May 27, 2023 · 1h 36m · news
Tomer Cohen: Why LinkedIn Stories Failed; How LinkedIn's Feed Was Born; AI Startups | E1019 · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this interview, LinkedIn's Chief Product Officer Tomer Cohen joins Harry Stebbings to discuss his journey to product leadership, the core frameworks driving LinkedIn's evolution and feature pivots, and the transformative impact of generative AI on business, organizational structures, and the future of knowledge creation.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 16.7% of the talking time here. How this is scored →
Harry as informed peer 3.0 Guest teaching 2.8 Guest disagreement 1.2 Harry pushing back 2.4
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 5:26 Tomer rejects host's binary framing
When Harry asks whether product development is art or science, Tomer explicitly pushes back, noting the question created tension because delineating art from science is impossible.
Hardest push from Harry ▶ 1:15:55 Harry rejects comparison between search engines and AI scrapingHarry refuses Tomer's framing that AI content scraping is like search indexing, pointing out that AI directly steals user traffic rather than routing users back to the publisher.
Biggest teaching moment ▶ 12:03 Tomer re-educates on the hidden emotional drivers of B2B software purchasingTomer breaks down how purchasing B2B software is not a functional spreadsheet task but a high-stakes, career-risking emotional job centered on building internal consensus.
Harry holds his own ▶ 1:10:08 Harry challenges Tomer's assertion on diminishing data moatsHarry uses concrete industry examples like Navan/TripActions to challenge Tomer's assertion that pre-trained LLMs diminish the power of proprietary data.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Tomer Cohen's Journey to LinkedIn CPO | 0 | 0 | 0 | 0 | Harry sets a welcoming tone and asks standard background questions about Tomer's career journey. Tomer recounts his trajectory from chip design to meeting Reid Hoffman and joining LinkedIn. | |
| Profound Lessons from Reid Hoffman | 2 | 3 | 2 | 1 | Harry asks whether product design is art or science. Tomer gently rejects the binary premise, noting it created tension for him because art and science are deeply interwoven. | |
| The Three Product Profiles: Visionary, Craftsman, or Operator | 4 | 2 | 1 | 2 | Harry references Shreyas Doshi's three product archetypes (Visionary, Craftsman, Operator) to prompt Tomer. Tomer avoids locking himself into a single box, emphasizing his love for all three aspects. | |
| The 'Job to Be Done' Framework at LinkedIn | 1 | 4 | 0 | 0 | Tomer schools the host on humanizing product needs, explaining that B2B buying is deeply emotional and social rather than purely functional. Harry listens intently while prompting with quick follow-ups. | |
| The Role of Founders in Corporate Innovation | 4 | 3 | 2 | 2 | Harry cites Shopify's VP of Product claiming innovation stops when founders leave the CPO role. Tomer politely disagrees with the absolute claim, citing counterexamples like Microsoft, Intuit, and Netflix. | |
| Evaluating LinkedIn's UI and the Impact of Generative AI | 3 | 2 | 2 | 6 | Harry asks a direct and blunt question about LinkedIn's UI feeling outdated. When Tomer pivots to corporate growth statistics, Harry pushes back asking him to go deeper on where product clarity resides. | |
| Rethinking the Social Graph with Topical Recommendations | 4 | 3 | 1 | 3 | Harry references TikTok and Facebook shifting away from social graphs and queries Tomer on connection utility. Tomer highlights the shift toward topical recommendations and follow relationships. | |
| Product Failures: The Ephemeral Story Experiment | 4 | 4 | 2 | 3 | Tomer breaks down the failure of LinkedIn Stories as misunderstanding creator intent. Harry presses on the importance of speed to market citing Alex Schultz, but Tomer reframes the key metric as product-market fit rather than launch speed. | |
| Evaluating Post-Launch Data with Conviction | 2 | 3 | 0 | 0 | Harry asks how product teams decide between early data signals and product conviction. Tomer outlines the distinction between building on pre-existing evidence versus hypothesis-led conviction. | |
| Structuring Collaborative Product Jams | 1 | 2 | 0 | 1 | Tomer outlines how LinkedIn rebranded product reviews into collaborative Product Jams. Harry asks about remote versus in-person discussion quality. | |
| Maintaining Priority and Accountability Post-Jam | 2 | 2 | 0 | 1 | Harry asks how accountability is maintained post-jam when multiple parties are involved. Tomer explains decision framework roles like recommenders, approvers, and single decision-makers. | |
| The Turnaround of the LinkedIn Feed | 3 | 2 | 3 | 6 | Tomer shares the turnaround story of the LinkedIn Feed. When Harry repeatedly presses Tomer to give his feed product a numerical rating out of 10, Tomer persistently deflects to qualitative measures. | |
