AI can make sophisticated analysis accessible, but it does not guarantee that the analysis is correct. Verrilli argues that organizations need strong data engineering, taxonomies, attribution, and labeling, while users remain responsible for validating AI-produced conclusions.

AI can amplify weak analysis as easily as strong analysis
Reliable self-service depends on sound data structures
Data scientists can improve tracking and labeling systems
Tool use does not transfer accountability away from the analyst

“using an AI tool to find a piece of data, much like using an AI tool to write code, doesn't absolve you of responsibility to…”
Tom Verrilli · 42:30

#ai#data-quality#data-science#accountability

Myth Buster1:13:30

A Feature Used by 3% Can Be Essential to Every One of Them

Verrilli identifies aggregate averages as a recurring source of his most disappointing product mistakes. Low overall adoption can hide a segment for whom a feature represents the entire use case, and removing it can damage users' businesses and trigger network effects.

Average adoption does not describe individual dependence
Small cohorts can have extreme feature intensity
Deprecations can destroy a group's core workflow
Marketplace changes can directly affect livelihoods
Anecdotes and segment-level analysis can reveal what averages hide

“averages mean nothing to the individual”
Tom Verrilli · 1:14:00

“averages without thinking about the individual use cases that are hidden underneath.”
Tom Verrilli · 1:15:00

#metrics#segmentation#deprecation#marketplaces

Hot Take· 3

Hot Take22:00

Promoting the Best PMs Out of Product Work Was a Mistake

Verrilli argues that product organizations historically rewarded excellence by moving their strongest builders into coordination-heavy management. At Whatnot, even PM managers spend more than 90% of their time on individual-contributor work, while he estimates that roughly half of his own time remains hands-on.

Traditional promotion paths removed top performers from building
Senior judgment can compress decision cycles
Broad IC ownership reduces cross-team political conflict
Hands-on work keeps leaders connected to ground truth

“We took all of our A players and then promoted them out of doing things.”
Tom Verrilli · 22:30

“All of them would spend 90 plus percent of their time doing IC work.”
Tom Verrilli · 23:00

#leadership#ic-work#career-paths#product-management

Hot Take43:00

The Rise of the Engineering Manager Light

As experimentation becomes cheaper, Verrilli expects more small, focused teams led by hybrid technical leaders. These groups can take swings at previously uneconomic or technically difficult ideas without requiring a full traditional management structure.

Lower experimentation costs make small incubator teams viable
Hybrid technical leaders can guide compact teams
The role sits between technical staff and a full EM
Small teams can pursue historically difficult opportunities

“engineering manager light, for want of a better term, uh, is the thing that I'm definitely think that we'll see more and more of.”
Tom Verrilli · 44:30

#engineering-management#ai#team-design#incubation

Hot Take46:00

AI Makes Core Product Skills More Valuable—and Product Theater Less Safe

Verrilli agrees that understanding customers, business, and technology—and translating among them—becomes more valuable as building gets cheaper. However, he distinguishes those durable skills from storytelling, alignment, and leadership-facing framework communication that organizations often rewarded under the PM title.

Customer, business, and technical understanding create durable leverage
Cheaper execution raises the value of deciding what to build
The PM title does not guarantee core product capability
Hands-on case studies reveal gaps hidden by polished communication

“product skills are probably the most durable.”
Tom Verrilli · 48:00

“there's not a lot of place to hide in that anymore.”
Tom Verrilli · 48:30

#product-skills#ai#product-theater#career

Explainer· 2

Explainer17:30

Why 31,832 PM Applications Produced One Hire

Whatnot looks for product managers who can connect a broad system model to fast, specific execution. Candidates lose ground when they emphasize stakeholder politics or describe maintaining established products without showing the decisions they personally made.

Political alignment is becoming a PM hiring anti-pattern
Strong candidates combine macro and micro thinking
Specific decisions matter more than proximity to a famous product
Whatnot uses hands-on case studies to expose product theater

“In the last two years, 31,832 people applied to be a product manager at whatnot. We hired one.”
Lenny Rachitsky · 17:30

“can we see both the macro thinking and the microthinking?”
Tom Verrilli · 18:30

#hiring#product-management#interviews#judgment

Explainer1:06:30

Agentic Commerce and Live Shopping Serve Different Intent

Verrilli expects agents to handle routine replenishment and precise, high-intent searches, but argues that much retail remains exploratory and social. Live commerce combines internet scale with human curation, entertainment, and discovery rather than competing directly with automated purchasing.

