Tomer Cohen: How LinkedIn turned cringe feeds into signal

Cohen’s mantra, “We might be wrong, but we’re not confused,” pushes teams to align on a clear problem, a sharp opinionated solution, and explicit tradeoffs—then execute fully in one direction instead of hedging and stalling.

At LinkedIn, product leaders are expected to own algorithm objectives, features, data collection, fine-tuning, and even infrastructure choices, recognizing AI as the core matchmaker that drives value in every marketplace.

To reinvent the feed without destabilizing company-wide metrics, LinkedIn isolated 2 million members, ran bold AI-driven changes there, and used the dramatic behavior shifts in that cohort to justify rolling the new model out broadly.

Cohen sets ambitious “mountain peak” visions (e.g., LinkedIn as a daily destination for millions, or a billion-plus professionals sharing expertise) and then works backward, using current numbers only to plan base camp, not to cap ambition.

When the new wave of LLMs hit, LinkedIn asked every team to drop their roadmaps, restate their core objectives, and then reimagine better solutions with AI—avoiding random “AI features” that don’t solve real problems.