| Balancing Short-Term Revenue and Long-Term Innovation | 5 | 3 | 2 | 5 | Harry challenges grandiose corporate visions like Stripe's 'increasing internet GDP', noting it means nothing to a freelancer needing clients. Tomer explains how job-to-be-done frameworks prevent losing touch. | |
| Leading Product Teams in the Age of AI | 2 | 3 | 0 | 0 | Harry asks how AI alters product leadership skills. Tomer uses analogies of white-water rafting and shift from chef to ingredient provider to explain loss of deterministic control. | |
| Enterprise AI: Bundled vs. Unbundled Preferences | 4 | 2 | 1 | 2 | Harry puts forward a strong thesis that enterprise buyers are inherently lazy and prefer bundled AI suites. Tomer agrees, elaborating on how central AI models collapse multi-step workflows. | |
| AI-Driven Productivity and the Future of Code Creation | 6 | 4 | 3 | 6 | Harry cites GitHub stats and asks where value accrues in AI. When Tomer suggests data advantage is diminishing due to pre-trained models, Harry directly pushes back using specific proprietary data examples like Navan/TripActions. | |
| New Team Structures and the Craft of Prompt Engineering | 1 | 3 | 0 | 0 | Tomer discusses prompt engineering as a core competency, giving an example of asking AI to explain quantum physics in a PBS style for a 10-year-old. | |
| Content Discovery, Scraping, and the Evolving Internet Business Models | 5 | 3 | 2 | 6 | Harry confronts Tomer on AI models scraping publisher data without compensation. When Tomer equates it to traditional web search indexing, Harry refuses the analogy, pointing out LLMs consume the user visit directly. | |
| AI Safety, Regulation, and Responsible Principles | 3 | 3 | 2 | 2 | Harry asks whether AI models should be politically correct. Tomer separates fundamental model objectivity from application-level safety guidelines. | |
| The Next Frontier of AI: Generating New Knowledge | 2 | 4 | 0 | 1 | Tomer explains the transition from AI analyzing existing knowledge to generating net new scientific knowledge. Harry asks if generating new knowledge is synonymous with AGI. | |
| Education in the Age of AI: The Growth Mindset | 4 | 3 | 1 | 3 | Harry questions traditional schooling relevance, sharing how ChatGPT produced a complex German real estate analysis in seconds. Tomer argues growth mindset is the ultimate enduring skill. | |
| Quick Fire Round | 3 | 3 | 2 | 3 | In the quick fire round, Tomer answers on hiring, embarrassing products, intermittent fasting, and Microsoft's OpenAI partnership. Harry gently pushes back on intermittent fasting causing cognitive decline. |
Statements from this episode (33)
Reid Hoffman reframes binary choices with unexpected 'option orange' alternatives
“I would come to Reed and I'll say, Hey, there's option A and option B and help me think for those. And he'll give me option orange that I haven't even thought about before.”
Cohen: Product management cannot be learned in a classroom
“You can't sit in a classroom learning how to do product. You have to build.”
Cohen: Most product launches are 0.6, not zero-to-one successes
“Most launches are never zero to one. They're like zero to .6. They're never great. And like moving it from .6 gradually to really finding product market fit.”
Tomer Cohen: Great product leaders love building, not decision-making power
“You have a lot of people who are coming to product to make decisions. I don't think to think of those as the right product people. I think it's inherently in the love of building and trying to create change.”
Cohen: B2B purchasing is highly emotional and stressful, not just feature-based
“When you buy a B to B product, it's actually a highly emotional, stressful job, because if you're going to buy a widget for a company, for your company, that's going to cost 50,000 dollars, a 100,000 dollars, a million dollars, and it's going to take months to…”
Cohen: 50% of early LinkedIn creators got commercial inquiries within a month
“We had a creator. It actually started with some early creators and we found that 50% of them in the span of a month after creating on LinkedIn, we're getting reached out to people saying, Hey, maybe you can do a session with my team, do a workshop on your craf…”
Cohen: LinkedIn does not offer revenue sharing to content creators
“We don't do ref share. Your opportunity of LinkedIn is your voice. It's what you create. It's your reputation. It's your brand.”
Cohen: Early-stage founders must stay intimately involved in shaping product
“Now, if you take it all the way back to a startup, Or a young company, then yes, I will be very worried if the founder is not intimately involved in the product, either as the head of product or playing a key role in shaping the product, because ultimately it …”
Cohen: Mature company founders shift from operations to spiritual guidance
“But if you think of a company evolving with multiple founding moments that are still shaped by the original vision, then you have a more expensive way to think about The role of a founder. And I think it's becomes a lot more in spirit and guiding than in opera…”
Cohen: LinkedIn's annual revenue more than tripled over six years
“Revenue for LinkedIn more than tripled.”
