31,832 PMs Applied. Whatnot Hired One. The Premise Is the Point.

Over two years, 31,832 people applied for a product manager job at Whatnot. The company hired one. That number sounds like a broken recruiting funnel until you hear the founding premise of the product org, stated by its CPO Tom Verrilli on Lenny's podcast this month: "We regret that product management exists."

I have been writing a version of this argument all year, so I want to be careful here. It would be easy to read Whatnot as confirmation and move on. The more useful reading is that Verrilli, who ran product at Twitch for seven years and did growth at Twitter before that, has been running the operating model I keep describing, in production, at an $8 billion GMV marketplace, and his account both confirms the thesis and corrects it in one place where I have been leaning wrong. Both halves are worth your time.

The short version

Whatnot's product org was founded on the premise "we regret that product management exists," and the numbers make it an operating model rather than a hot take: roughly 22 senior PMs across the fastest-growing US marketplace, one hire from 31,832 applicants, PM managers spending 90%+ of their time on IC work, and PMs assigned to problems each six-month cycle rather than to pods. Verrilli's target is not product management as a craft, it is the pod-ratio default that hired a PM per six engineers, infantilized engineers and designers, and produced a generation whose specialty was politics. His sharpest structural claim is the one I have argued from the theory side: the title is getting scarcer while the trade gets more durable, and AI removed the places to hide. His "verify then trust" leadership, micromanagement redefined as managing without ground truth, is the missing answer to the calibration problem I wrote about in the ladder posts. And his claim that AI's biggest PM unlock at Whatnot was data science, not prototyping, corrected me: I have been preaching build-speed while his binding constraint was learn-speed. The caveats are real, extreme talent density is the enabling input and most companies cannot buy it, but the premise, the problem-mapping, and the verification discipline transfer to anyone.

The target is the default, not the discipline

Start with what Verrilli is actually attacking, because the podcast title invites a lazier reading than the argument deserves. Product management, he notes, did not exist in early internet companies. Founders talked to engineers and designers directly, and execution was shared. Then scale forced delegation, and delegation calcified into a formula he calls the pod ratio: every six engineers triggers a PM, a designer, and an engineering manager. The ratio became a default, and defaults hire PMs for things that do not need them.

His two best lines carry the whole case. First: "Hiring so many PMs infantilizes the engineers and the designers who are perfectly capable of making good decisions but just never had to because there was always a PM to babysit them." Every decision-rep given to a PM is a rep taken from someone else who could have built the muscle. Second, on what the pod world selected for: "There's definitely a group of PMs whose specialty wasn't technical or customer oriented. It was politics."

The counter-model is specific, and the specificity is what makes it worth studying. Every six months, the CEO, Verrilli, and senior leads decide what must be true over the half-year, stack-rank the critical projects, and assign a named DRI to each. The DRI is usually a PM but can be an engineer or designer, and everyone goes through the same product review regardless of role. PMs map to problems, not pods. A team can run a full roadmap for a year with no PM attached. The premise "we regret that product management exists" operates as a forcing function: nobody gets a PM by default, ever, and so every PM that exists is load-bearing.

This is the operating model I described in the ledger, running in production with a P&L attached, and that matters because the strongest objection to the Product Builder thesis has always been "show me a real company." Whatnot is a real company. The fastest-growing marketplace in the country runs product with roughly 22 senior people, and its CPO ships production code, in his words, "me and Claude Code," while conceding the engineers quietly fix his linting.

Messi, the ladder, and the promotion machine running in reverse

The part of the conversation that connects hardest to what I wrote this week about artifacts as promotion packets is Verrilli's demolition of the promotion logic that built the old pyramid: "We took all of our A players and then promoted them out of doing things. Why wouldn't you want Messi playing for your team rather than trying to have the academy coming along all the time?"

At Whatnot the four or five PMs who manage other PMs spend over 90% of their time doing IC work. Verrilli himself is roughly half IC. And the economics are stated with a bluntness I appreciate: collapse the pyramid of five mid-level PMs under a manager under a director into three senior ICs paid what he calls "D2-VP money," and you get more impact at lower total cost. His evidence that the market is already moving: former CTOs of major companies joining AI labs as plain members of technical staff. His stated dream is that the Whatnot product bench looks like that.

This is what replaces the PM ladder, arrived at from the practitioner side rather than the theory side, and his advice to displaced PMs is exactly the advice the ladder post ends on: start doing IC work in the role you already hold. Get back to support tickets. Pull your own data. Write the spec. The line he uses to recruit product VPs who have spent five years in alignment meetings is the whole thesis in one sentence: "Don't you miss actually doing things?"

And the resolution of the apparent paradox, fewer PM jobs while PM skills become more valuable, is the cleanest formulation I have seen. The title is deflating. The trade, understanding customer, business, and technology and translating across all three, he calls "probably the most durable" skill set in tech. What died is the ability to hold the title without the trade, because AI compressed the artifacts that theater hid behind. Marty Cagan named the phenomenon product theater years ago. Verrilli's update is that the theater no longer has a stage: "There's not a lot of place to hide in that anymore."

Verify then trust is the answer to the calibration problem

Here is where the podcast gave me something I did not have. In the ladder post I argued that the new leveling evidence, judgment under uncertainty, is harder to grade than documents, and that most calibration rooms cannot read it. I named the problem and gestured at the fix. Verrilli has the fix, and it is culturally expensive, which is why most orgs will not copy it.

