"We Might Be Wrong, But We Are Not Confused": How LinkedIn's Tomer Cohen Rebuilt the Feed Around AI

Colleagues told Lenny to ask Cohen about his best-known "Tomerism": "We might be wrong, but we are not confused." Cohen says it came from a startup founder he met years ago whose company was near failure. After the team chose a path, people kept hedging in different directions, and the founder realized that unless everyone pulled the same way there was no chance. Cohen reduces it to clarity of thought and clarity of execution, both equally important.

On clarity of thought, he says ambitious people often get attached to being right or wrong, which leaves confusion in the system. He prefers to be attached to clarity and focus. His teams spend significant time defining the problem precisely, such as which kind of video for which audience, and he likes solutions rooted in first principles that have "teeth", meaning they state a real trade-off rather than just "simple". A mentor also taught him to ask whether someone is disagreeing or merely misunderstanding, a distinction he learned after moving to the US from Israel.

Clarity of execution, he says, is even more important, because many organizations reach a decision and never act on it. If a leader names an initiative as the top priority but most engineers are on a migration, the migration is the real priority. Resourcing has to reflect what you say matters, including where your top talent works. Lenny notes that a colleague found the phrase liberating, and Cohen agrees: it frees people from the right-or-wrong game and makes it possible to learn from decisions.

Cohen believes in setting ambitious goals and then over-delivering, rather than underplaying them. He pictures a mountain: you can see the peak and the base camp, while the middle may be blurry and gets figured out along the way. Sharing the peak, he says, makes building more inspiring for the people involved. Lenny compares it to Brian Chesky's habit of asking for ten times the goal.

For the LinkedIn feed, he says it was hard at first to imagine more than a promotional product, yet he wanted millions of people to come daily. His reasoning started backwards from the potential: how many professionals exist, and how many would love a place to share and engage with content. He says it is not detached from reality, but it is not hooked to it either. He argues that without that ambition you will not reach it, and describes LinkedIn as an economic platform where one billion members is still small.

Lenny says he was skeptical that anyone would want to post on LinkedIn, and Cohen notes that there is much talk about zero-to-one and scaling but little about "minus one to one" turnarounds. These are often harder internally, he says, because of entrenched flows, processes and metrics. He sees LinkedIn as a platform for economic opportunity on top of a strong social graph, where knowledge exchange is one of the most powerful transactions. LinkedIn started with an activity feed of who changed jobs and who connected to whom, and it drifted into promotion.

When Cohen asked to lead the feed there was no unified feed team, and he gave it a new purpose: not a springboard for other products or a traffic jumpstart, but people who matter talking about things you care about professionally. The first move was to make it AI-first by bringing a centralized AI team, previously not part of any product team and pointed at different goals, into one unified team. He spent most of his time on objectives, algorithm features and data training.

Experiments at scale made other teams, which relied on the feed for traffic, nervous about their numbers, and he found himself stuck in escalations. So he carved out two million members as "his" members and built for them with full liberty. Over months their behavior changed noticeably, giving strong evidence that the pie could grow rather than just be re-sliced. He also ran negative tests to prove points, then brought the experience to everyone. He says the target was the right views reaching the right people, not volume, and that shifting the AI objective from click-through to downstream conversations also attracted spammers that had to be removed.

Because the feed is the first screen, Cohen says counting everyone who touches it counts bystanders, so the team measured active, engaged and high-value downstream engagement. On how he spots opportunities, he starts from beliefs. His biggest career shift, he says, was moving from asking what is in demand to asking what he cares about and has conviction on, and he finds it hard to be a strong product leader without conviction. He says he leads with beliefs and then tries to prove them with evidence, not the other way around.

He does not schedule time to dream, but he is reflective and likes LinkedIn's "vision to values" exercise from a former CEO, which asks what change will happen in the world if you succeed. As for what made the feed social, he points to AI as the ultimate matchmaker. In a marketplace, satisfying the need on the other side creates magic. He says AI is underutilized and misunderstood, and in the past was run separately from product teams.

He gives examples from both sides. A creator who posted an article criticizing the Olympics learned that NBC executives read it, which meant the content reached people that mattered. On the receiving side, he cites a professor who started posting daily and told him that speaking engagements were bringing in roughly half his university salary and that he had been invited to advise prime ministers on investment strategy.

