instinct to save someone drowning is human nature, and he does the opposite: give it to the busiest person, ride your racehorses. The counterweight is always defending those people in public and never criticizing them in front of executives. He admits he lacks a good mechanism for helping people improve and lets strugglers fail on their own.
Check where your one-on-one hours actually go; shift them toward your top performers and hand them the next hardest decision.
Principle
Collaboration has an uninternalized coordination cost: you move as slow as the slowest node
Treat collaboration as a cost to be minimized, not a virtue to be maximized.
Sellis calls it a warning sign that no executive ever says stop collaborating. He tries to aggressively internalize the cost, and pairs that with the ideology-plus-decision-rights design so outcomes match what collaboration would have produced.
For each cross-team dependency on a project, ask whether shared ideology plus a single owner could replace the meeting.
Principle
Network-effect businesses want to grow implicitly: growth comes from deepening existing core users
Default growth strategy for network products: make the core use case perfect for current users rather than chase new segments.
Learned twice, at Snap (post-2018 Android performance) and Discord (2024 focus on intentional multiplayer gamers). Sellis encoded it as there is always money in the banana stand: return to your core users and make sure it is perfect for them.
Before a new-segment bet, quantify the growth from moving core users one more active day per month.
Principle
At high-growth companies, PMs should be making decisions the founders would have made
The PM role at founder-led companies is to absorb founder judgment and exercise it on decisions founders used to own.
Sellis worked directly with founders at Snap and Discord and describes the head of product as a vessel for their vision; the test of the org is whether decisions that required a founder 6-24 months ago now happen correctly without one.
List decisions your founder made a year ago; each should now have a named PM owner who makes it the way the founder would.
Principle
Consumer products must optimize for daily use: one monthly user is worth 1/30 of a daily user
Almost every consumer product should tie itself to a daily (or hourly) cycle and track DAU/MAU as its core retention measure.
DAU/MAU is just next-day retention: given you used it today, what is the probability you use it tomorrow. The most effective consumer products have extremely high DAU/MAU, and moving that ratio a few basis points often beats acquiring new MAU and hoping they retain.
Report DAU/MAU next to user growth in every consumer review, and fund frequency work before acquisition work.