Lenny's Podcast: Why LinkedIn is turning PMs into AI-powered full-stack builders

Published on December 4, 2025, this episode of Lenny’s Podcast features Tomer Cohen, Chief Product Officer at LinkedIn, discussing the organizational redesign LinkedIn has applied to its own product teams — and why he believes the traditional PM structure is no longer the right model for AI-native product development.

Who it is for

The episode is most relevant for senior PMs, product leads, and anyone managing or building a product team. It is also worth watching for early-career PMs who want to understand where the role is heading and which skills to prioritize.

What the episode covers

Cohen describes LinkedIn’s decision to replace its Associate Product Manager program with an Associate Product Builder program — a track that trains participants across coding, design, and product management simultaneously rather than as separate sequential specializations. The program introduces a formal Full Stack Builder title and career ladder. The change is framed not as a cost-cutting move but as a structural response to what Cohen calls a “broken traditional model” where feature development takes approximately six months from idea to release, largely due to coordination overhead between departments.

Key takeaways

1. Coordination overhead is the actual bottleneck, not individual contributor speed. Cohen argues that the cost of moving work between PM, designer, and engineer — resolving mismatched assumptions, re-explaining context — is large and largely invisible. When one person can carry a product from idea to launch, that overhead drops out entirely, and LinkedIn found this to be the primary source of latency in their development cycles.

2. AI tools raise the floor for individual contributors, but they do not level the field. Top performers adopt AI tools fastest, which contradicts the assumption that AI will reduce the gap between strong and average contributors. For hiring and team building, AI fluency is functioning as a differentiator, not a baseline requirement everyone reaches at the same pace.

3. Cultural adoption requires deliberate change management. LinkedIn updated performance review criteria and created explicit recognition tracks for AI tooling adoption because informal adoption signals — waiting to see who picks things up — fail when the change touches job identity. Celebrating early wins and creating visible success stories mattered more than announcing the tools.

4. Off-the-shelf AI agents were not sufficient for LinkedIn’s internal workflows. The team trained agents on internal processes, documents, and decision history. Generic models produced output that required enough correction to negate the time savings. Enterprise-specific training was necessary before the efficiency gains materialized.

5. The program includes AI agents for idea critique — automated devil’s advocates that pressure-test product ideas before they reach human review. Cohen describes this as a way to improve the quality of decisions that do reach senior reviewers, not to replace human judgment, but to ensure that reviewers are engaging with ideas that have already survived basic scrutiny.

Worth watching if you are evaluating whether your organization’s feature velocity problems are structural (coordination overhead between functions) rather than talent problems, or if you are building out a product team and deciding which skills to hire for versus which to train internally.