Tomer Cohen on LinkedIn’s Feed Transformation, AI-First Product, and Career Conviction
Tomer Cohen, Chief Product Officer at LinkedIn, in conversation with Lenny Rachitsky. The episode covers the turnaround of LinkedIn’s feed, the two-million-user carve-out strategy, what AI-first product leadership actually looks like, the LLM wave transition, and a career philosophy built on conviction rather than demand.
Key ideas
- “Might be wrong but not confused.” Tomer’s operating principle. Strong conviction about where things are going is a prerequisite for making consequential product bets. Being uncertain about whether you’re right is normal; being uncertain about your direction is not.
- The carve-out as an organisational firebreak. When existing metrics and incentives made it impossible to test a radically different feed experience at scale, Tomer carved out two million users — small enough not to move headline numbers, large enough to prove the concept. The experiment ran with full latitude and produced the evidence that broke the organisational deadlock.
- AI-first means owning the objective, not delegating it. Most product leaders treat the algorithm as a black box operated by engineers. Tomer’s test: can you write the algorithm’s objective on a whiteboard? What parameters does it learn on? What is your data collection and fine-tuning investment? These are product questions, not ML questions.
- The LLM wave required letting go of the roadmap. In autumn 2022, before ChatGPT was public, Tomer convened his leaders and asked them to set aside their roadmaps and return to the drawing board — starting from problem statements rather than features. Diverge first (several weeks of broad exploration), then converge top-down on four to five bets.
- Career path: follow conviction, not demand. Tomer did not apply for his role at LinkedIn; he met the then-head of product and pitched how LinkedIn mobile should be built. Moving from startup founder to LinkedIn mobile lead to feed to ads to CPO was driven by what he had conviction about, not by what the market valued most at any given time.
The Feed Transformation
LinkedIn’s feed before Tomer’s intervention was a promotional stream that ranked primarily on predicted click-through. The hypothesis was that knowledge sharing — members sharing professional expertise with the people most relevant to receive it — was both more valuable to members and more durable as a business.
The two-million-user carve-out gave the team full freedom: different metrics, different objectives, different product decisions. Results were dramatic behavioural change over months. The evidence was unambiguous enough to drive company-wide adoption despite the organisational friction the experiment had been designed to avoid.
The principle Tomer articulates: in a marketplace, if you can satisfy the need on the supply side and the demand side simultaneously, both sides get more value. AI was the mechanism. The feed objective shifted from click-through to downstream engagement — the difference between measuring whether someone tapped and measuring whether someone acted professionally as a result.
AI-First Product Leadership
Tomer frames the shift to AI-first product as a change in what the product leader owns. In a deterministic product, you own the experience. In an AI-first product, you own the ingredients and the guidelines — the model takes care of the experience. This is threatening to product leaders accustomed to controlling every decision; Tomer’s view is that the discomfort is the signal to act, not to resist.
At LinkedIn, this took institutional form: an AI academy that every PM had to complete, modelled on the mobile-first transition of 2014. Tomer’s review sessions focused on three questions per team: What is the algorithm’s objective? What features have you added to the learning process? What is your data collection and fine-tuning roadmap?
The autumn 2022 reset extended this: with the LLM wave incoming, teams were asked to start from their core problem statements rather than from their existing roadmaps. After a period of broad experimentation, Tomer selected four to five top-down bets and consolidated resources behind them. The result was LinkedIn’s current generation of AI features, built when most companies were still assessing the opportunity.