Tomer Cohen: LinkedIn's member base more than doubled over six years
“Our memo base more than doubled.”
Cohen: LinkedIn is growing at its fastest pace in 20 years
“We're growing the fastest we ever grew right now.”
Cohen: LinkedIn's 'follow' feature is growing 200% year-over-year
“Actually, follows right now is one of the most impressive growth trajectories at LinkedIn. It's growing at 200% year over year.”
Cohen: LinkedIn Stories failed because users want permanent professional content
“When we did follow-up sessions with members after that, it was clear that we completely misunderstood the job to be done for creators on LinkedIn. It wasn't about wanting things to disappear. It was almost like the opposite. People wanted things to last on Lin…”
Cohen: Define success metrics before building, not retroactively post-launch
“I always like to start with before you launch, before you even start building, what is success? Like, you know, once you launch it, what should be going up into the rights that should be excited about? Don't tell me in retrospect because, you know, then you're…”
Cohen: Retention is the true test of product-market fit, not adoption
“Ultimately you have success with a product when you have real adoption and retention. People actually find product market fit. They're coming back to the product to engage with it. That's when, you know, you really have something and adoption by itself, you kn…”
Tomer Cohen: Talent hiring has historically favored pedigree over measurable skills
“Talent marketplace has been built on pedigree and experience and not necessarily on the skillset you have, because that was not easy to measure.”
Cohen: In-person product reviews outperform remote sessions in creativity and velocity
“I actually do my product jams in person. I feel there is an energy of creativity and discussion that you get in the room in person. It's, you can do it remote as well. It's, there's no, It's not that remote does not work, but I think there is an edge to doing …”
Cohen: Platform growth and trust exist in a fundamental trade-off
“Usually there is a trade-off between trust and growth, right? Because you want to allow for a lot more growth in the system. You want to reduce friction. When you read restriction, you also allowed, allow usually for bad activity to engage as well. And the mor…”
Cohen: LinkedIn uses a 'brief back' process after product reviews
“We have a new process we started a while back, which is called brief back. The team itself, which presented, they send a summary of the session with the feedback and all the key action items they have and the ETAs for the key areas they're going to work on.”
Cohen: Resolving team disagreements quickly requires a single designated decision-maker
“Whenever there is an escalation or a disagreement, we try to solve it really quickly with one person. It's a person on the team. It's always a person that has the decision On how to resolve the situation.”
Cohen: LinkedIn intentionally curtailed record viral feed growth to prioritize quality
“We made a core decision to curtail growth even at the expense of engagement that we've never seen at LinkedIn growing at such record levels to make sure we go back to the drawing board and we focus on trust and quality.”
Cohen: Gen Z and entry-level professionals are LinkedIn's fastest-growing demographic segment
“So for example, at LinkedIn, one of our fastest growing segments are Gen Z's and entry level professionals.”
Cohen: Product leaders must give up deterministic control when building AI
“You don't control the experience. AI is not deterministic. If you ever had an interaction with JGPT, it's, you know, you're, most of the time you're trying to, hard to anticipate how it would respond. And I give the analogy of a chef in a restaurant where befo…”
Stebbings: Enterprise software buyers are inherently ignorant and lazy
“I think enterprises are inherently ignorant and lazier than we give them credit for. And they like simple solutions, especially in Europe.”
Stebbings: Enterprise AI will be dominated by bundled products in 3-5 years
“I think we enter a world of bundled AI products, especially from the next three to five years.”
Cohen: Centralized AI models will trigger a massive tech software bundling wave
“So we are, in my opinion, going into a transformation of bundling when you'll see better, more efficient experiences done by one centralized brain, which is the AI model itself.”
Cohen: Incumbents hold major AI advantages in compute, talent, and distribution
“If you're an incumbent and you already have the infrastructure for AI, the talent for AI, you're already have a leg up in this race and you should be using that. Incumbents have existing market share. They have already an existing customer base and that custom…”
Cohen: LLM pre-training on public data diminishes historic incumbent data moats
“Technologies like GPT are already trained on all public data. They've read every book, every healthcare book. They've trained on every piece of literature, art, best practices. This, the whole idea of the pre-trained is one of the biggest inflection points of …”
Cohen: Information AI models should prioritize factual objectivity over political correctness
“If you're trying to access information, the models should try to be as factual and objective as possible. And I think that kind of puts aside the notion of being politically correct.”
Cohen: Job skill requirements changed 25% over five years
“If you look at the skill set you needed for a job today versus five years ago, that has changed by 25%.”