Cohen says AI-first is a mindset, not a technology: it begins with strategy, then flows to product and the talent you hire. He says that when he took on the feed there was no one on the product side thinking about AI, and that this was rare in company strategies even two years earlier. He uses a rafting analogy: AI is the two pedals that steer the boat, and the guide holding them should be the product leader. Too often, he says, product people treated AI as a black box and delegated it.

He challenges product leaders whose products contain algorithms to write down the algorithm's objective on a board, name the features it learns from, and describe their investment in data collection and fine-tuning. Fine-tuning, he says, was once seen as engineering's job, but a whole strategy can be built on data collection and fine-tuning. Lenny summarizes that PMs should ask what the objective is, which features go into it, and what the data strategy is, and Cohen adds that product people should also think about infrastructure, which can be a bigger lever than another button.

To scale the mindset, LinkedIn set up an AI academy in early 2020 that every PM had to attend, like the mobile-first shift in 2014, and he spent review time on AI strategy and objectives. After the arrival of LLMs, AI moved from the background of LinkedIn's marketplaces to the front. In fall 2022 he asked teams to let go of their roadmaps and return to the problem they were solving, then tell him how this technology could solve it better. That, he says, is very much AI-first.

A major shift for AI-first leaders, Cohen says, is realizing you no longer control the experience but the ingredients. He compares product leaders to a chef used to dictating every detail, from ambiance to the temperature of the broccoli, who now has a technology that says, give me the ingredients and the guidelines and I'll cook. Many find that scary, he says. Safety guards and responsible AI are critical, but because AI is not deterministic, giving it room to learn will lead to a better result.

To avoid shipping pointless AI features, Cohen began in fall 2022 by asking leaders to go back to the objectives they were already solving rather than invent new ones for a cool technology. He then let teams run and be creative, even building overlapping ideas, because there was no playbook and he wanted to learn; prompt engineering became an internal playbook. After a few weeks he converged top-down on a handful of best bets, since capacity and cost are constraints, and reviewed those bets weekly so everyone knew what he cared about.

Lenny notes that room to explore also gives enthusiastic employees an outlet, and Cohen says his intent was learning but welcomes the point. Comparing it to Chesky's practice of injecting chaos when things feel too comfortable, Cohen says people settle into lanes and inertia, and sometimes the disruption has to be triggered from outside.

Asked about his rise from senior PM to CPO, Cohen shares his own path rather than general advice. Early on he focuses on learning from great people, some of whom did not know they were mentors. He says he grew up choosing what was most in demand and most challenging, like the toughest engineering role or army unit, and after moving to the US for graduate school in 2008 he realized that path was dictated by society, not by him.

He then shifted to asking where he had strong conviction, what he was passionate about, and where he could make a dent. That led him to start a company despite student debt, and he joined LinkedIn without applying, after explaining to a leader how he thought LinkedIn mobile should be built. He adds that if you are not genuinely excited about what you build, it is hard to have a big impact, and that product people are fortunate because they are judged on the products they built and their impact, not titles or logos.

When Lenny points out that some people only chase what excites them, Cohen agrees and says that between a thrilling project on the fringes and a grindier one at the core of the company, he would pick the core: impact first. In interviews he asks what a person built, what they learned, and how well it did.

Cohen is excited about his podcast, BuildingOne, which grew from internal fireside chats with builders from different disciplines, from Pixar co-founder Ed Catmull to product leaders and chef Dan Barber. His main lesson is that everybody builds differently and succeeds by being authentic to who they are. Lenny agrees with the power of focusing on strengths. In the lightning round he recommends Carol Dweck's book on growth mindset, Thinking, Fast and Slow, and High Output Management, and he picks the motto "becoming is better than being".

On LinkedIn features worth trying, Cohen points to video and to the job-seeker coach, an LLM experience personalized and private to the user that helps with fit, how to apply and comparing opportunities. He recounts a job seeker who cried during a session because the search felt lonely and who wanted a non-judgmental buddy they could not afford to pay a coach to be. He likes to measure impact by emotion, and says seekers afterward felt supported. The coach now sits at the top of the Jobs tab.

Looking ahead, Cohen predicts that AI will feel so intimate that being away from it will cause anxiety, like "nomophobia" about being without a phone. He says we are just getting started on an incredible revolution. He also mentions two free AI-first courses for people who want to go deeper.