Agents suit routine and high-intent purchases
Most US retail spending still occurs offline
Low-intent shopping depends on taste, curation, and discovery
Commerce can support economically viable streams with small audiences
Live shopping addresses a different need from automated procurement

“most of commerce in America isn't actually high intent.”
Tom Verrilli · 1:07:30

“I think it's a completely different kind of like customer need.”
Tom Verrilli · 1:09:30

#agentic-commerce#live-commerce#retail#whatnot

Story· 3

Story38:30

Engineers Changed a Live Auction Timer While Watching a Seller

During Verrilli's second week at Whatnot, a seller complained that seven-second auctions moved too slowly. Two engineers recognized that the timer was configurable, changed it during the stream, and asked the seller to refresh, immediately producing a three-second auction.

Direct customer observation exposed a concrete need
Engineers recognized the change as configuration rather than a project
The team modified the experience during the live stream
The episode illustrates Whatnot's bias toward responsive execution

“that's a config, we could totally do that.”
Tom Verrilli · 39:00

“I'm I'm with my people, I'm in like the right place”
Tom Verrilli · 39:30

#whatnot#live-commerce#engineering#customer-feedback

Story1:09:30

Twitter Taught Verrilli What Real Product-Market Fit Feels Like

Despite organizational chaos and repeated product-leadership turnover, Twitter retained unusually strong emotional attachment and network effects. Verrilli also learned that apparently complex product debates can persist because leaders avoid making a decision, as happened with the long-delayed expansion beyond 140 characters.

Strong product-market fit can survive severe organizational dysfunction
Emotional customer fervor is stronger evidence than merely adequate graphs
Network effects preserved Twitter through extensive change
Some complex-looking debates are failures of leadership
The 140-character decision was delayed long after the end state was understood

“if you manage to bottle lightning, doesn't matter how badly you screw up the organization uh of it, like it's huge.”
Tom Verrilli · 1:10:30

“sometimes things aren't actually that complicated. It's just weak leadership.”
Tom Verrilli · 1:12:00

#twitter#product-market-fit#network-effects#leadership

Story1:21:30

The Livestream Purchase That Ended With a Lobster at the Door

A San Diego seafood seller streams crates as fishing boats arrive, auctions the fresh catch on Whatnot, and ships purchases overnight. Verrilli bought a California spiny lobster he had not intended to purchase that morning, illustrating the serendipity that automated commerce may not replace.

The seller broadcasts seafood arriving directly from boats
Customers bid during the livestream
Purchases ship overnight in temperature-controlled packaging
The experience creates demand rather than merely fulfilling intent

“The most fun thing I bought recently, I'm not joking, is a live lobster.”
Tom Verrilli · 1:21:30

“I had no intention of buying a spiny California lobster that morning.”
Tom Verrilli · 1:22:00

#whatnot#seafood#live-shopping#discovery

Tool· 1

Tool35:30

AI Lets PMs Do Ten Times More Data Work

Verrilli identifies AI-assisted data analysis as Whatnot's largest productivity gain, citing Hex threads for cohort analysis, user-level investigation, forecasts, and regression models. AI also helps PMs inspect code without interrupting engineers and connect live customer behavior to the underlying implementation.

Hex threads accelerate nuanced cohort analysis
PMs can inspect individual user logs and estimate impact
AI helps detect regressions and product interactions
Code assistants reduce engineering interruptions
Live observation can connect customer confusion to code behavior

“the first one I think by our country mile is data science.”
Tom Verrilli · 35:30

“I've probably spent 10 times more time in data and understanding actually how the product's working than I ever have in my career.”
Tom Verrilli · 36:30

#ai#data-science#hex#productivity

Takeaway· 2

Takeaway25:30

Product IC Work Is More Than Shipping Code

Hands-on product leadership means knowing support problems, pulling data, examining the codebase, writing specifications, and working directly with engineering and design. Shipping production code can be educational, but Verrilli says it is usually not the highest-leverage use of a PM's time.

Start with customer tickets and direct evidence
Pull and understand data personally
Query the codebase to learn how the product works
Write specs and participate directly in delivery
Use coding for comprehension rather than status

“whatever is required to kind of like most effectively ship is the short answer.”
Tom Verrilli · 26:00

“do I think that's really the best use of my time? Not really.”
Tom Verrilli · 26:00

#ic-work#product-management#customer-research#code

Takeaway53:30

Avoid the Two Dads Problem in Founder-Led Product Teams

When a founder is already deeply involved in an initiative, Verrilli steps out rather than adding another product authority. He advises product leaders to remember that the founder owns the company's vision and to learn how to offer evidence-based challenge without competing for control.

Avoid overlapping senior product direction
Clarify which leader is accountable for each initiative
Ask whether a founder is open to feedback before pushing
Seek missing context before assuming disagreement
When both sides only have opinions, the CEO decides

“if it's nobody's got data, it's just two opinions, the CEO's opinion is going to win.”
Tom Verrilli · 1:01:00

#founders#cpo#leadership#influence

Frameworks in this episode

Need-Based Product Management

No, Then Go

Play the Accordion

Verified, Then Trust

Insights & moments

⚡Myth Buster· 2

AI-Generated Analysis Still Requires Data Judgment