Whatnot's CPO: "We regret that the PM function exists"
Lenny’s recent interview with Whatnot CPO Tom Verrilli (ex-Twitch CPO, Twitter product growth) comes with a sharp title: “We regret that the product-management function exists.” He is not arguing to fire every PM. The real claim is:
Stop treating “one PM per N engineers” as the default org chart.
In one sentence
PM is a craft you learn by doing, not a credential. Defaulting to more PMs steals reps from engineers and designers who should be making decisions. A better shape is fewer, more senior, staffed to problems; managers still spend most of their time as ICs. Once AI drives the cost of validation down, “alignment-only” product theater has nowhere to hide.
How Whatnot staffs
Whatnot, a livestream shopping platform, deliberately runs the product org lean: about 20–22 PMs cover buyer, seller, and trust/risk. In two years they saw 30,000+ PM applications and hired one. Interviews are not “how I drive alignment.” They hand you a problem and a dataset → write a POV → defend it live. People who tell a good story but cannot think get exposed on the first follow-up.
| Dimension | Typical pod | Whatnot |
|---|---|---|
| How PMs attach | Bound to a team | Bound to a problem/project; re-staff about every six months |
| Who owns a new product | Default: PM | Engineer, designer, or PM — same product review |
| Manager time | Promotion means leaving the work | PM managers ≥90% IC; CPO about half IC |
| Headcount logic | Tracks engineer count | Tracks truly critical projects |
Some engineering teams can go a year without a dedicated PM. A PM may get pulled onto the next hardest problem — buyer, seller, trust — instead of babysitting notifications for three years.
What “regret” actually means
Early internet companies often had founders talking straight to engineers and designers. Scale requires delegation, but the first people you delegate to are already in the room, not a separate caste of decision specialists. The industry later froze a pod ratio of roughly “6 engineers + designer + PM + EM,” and then stuffed PMs into places that did not need them.
Tom’s line is blunt: hire too many PMs and you babysit engineers and designers who could decide. They are not incapable. They were never forced to practice.
So “we regret that PM exists” is a forcing function: hire a PM only when there is a clear need, not to complete a template. The real variables are culture and founder style, not a simple B2B vs B2C split.
AI unlocks data, not decks
His bet: the biggest unlock for PMs is data science, not prototypes.
In 2017, an Amazon L7 data scientist needed a week or two for a cohort or regression. A PM can now pull the same cut in Hex in minutes to an afternoon. He spends more time in data than at any point in his career, and fewer hours in meetings with DS. He uses Claude Code to estimate timelines and chase the order of a new-user modal; on a livestream he hears a user complain and checks the code on the spot to separate bug from misunderstanding.
The side effect is honest: traditional DS can become “reviewers of amateur analysis.” A better destination is upstream — data engineering, attribution, tracking, label quality. Using AI to fetch numbers does not excuse you from owning the quality of the analysis.
Product theater (stakeholders, alignment, storytelling) was rewarded for five years. Theater survives only while the real work is hard to verify. When verification gets cheap, people who only present have no cover.
Skills going down vs skills going up
| Going down | Going up |
|---|---|
| ”How I drive alignment” | Systems thinking: macro and micro in the same person |
| Political narrative | Being able to say what you built and where you decided |
| Pretty presentation | What if it goes green? What if it goes red? If you cannot answer, you have not thought it through |
The product-review mantra:
What do we do if the experiment is green? What do we do if it’s red?
A harder line: when you hear “this is complicated,” it usually is not — leadership will not call it. Everyone knew Twitter’s 140-character limit should go; a working group killed the work with delay. It shipped about two years after he left. Nobody died.
Three mental models that hold up
- The accordion: pull it open to get strategy and mechanism clear → compress to the smallest V1 and ship → pull it open again to review. Only opening is talk. Only compressing is throwing spaghetti.
- Know then go: think through what explodes if you scale 1,000×, then ship anyway.
- Averages lie: a 3% adoption rate can look like a kill; for one group it may be 100% of the use case. When data and anecdotes collide, he takes Bezos’s rule — believe the anecdote. Hitting internal goals 50% of the time is enough; assume half your calls are wrong.
Leadership posture flips too: not “hire and get out of the way,” but verify then trust. A CEO can clear the calendar and sit with the team in tickets, code, and data. Review becomes “us vs the problem,” not “you vs the reviewer.”
For indie products and small teams
- Ask whether you need a PM before you ask who to hire.
- Do not promote your best IC into a person who only attends meetings.
- Get into the data and the code yourself — AI has already collapsed the old “ask DS / ask eng for an estimate” gate.
- Staff to problems, not to a headcount template.
- Watch for product theater: in a world you can verify, alignment-only value is collapsing.
The caveat is real: Whatnot already has a strong product culture and an extreme hiring funnel. A huge org with weak alignment should not copy the org chart blindly. He says it himself — there is no single correct way to do product management.
Full notes live in the second brain: 创业商业/Whatnot-CPO后悔产品经理存在-Lenny访谈要点.md. Original episode: Lenny’s Newsletter · YouTube.
Comments