Tom Verrilli

AI Summary

→ WHAT IT COVERS Tom Verrilli, CPO of Whatnot and former CPO of Twitch, argues that product management as a default organizational structure has damaged engineering and design teams by removing their decision-making reps. Whatnot runs 20 PMs across a fast-scaling marketplace, hired one PM from 31,832 applicants in two years, and maps PMs to problems rather than teams. → KEY INSIGHTS - **PM-to-team ratios cause skill atrophy:** The standard pod model—one PM per six engineers—creates a dependency that prevents engineers and designers from developing product judgment. Whatnot deliberately avoids mapping PMs to specific teams, instead assigning them to high-priority problems every six-month planning cycle. This forces engineers and designers to build the decision-making muscle that constant PM coverage would otherwise eliminate entirely. - **Hiring filter: IC output over alignment theater:** Verrilli screens out candidates who center their interviews on stakeholder management and driving alignment, treating these as signals of political rather than product skill. The case study stage—required for all whatnot hires—reveals this quickly: candidates who present well often fail when given real data and asked to defend a specific point of view under verbal questioning with no preparation time. - **Senior PMs as ICs multiplies output:** Promoting top PMs into director roles that are purely managerial removes the highest-judgment people from actual building. At Whatnot, managers and directors spend over 90% of their time on IC work; Verrilli himself spends roughly 50% doing IC work. One senior PM with strong instincts can cover the workload of multiple junior PMs while making faster, higher-quality decisions with less coordination overhead. - **Play the accordion—zoom out before every build cycle:** Before shipping anything, teams should fully expand their thinking to understand the systemic implications of a feature, then compress back into the smallest testable version. Verrilli uses Whatnot's listings problem as a concrete example: removing listing requirements speeds sellers but breaks search and discovery for buyers, a knock-on effect only visible when zooming out before committing to a direction. - **AI enables data science without data scientists:** Hex threads and similar AI-powered data tools now let PMs pull nuanced cohort analysis, trace individual user logs, and build sensitivity models in hours—work that required a senior data scientist weeks in 2017. Verrilli reports spending less time with data scientists than at any point in his career, while spending ten times more time actually inside the data than before these tools existed. - **Know then go—think through failure modes before shipping:** Rather than slowing down for risk reviews, Verrilli's teams run a mental exercise before every build: what breaks at 1,000x scale, what are the unexpected knock-on effects, what happens if the experiment is green versus red. Teams that cannot answer what they would do differently based on each experimental outcome have not thought through the problem sufficiently and should pause before writing a single line of code. → NOTABLE MOMENT Verrilli describes watching two engineers at Whatnot notice mid-livestream that a seller wanted a shorter auction timer. They recognized it was a configuration change, updated it in real time, told the seller to refresh, and the change was live within minutes—no PRD, no alignment meeting, no PM involved at any stage. 💼 SPONSORS [{"name": "WorkOS", "url": "https://workos.com"}, {"name": "Mercury", "url": "https://mercury.com"}] 🏷️ Product Management, Org Design, AI Productivity Tools, Marketplace Products, Hiring and Recruiting, IC Culture