He rejects "hire great people and get out of their way" outright, reading it as "code for devolving the entire roadmap and disappearing." Whatnot's stated posture instead: "We tend to live in a verify then trust land as opposed to a totally trust or even trust-but-verify." The defining anecdote is his CEO, Grant LaFontaine, saying "I don't think this is right" in a product review, then: "I'm going to clear the rest of my day. Let's sit and figure it out," and sitting with the team pulling tickets, code, and data line by line. The effect, Verrilli says, is that reviews become "us versus the problem" rather than "you versus the reviewer."

The fraud anecdote makes the discipline concrete. Someone in a growth meeting says "that was fraud." How do you know? It is labeled fraud in the data set. Who labels it? Someone in ops, presumably. Do you know the SOP? No. "Then you don't know it's fraud." And the follow-through is a senior person actually going and improving the labeling system. His redefinition is the line I am keeping: micromanagement is managing without ground truth. Being in the weeds with the same data your team has is unblocking, not meddling.

This is what a calibration room that can read the new evidence looks like. Leaders who go deep enough on the micro to grade a bet on its reasoning, not its formatting. "You can't make good macro decisions without the micro" is Verrilli's aphorism for it, and it prices the real cost of the senior-IC model: it only works if leadership stays close enough to the work to verify, which most executive layers stopped being able to do a decade ago. The model does not just demand more of ICs. It demands more of the people above them, and that is the half nobody talks about when they praise flat orgs.

Where this corrected me: learn-speed beats build-speed

Now the part where I update, because an opinion post that only confirms its author's priors is a mirror, not an argument.

I have spent the year evangelizing prototype-speed: build the thing in an afternoon, put it in front of customers, let the artifact end the debate. Verrilli's claim is that at Whatnot, AI's biggest unlock for PMs was not prototyping. It was data science. A nuanced cohort analysis that would have taken a senior Amazon data scientist a week or two in 2017 now happens in a Hex thread in an afternoon, done by the PM, during the meeting where the question came up. His PMs spend roughly ten times more time in data than before, and watching a user stream while querying the codebase in real time is, in his words, "a feedback loop on steroids."

He is right, and I was underweighting it. The loop I care about has two speeds, build-speed and learn-speed, and which one binds depends on the company. I build a new product, so building was my constraint and prototyping was my unlock. Whatnot operates a live marketplace with millions of daily events, so their constraint was always synthesis, and their unlock was collapsing the distance between question and answer. The general rule that covers both: AI's value is wherever your loop currently stalls. If your roadmap debates die for lack of a working artifact, the unlock is prototyping. If they die for lack of an answer about what is actually happening, the unlock is analysis, and a PM who cannot pull their own cohort curve in 2026 is as handicapped as one who cannot ship a prototype.

His warning label belongs on both: "Using an AI tool to find a piece of data, much like using an AI tool to write code, does not absolve you of responsibility to make sure that was good analysis, good code." And the second-order consequence he describes for data science, the best people migrating upstream into data engineering, attribution, and labeling correctness because AI-assisted analysis is only as safe as the data layer under it, is the same move evals made in engineering. The ground-truth layer becomes the scarce, senior work.

Where I push back

Two honest cautions, offered in the spirit of Verrilli's own closing line, "Assume that at least half of what I've said is wrong."

Talent density is the enabling input, and it is not evenly available. The model runs on 22 people who survived a funnel that rejected 31,831 of 31,832 applicants, at a company every senior PM in the country wants to join. That is not a neutral condition. A mid-market company with normal gravitational pull cannot staff this model by deciding to, and a leadership team that cuts its PM org to Whatnot ratios without Whatnot's bench will get the failure mode I described in who owns landing: critical problems with no owner, renamed as empowerment. The premise transfers. The ratio does not, until the bench does.

And the model quietly assumes founder-grade context at the top. Verify-then-trust works because Verrilli and LaFontaine can actually verify, line by line, in systems they have total context on. He concedes very large orgs may legitimately need layers, and his own "two dads problem," where two strong leaders issue conflicting instructions and a PM shuttles between reviews, is what happens when deep-diving leadership loses coherence. The discipline scales down beautifully. Whether it scales up past founder-context size is the open question, and I notice Whatnot has not had to answer it yet.

Try this week

Three moves, sized by seat.

If you run product: audit your org against the pod default. For every PM, answer one question, what specific problem is this person the DRI for this half? Any PM whose honest answer is "they cover a team" is a ratio hire, and either needs a problem or is the person Verrilli would not have hired. Do the same for the next PM req before you open it.

If you are a PM: adopt his green-red question as personal discipline. Before your next experiment ships, write down what you do if it is green and what you do if it is red. If you do not know, the thinking is not done, and no amount of velocity fixes thinking that has not happened.

If you lead leaders: run one verify-then-trust review this month. Pick the decision you were about to wave through, clear the time, and go to ground truth with the team, tickets, data, and code on the table. Not to catch anyone. To find out whether you still can. If you cannot read the work anymore, that is the most important thing the review will teach you, and it is the gap between your org and the one Verrilli is running.

The whole conversation is worth your time: This CPO regrets that product management exists, on Lenny's podcast.

Sources: This CPO regrets that product management exists, Tom Verrilli, Lenny's Podcast, August 2, 2026. Whatnot CPO Tom Verrilli: "We regret that product management exists", BigGo Finance recap, August 2026. Product management theater, Marty Cagan on Lenny's Podcast.