32,000 Applied. Whatnot Hired One PM.

Source: Lenny's Podcast | Published: 2026-08-02T12:30:29Z

Whatnot CPO Tom Verrilli says most of the 32,000 applicants excelled at office politics, not product judgment — and mentioning "driving alignment" in an interview is the fastest path to rejection.

Tom Verrilli said this himself on a podcast last week. He's Whatnot's CPO, and before that spent seven years as Chief Product Officer at Twitch. He wasn't sharing the number to brag about the selection rate. He wanted to make a point: a huge proportion of people carrying the PM title simply don't have the capabilities he believes a product manager should have.

Whatnot is America's fastest-growing live commerce platform. Tom has been there since the early days, and he holds a position on product management that sounds strange to outsiders —

"We regret that the product manager role exists."

How the PM role got bloated

Tom traces it back to history. Early internet companies had no PMs at all — founders or CEOs talked directly to engineers and designers, explained what needed to be built, and everyone executed together.

As companies scaled, cracks started to show. At some point, HR's headcount formulas took over hiring: for every six engineers you bring on, add a designer, a PM, and an EM. This "squad configuration" became the default template. When you're running a product at a billion daily actives, engineer headcount explodes — and so does PM headcount.

But here's the problem: does the push notification infrastructure team actually need a PM? Engineers are perfectly capable of understanding how that system should work.

Tom used the word "infantilizes." Hire too many PMs, and engineers and designers — who are fully capable of making good decisions — gradually lose that muscle because there's always a PM "taking care of them."

His core argument: product decision-making is a muscle that needs constant exercise. The more you pull engineers and designers away from that work, the more that muscle atrophies. "Regretting the PM role" isn't saying PMs are useless — it's saying: if you can avoid it, do. Let other functions develop that muscle directly.

Whatnot's actual approach: PMs aren't permanently attached to specific engineering teams. They're assigned dynamically by project and problem. Every six months, they do a planning cycle, identify the most important problems, then decide who owns each one. That owner might be a PM — or it might be an engineer or designer.

31,832 applicants, one hire

"Over the past two years, 31,832 people applied for PM roles at Whatnot. We hired one person."

Tom says this isn't meant to discourage job seekers. The problem is that the mass expansion of product management as a profession didn't produce a corresponding supply of people with genuine core capabilities. A huge number of people have PM titles but spent five years sharpening an entirely different skill set.

The fastest path to rejection in interviews? Spending most of the conversation talking about "driving alignment" and "managing stakeholders." Tom says bluntly: their core competency isn't technical depth or user insight — it's politics.

What Whatnot actually cares about is two things. First: can the person demonstrate both macro thinking and micro execution — can they explain how a system works and immediately tell you what the smallest validation step would be? Second: have they actually made decisions — not just drifted along on a big company's momentum, but genuinely moved things forward through real complexity.

On this point, he mentions that Netflix CTO Elizabeth Stone emphasizes the same thing: systems thinking, the ability to zoom out one level and understand what your manager cares about and what impact your work has on the broader business.

Case exercises are mandatory at Whatnot. Tom says many candidates perform brilliantly in interviews — but once you give them data and a problem, ask them to come back with answers and defend them live, the quality of thinking falls off sharply. "You can immediately tell who's good at performing and who can actually think."

Senior PMs should be writing code, not running alignment meetings

Whatnot has about 21 or 22 PMs right now. Tom himself still spends 50% of his time on IC (individual contributor) work. The four or five directors who manage other PMs spend over 90% of their time doing IC work.

There's genuine frustration behind this. He explains that the old promotion logic worked like this: the better you performed, the faster you got bumped to director — and then your job became coaching, mentoring, and sitting in review meetings. Completely removed from the front lines. "We took all the A-players and promoted them, then made them stop doing things."

What he actually wants is a different model: put Messi on the field, don't make him run the youth academy all day.

What does IC work mean for a PM? Not necessarily shipping production code. Tom has pushed some production code at Whatnot, but admits that 80% of it was probably quietly refactored, because his code didn't pass lint checks and localization wasn't handled properly. Real IC work means: are you going through support tickets yourself, pulling data yourself, sitting next to engineers to understand how the system actually works, writing the spec yourself, running standup yourself.

