Whatnot

Every PM added takes decision-making reps away from engineers and designers. Tom Verrilli, Whatnot's CPO, says product management became a default role through scale and fixed ratios like one PM per six engineers, not through necessity. Decision-making is a muscle, and abstracting it away leaves engineers and designers who never had to make calls.

The ideal is engineers and designers close enough to users to make the same calls a PM would. A notifications infrastructure team, for example, probably needs no PM because the engineers understand how it works. Verrilli says culture, not consumer versus enterprise, decides whether PMs are needed, and admits the model comes from one fast-growing, founder-led company and may not hold for very large organizations.

Lenny's Podcast

Hire a PM only for a specific need and map them to problems. Whatnot runs about 21 PMs across buyer, seller, and trust and risk while moving billions in GMV. PMs are matched to problems and critical projects rather than teams, reassigned each 6-month planning cycle, and some teams go a year without a PM.

Engineers and designers can be the DRI of product work, but everyone goes through the same product review. "You don't hire a PM just for the sake of hiring one, you hire one with a really specific need."

Lenny's Podcast▶ 9:55

Leaders set outcomes top-down, then name one owner per priority. Each cycle the CEO, the CPO and senior leads list what must be true and the critical projects, then assign a named DRI to each. The assignment is a problem to solve, not a feature to build.

Not every owner is a PM; those who are get matched on skill such as financial work, recommendations or core user features. An important item with no owner goes to a PM "with a congratulations".

Lenny's Podcast

A hands-on case study exposes candidates who are good at theater, not specifics. Whatnot received 31,832 PM applications in two years and hired one. Every hire gets a prompt and data, writes a PRD, and defends it verbally.

Many candidates were strong at alignment and storytelling but their thinking decayed when pushed for specifics. Verrilli looks for both macro system thinking with an end state and micro impatience about validating fast, and sees PMs who lead with politics and stakeholder management as trending down. Big-company PMs often polished existing things instead of deciding on a novel problem.

Lenny's Podcast▶ 49:11

Keep top PMs doing hands-on work instead of promoting them out of it. Whatnot's four or five PM managers spend 90% or more of their time on IC work, and Verrilli spends about 50%. A leader with 10 to 15 years of experience decides faster than a chain of PMs and engineering teams debating.

Directors who only did alignment and review produced a yo-yo of work, rejection and rework. Verrilli argues the combined pay of a layered org (five L5s under one L7) funds a few high-impact senior people. "We took all of our A players and then promoted them out of doing things."

Lenny's Podcast▶ 22:46

One owner across competing areas removes months of politics. At Twitch, ads and discovery fought for months over feed impressions. Verrilli put ads inside discovery under one PM, who traded organic against paid GMV, and the conflict went away.

He says separate owners drift from "company first" to "career first".

Lenny's Podcast

Verify, then trust: leaders dig into the details when something looks wrong. Verrilli rejects "hire great people and get out of their way". Whatnot's CEO Grant clears the rest of his day when a review feels off and works through the data with the team.

This sets a culture of seeking truth rather than "listening for yes" in reviews. "You can't make good macro decisions without the micro."

Lenny's Podcast▶ 51:10

Ground truth means tickets, data and labels, not shipping code. IC work for a PM means reading support tickets, pulling data, querying the code base, writing the spec and running stand-up. When someone says an outcome was fraud, Verrilli asks how it was labeled; if nobody knows the SOP, they don't know it was fraud.

A good director goes and finds out, which strengthens the labeling system agents use. Verrilli has shipped production code but says engineers quietly rework it. "It's too easy otherwise for growth to hide all sins."

Lenny's Podcast▶ 30:10

Check the small group that depends on a feature before cutting it. A feature used by 3% of people may be 100% of what a small group does, and in a marketplace it can be someone's business. Verrilli says his worst decisions came from relying on averages.

Deprecating such a feature for expediency can break that group's use case and spiral through the network. "Averages mean nothing to the individual."

Lenny's Podcast▶ 1:14:55

PMs pull cohorts and models themselves with AI data tools. Data pulls that took an Amazon L7 data scientist a week or two in 2017 now happen in a Hex thread. Verrilli says he spends 10 times more time in data while talking to data scientists less than ever.

PMs pull nuanced cohorts, inspect one user's logs and build sensitivity and regression models. Data scientists shift toward fixing tracking, attribution and data structure, and an AI answer does not absolve anyone of checking the analysis.

Lenny's Podcast▶ 35:58

Query the code base with Claude instead of asking engineers. PMs use Claude to understand how the system works, get rough LOEs, and inspect logic such as feed ordering. A senior person who spots bad feed ordering can just change it, skipping a kickoff, an alignment meeting and a week of PRDs.

Watching user sessions while reading the code live tells whether a complaint is a bug or a comprehension gap. Verrilli calls it "a feedback loop on steroids".

Lenny's Podcast▶ 38:18

Live observation plus config lets fixes ship during a show. In Verrilli's second week, a seller wanted a 3 second auction timer instead of 7. Two engineers saw it was a config, changed it, and told the show to refresh the app.

Watching sellers live exposes problems instantly, and the short loop from observation to shipped change is a core advantage of the format.

Lenny's Podcast▶ 38:55

Play the accordion: zoom out to plan, zoom in to ship. PMs alternate between re-planning from what they just learned and shipping the smallest thing that proves the belief. Zooming out showed new buyers expect search, while forcing sellers into 3 and a half minute listings would cut how much they sell per hour.

It avoids both throwing spaghetti at the wall and two to three year roadmaps that waste what A/B tests teach. If V1 works, stay on plan for V2; if not, V3 must differ.

Lenny's Podcast▶ 1:01:38

Play out consequences before building: know then go. In product review Verrilli asks what the team does if the experiment is green and what it does if it is red, and has them picture 1,000 times the expected usage.

The motto "know then go" means thinking through risks, not solving all of them. A high-growth business wants moves that compound, not a 5% statistical win.

Lenny's Podcast