Fireside PM: five senior product leaders on the future of PM in the AI era

In December 2025, Tom Leung convened five senior product leaders for a conversation about how AI is reshaping product management from the inside. The panel included Rami Abu-Zahra from Amazon’s Kindle and Prime Video teams, Todd Beaupre from YouTube’s home and recommendations group, Joe Corkery (CEO and co-founder of Jaide Health), Lauren Nagel (VP of Product at Mezmo), and David Nydegger (Chief Product Officer at Oviva). The discussion surfaced tensions that rarely appear in headline coverage: where AI genuinely changes the job, and where it does not.

The panel agreed that human judgment is not replaceable in the near term. One speaker put it directly: “AI cannot feel the pain your users have.” Lived experience, emotional attunement to user problems, and the ability to make conviction-based bets on uncertain information remain distinctly human. What AI changes is the execution layer — prototyping, documentation, query resolution, and synthesis of large information sets all move faster. But faster execution without good judgment produces wrong things more quickly, not better products.

On the technical side, the panel identified two new skills that product managers building AI features need regardless of their background. The first is context engineering: understanding how to structure inputs to a model to get reliable outputs, and knowing how context window management affects product behavior at runtime. The second is evaluation: building test sets, defining success metrics for probabilistic outputs, and monitoring for quality drift over time. Neither skill requires coding ability, but both require more direct engagement with how models actually behave than most PMs currently have.

The panel also raised concerns about wrapper applications — products built as thin layers on top of foundation models without proprietary data or deeply embedded workflows. Several speakers expressed skepticism about whether these products have durable competitive positions. The products most likely to hold value are those that accumulate proprietary interaction data over time and integrate AI into workflows that are genuinely difficult to replicate with a different provider.

The piece is most useful for senior individual contributors and product managers who want a grounded, practitioner-level discussion of where the job is heading, framed by people actively building in AI rather than predicting from the outside.