Why design things this way? He gives a concrete example. At Twitch, he kept finding that the in-stream ads team and the content discovery team were constantly fighting over feed real estate — ads needed positions in the feed, and so did the discovery algorithm. The first thing he did after joining was put both problems under a single PM, simultaneously accountable for GMV on both sides. That PM naturally made the right tradeoffs and eliminated months of back-and-forth meetings.

When one person owns multiple things, inter-departmental politics disappear — because politics is fundamentally about putting your own interests ahead of the company's, and when someone is accountable for both things at once, they have no incentive to play that game.

AI made this finally practical

Tom doesn't think this direction is AI-driven, but he acknowledges that AI made it ten times faster.

He uses data science as an example. A user cohort analysis that used to take an L7 data scientist a week or two can now be run by a PM in a few hours using tools like Hex. More importantly, you can pull logs for specific users directly, understand their complete behavior paths, then build regression models to estimate impact. "Over the past year, I've talked to data scientists less than in any previous year of my career — but my actual depth of understanding of product data is ten times what it used to be."

The second high-frequency use case is understanding codebases. PMs used to have to constantly interrupt engineers to ask "roughly how long would this take?" Now you can talk directly to Claude, get a clear picture of the system's structure and logic, and only then go find someone to align on execution details. "Before, my goal was to reach 'boxes and lines' understanding — knowing which system drives which thing. Now there's no reason not to go one layer deeper."

There's a third use case he finds particularly compelling, because Whatnot is a live platform. When a user reports an issue during a stream, he can simultaneously watch the stream, dig into the codebase, and talk to AI to determine whether it's a bug or a user misunderstanding. User side, code side, observation side — three angles converging in real time. "It's a feedback loop with a turbocharger."

Direct evidence: the story of a seven-second countdown

In his second week at Whatnot, Tom sat in the office watching a seller run a live auction. The seller complained that the seven-second countdown was too long — could it be changed to three seconds? They'd be able to sell more that way.

Two engineers looked at each other. One said: that's a config value, we can change it.

And they changed it right there, jumped into the seller's live chat, and posted: "Refresh the app."

A moment later, the countdown was three seconds.

Tom says the only thought in his head at that moment was: good, I'm in the right place.

This isn't some extraordinary technical feat. It's a culture: when you know something is the right call, you just do it — no meeting, no PRD, no waiting for approval.

"Playing the accordion": toggling between macro and micro

This is one of Tom's internal mental models. He says he doesn't like frameworks, but he makes an exception for this one.

The accordion logic: you have to fully expand it before air can get in. But the music only happens when you push it back. Expanding is thinking; pushing back is acting — get clear on direction first, then take the smallest verifiable step. See the results, then expand again to understand what those results mean for your direction, then push back and take the next step.

He uses Whatnot's product listing problem as the example. Live commerce's defining feature is that sellers don't need to list items in advance — hold up a pair of AirPods to the camera and buyers know exactly what's for sale. Zero listing overhead. But expand that out: if a buyer searches for "headphones," how does the system know that a particular seller is currently selling AirPods? It doesn't. So search and recommendations break down entirely.

Fine, make all sellers list in advance. Expand that: each listing takes three minutes, and if a seller moves 20 items per hour, that directly caps their throughput. Real harm to a seller's business.

He says failing teams tend to fall into one of two extremes: either they iterate endlessly without knowing where they're going, or they write two- to three-year roadmaps that can't be updated, losing all the learning that A/B testing is supposed to deliver. "The key to the accordion is that expanding itself creates no value. All value happens when you push it back. But you have to keep doing both."

Average data is the biggest lie

In the Fail Corner segment, Tom was asked about the most painful failure of his career.

He didn't cite a specific product case. Instead, he described a mistake pattern that has followed him throughout his career: trusting averages.

A feature with only 3% of users using it. Looks like it should be cut. But if you don't dig one level deeper — for a certain type of user, that feature covers 100% of their use case. It's the foundation of how they run their business on your platform. Take it down, and you've cut their business off entirely.

He used an analogy: imagine a shopping mall going dark on Christmas Eve. Nobody actually wanted that outcome, but you didn't think about what was behind that 3%. Averages lie constantly. They hide the shape of the distribution.

