On Lenny’s Podcast, Snap’s first PM and former Discord Head of Product names a cost that every organisation carries and almost none of them price.
Editorial note: This article summarises a podcast conversation on Lenny’s Podcast. Every claim about collaboration costs and distribution is Peter Sellis’s own. This is not anti-collaboration advice — his prescription is to price the cost of collaboration, not to prohibit it. The distribution terminology he uses is informal rather than technical; both that fact and the conclusion it supports are addressed below.
Peter Sellis — Snap’s first product manager and, later, Head of Product at Discord — appeared on Lenny’s Podcast and made an observation that is easy to agree with in the abstract and hard to act on in practice. There is, he says, “a hidden cost to collaborating that no one really internalizes extremely well.” His evidence for the claim is structural, and it comes from an absence rather than a data point: “you’ll never hear really any executive at any company be like, stop collaborating. And to me that’s kind of like a warning sign.”
The cost he names is not vague. Collaboration carries coordination costs, and those costs, in his framing, “essentially assure that you’re moving as slow as the slowest node in the system.” That is the mechanism. When a decision requires several parties to align, its latency is determined not by the average participant but by the slowest one. Adding a thoughtful, well-intentioned reviewer to a process cannot make it faster. It can only leave the speed unchanged or reduce it.
His second argument concerns how skill is distributed inside professional populations. He reasons that because skill is roughly normally distributed across people at large, those who actually practise a craft professionally sit in the right-hand tail of that distribution. When you consider only that selected group, the shape becomes skewed — and in a skewed distribution, the median sits below the mean. The practical consequence is that reasoning about the “average” practitioner is unreliable, because a small number of extreme performers pull the average upward while most individuals sit below it.
The key insight: Every contested activity inside a company has an institutional counterparty — spending has finance, hiring has a budget, risk has compliance. Collaboration has none. Its benefits are advocated continuously by everyone; its costs are borne diffusely by everyone and owned by nobody. The ledger ends up with only one side filled in.
The Structural Read
The absence Sellis identifies is worth examining as an organisational design problem rather than a management platitude. Almost every material cost inside a company eventually acquires a monitor — someone whose job is to push back, to measure, to say the number is too high. Collaboration has never acquired one. No meeting exists to review whether there is too much of it. Nobody is evaluated on having reduced it. The activity is self-perpetuating precisely because the costs it generates are invisible in any individual ledger while the benefits are visible and attributable.
This is the same structural shape as a denominator that nobody produces because nobody has a reason to. In financial contexts the pattern shows up as a metric with a numerator everyone tracks and a denominator nobody maintains. Here it is organisational rather than monetary, but the architecture is identical: one-sided accounting produces systematic mispricing, and systematically mispriced resources tend to be overconsumed.
The slowest-node claim repays being taken literally. Each additional participant in a decision does not add a single link to the chain — it adds relationships, and the number of relationships grows faster than the headcount that produces them. The coordination cost therefore scales in a way that headcount alone does not reveal. Two people describing the same failure from different industries — one as an organisational design principle, one as an existential worry about becoming process-bound at scale — are converging on the same underlying mechanism.
Peter Sellis — Lenny’s Podcast
“…a hidden cost to collaborating that no one really internalizes extremely well… you’ll never hear really any executive at any company be like, stop collaborating. And to me that’s kind of like a warning sign.”
Builder-PM Framework
Pricing vs. Prohibiting
Sellis’s prescription is narrower than a headline would suggest. He does not propose less collaboration. He says he tries to internalize its cost — which is a different intervention entirely. Internalising a cost means putting it on somebody’s ledger so it trades off against something else. The meeting is not forbidden; it simply has to be worth what it costs, and somebody has to feel that cost. Pricing changes behaviour without requiring anyone to refuse permission, which is why it tends to survive contact with an organisation better than a rule does.
On the Distribution Argument
Sellis frames the skill argument using the language of a power distribution. That term is doing informal rather than technical work — truncating a normal distribution does not literally produce a power distribution. But the conclusion survives the imprecision. Selection creates skew, and in a skewed population the median and the mean come apart in a meaningful way. The practical implication holds: inside any selected professional population, the average practitioner is not the typical one, because a small number of extreme performers pull the mean upward while most individuals sit below it. This quietly affects hiring bars, performance calibration, and any benchmark expressed as an average.
Three Implications
IMPLICATION 1 — THE UNMONITORED COST
Any organisational cost that lacks an institutional monitor will be systematically underpriced. Collaboration sits in that category today in most companies. The absence of someone whose job is to reduce it is not a sign that it is optimally sized — it is a sign that its costs appear on no one’s performance review.
IMPLICATION 2 — LATENCY IS SET BY THE TAIL
Decision speed in any multi-party process is bounded by the slowest participant, not the average one. This means that the cost of adding a reviewer is not zero even when that reviewer is fast — it is the probability that they become the constraint multiplied by the delay they would create. That cost is almost never computed before the reviewer is added.
IMPLICATION 3 — AVERAGES MISLEAD IN SELECTED POPULATIONS
The distribution argument extends beyond productmanagement. Wherever a population has been selected — by hiring, by credentialing, by market survival — the distribution of outcomes is skewed, and the mean is not a reliable proxy for the typical case. Benchmarks built on averages in these populations quietly set bars that most practitioners cannot meet, and calibration built on them will be systematically off.
Business Engineer Framework
Builder-PM: Pricing Decisions, Not Just Making Them
The Builder-PM framework reframes product decision-making around cost internalization — treating every coordination choice as a trade-off with a real price tag, not a default to consensus. Sellis’s argument is a clean instance of this logic applied to collaboration itself: the question is not whether to collaborate but whether the collaboration is priced.
Sellis’s observation is not a case against collaboration — it is a case against carrying any cost that nobody is responsible for measuring. The organisations that move fastest are not the ones that collaborate least; they are the ones that have put the cost of collaboration on someone’s ledger, so every meeting, every sign-off, every alignment loop has to justify itself against what it actually takes to run. The denominator exists. The question is whether anyone is required to produce it.
Get the structural read, every week. Business Engineer covers organisational design, AI strategy, and the business models shaping tech — with the frameworks to make sense of them.
Source: Lenny’s Podcast (YouTube). All claims about collaboration costs, the slowest-node mechanism, and the distribution of professional skill are Peter Sellis’s, drawn from his appearance on the show. No Snap or Discord internal data, headcount, or performance figure is used or implied. The distribution terminology is informal; both that qualification and the conclusion it supports are noted in the body of this article.
91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
This summarises a podcast conversation, and every claim above is Peter Sellis’s.It is not anti-collaboration advice — he describes internalising the cost of collaboration rather than rejecting the practice, and nothing above argues that teams should collaborate less. His “power distribution” terminology is informal: truncating a normal distribution does not strictly produce a power distribution. The conclusion he draws from it nonetheless holds, because selection creates skew and in a skewed population the median and the mean come apart. No Snap or Discord internal practice, headcount, performance or outcome figure appears above, no statistic about product managers, and no formula. Nothing above claims his approach worked, and nothing is predicted or recommended.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