He brings up Jeff Bezos's line: "When you have data and a story at the same time, believe the story." He says that's exactly what this is about: aggregate data doesn't tell you what's actually happening to individuals, and individual truth is the foundation of real product decisions.

Twitter's legacy: product-market fit is more durable than you think

Tom's two years at Twitter were around 2015 to 2016. He says the experience is a bit like a therapist asking about your childhood — you know exactly what's being dug up.

In those two years, Twitter cycled through nine product leaders. A colleague from a partner team said it felt like their Market Street offices "lived in the news," with external chaos and internal disorder in perfect sync.

But he took away one profound positive observation: if you've genuinely achieved product-market fit, the durability of that thing exceeds anything you'd imagine. During those years, Twitter's organization was a mess, strategy kept shifting — but the product's pull on users was so real that you could "emotionally feel" how much people loved it.

He says viewing this alongside Elon Musk's later acquisition of Twitter makes it even clearer: the brand name changed, the domain changed, the team shrank dramatically, everything changed. What remained? Network effects. Everyone was still there, because everyone was still there. That's "lightning in a bottle."

His other takeaway was negative, and more practical: "It's complicated" is usually cover for weak leadership. He uses the 140-character limit as the example. While he was at Twitter, everyone knew the character limit had to go — Japanese users tweeted six times more frequently than Western users because kanji conveys so much more meaning per character. This was the obvious endgame. But nobody would pull the trigger. Working group after working group, design sprint after design sprint — it didn't actually ship until almost two years after he left. The result? Nothing happened. Nobody died. Twitter's soul didn't collapse.

The tweet edit feature took another two and a half years after that. "Sometimes things aren't that complicated. Someone just doesn't want to make a decision."

Being a CPO is fundamentally a translation job

Tom has spent almost his entire career working for founders with strong product instincts — first Twitter, then Twitch (under Amazon's culture), now Whatnot, with two co-founders.

He says the core of this role isn't "owning the product vision" — it's helping founders translate their intuitions into something that can actually be built. A lot of CPO-CEO relationships fall apart because the CPO starts competing for vision ownership, starts contending with the CEO. That's the end of the relationship.

In practice, his strategy is: wherever a founder is present, he steps back. If Grant or Logan — Whatnot's two co-founders — are looking at a project, he confirms someone is covering it and walks away. If half the team is working on something, he might not touch it for days, because a founder is there and that's their domain.

He thinks this arrangement actually lets him go deeper in his own work. He doesn't need to be everywhere, so he can truly focus within his own scope and dig into the details.

Where live commerce ends

Someone asked Tom whether Agentic Commerce — AI agents that shop autonomously — would threaten platforms like Whatnot that depend on human interaction.

His answer was direct: welcome to Agentic Commerce, but it's solving a different problem.

Light bulbs, AC filters, a specific model of charging cable — high-intent, low-decision-cost purchases. Handing those to an AI agent makes complete sense. But the thing is, this type of shopping was never the core of retail. E-commerce has been around for thirty years and has never exceeded 20% of total US retail. The majority of American shopping still happens in physical stores.

Why? Because a lot of shopping doesn't start with a specific intent. It starts with "I want to browse." Someone goes to a mall because they have a wedding to attend and don't know what to wear. What they need is a curator — the woman who runs the shoe shop, her taste, her displays, the pair she recommended after chatting with you for two minutes. These things an AI agent cannot replace.

What live commerce delivers, he argues, is the first true convergence of the internet's scale and convenience with the social, cultural experience of in-person shopping. Back in the Twitch era, conventional wisdom was that streams under a thousand viewers had no commercial value, because CPM advertising needs volume. But on Whatnot, a seller with 50 people in their stream is prime time — because you're selling per unit, and having 50 people simultaneously in your shop is better traffic than any physical retail store.

He recently bought a live lobster on Whatnot. A seller who runs a seafood stall on a San Diego dock goes live every morning when the fishing boats come in, auctioning off the freshest catch right off the boat, delivered cold-chain to your door the next day. Tom says he had absolutely no intention of buying a lobster that morning — but he stuck around in the stream for a while and bought one.

"Agentic Commerce won't make that decision for you."