Tomer Cohen | Lenny's Summary

1. Key Themes

The Pace of Change Exceeds Our Ability to Respond

LinkedIn's data reveals a dramatic shift: 70% of job skills will change by 2030, and 70% of today's fastest-growing jobs weren't on the list just a year ago. Tomer Cohen frames this as entering "a phase where the time constant of change is far greater than the time constant of response. Basically means that change is happening faster than we're able to respond to it." [00:05:54] This isn't just about individual adaptation—it's an organizational imperative. "Whether or not you're looking to change your job, your job is changing. The only question is, do you keep it?" [00:06:28]

Process Complexity is the Real Enemy, Not Work Complexity

The work itself hasn't become more complex—our processes have. Cohen explains that at LinkedIn, researching a problem requires pulling from "10 to 15 sources of information" [00:07:40], and "each one of those sub-steps actually has a valid reason to exist. But when you add a whole thing together, you're like, oh my god, this is why it takes to build a small feature, multiple teams, multiple code bases, multiple sprints." [00:08:09] This process complexity spawned organizational complexity through "micro-specialization," where every sub-step has dedicated people. The solution isn't to remove validation—it's to collapse the stack through AI enablement.

Culture Trumps Tools: The 3-Part Framework

Cohen emphasizes three critical components for transformation: platform, tools, and culture—with culture being "by far for me the biggest and most important thing to do." [00:17:18] The platform requires re-architecting code bases so "AI can reason over it" [00:17:32]. Tools must be heavily customized: "We haven't seen anybody be able to work off the shelf immediately on our code base" [00:18:28]. But culture is where most fail: "I see a lot of companies roll out their agents and just expecting companies to adapt. Doesn't work this way." [00:42:44]

2. Contrarian Perspectives

Top Performers Benefit Most from AI (Not Low Performers)

"There's always been this question: Is AI going to just make people better, not amazing, more amazing, or is it going to make amazing people even more amazing?" Cohen's answer: "Our top talent are the ones who are using this the most at LinkedIn...top talent has the tendency of continuously trying to get better at their craft." [00:37:32] This contradicts the popular narrative that AI will be the great equalizer. Instead, it amplifies existing capabilities, with top performers showing the highest adoption rates because they have "this innate need to be at the cutting edge of how you build." [00:38:48]

Don't Give AI Access to Everything—Curate Carefully

"It's not great to just give it access to your drive and say reason all of this knowledge base. It actually does a very poor job understanding importance of the past and putting weights on stuff." [00:28:27] Cohen spent "weeks of work, getting up with that golden sample of examples" when rebuilding LinkedIn's feed years ago, and the same principle applies to AI agents. "Just reasoning over your entire knowledge base does not work." [00:28:52] This runs counter to the common approach of simply connecting AI to all company data.

Specialization Still Has a Place (But We Need Fewer Specialists)

Unlike the full disruption narrative, Cohen maintains: "Some people do not want to be full-stack builders and that's completely okay. Some people see themselves in specialization and I think specialization has a place and a role." However, the key insight: "I don't think we need as many specialized people as we did in the past." [00:49:22] This nuanced view recognizes that not everyone should or will become a generalist, but the ratio must shift dramatically.

3. Companies Identified

Cursor, Devon, Lovable, Bolt (AI Development Tools)

Description: AI-powered coding assistants and development platforms
Why mentioned: LinkedIn experimented with multiple coding agents but found none work off-the-shelf at enterprise scale. "We kind of experimented with Figma and subframe and magic patterns...we saw people gravitating depending on the function, their level of visibility...to different tools." [00:24:49]
Quote: "We're working with multiple companies to try and understand which product works best for us. And interestingly enough...different teams gravitate to different products." [00:24:18]

Figma (Design Platform)

Description: Collaborative design tool with AI capabilities
Why mentioned: LinkedIn is working closely with Figma to integrate their design system with AI agents, though significant customization was required.
Quote: "They first need to know how to work with our design systems, which is something there's, there are, you know, in many ways a lot of those companies are working on." [00:18:44]

GitHub Co-pilot and ChatGPT Enterprise (AI Platforms)

Description: Microsoft and OpenAI's enterprise AI solutions
Why mentioned: Used as foundational platforms for LinkedIn's custom agents, particularly knowledge corpus agents.
Quote: "For a lot of those kind of knowledge corpus agents, we're using everything from co-pilot enterprise to ChatGPT enterprise by far though, the most important part was basically our own customization of it." [00:23:46]

4. People Identified

Reid Hoffman (LinkedIn Co-founder)

Description: Co-founder of LinkedIn, partner at Greylock
Why mentioned: Cohen credits Hoffman with the original vision that attracted him to LinkedIn
Quote: "I went to a lecture at Stanford about social networks in 2008 and Reid was there and he talked about the power of being a professional communities online and I was very nerdy about it and thought it was incredible vision." [01:05:09]

5. Operating Insights

Build Specialized Agents, Then Orchestrate Them

LinkedIn created distinct agents for specific functions: a trust agent (identifying security vulnerabilities), a growth agent (evaluating growth opportunities), a research agent (trained on user personas), and an analyst agent (querying the LinkedIn graph). "We did want to be able to rate and grade those agents really well on how they're doing...we're kind of building this as a way where you'll be able to mask a lot of those. You might not know that there's a trust agent." [00:25:45] The key is building discrete, gradable components before creating the orchestration layer.

Create Scarcity and Competition for Access

"The biggest thing right now is everybody wants access. Everybody wants access with the tools." [00:33:25] Rather than rolling out broadly, LinkedIn deliberately limited access to create demand and ensure quality feedback from early adopters. The condition: "They give feedback. As a response for that, they make the tool better." [00:34:08] This mirrors product launch best practices—start with engaged users who will help shape the product.

Change Performance Review Criteria to Drive Adoption

Cohen moved his direct reports to 360-degree reviews across functions rather than within specialties, then "broadly took it across so when we hire right now we look for those and then this upcoming cycle...that's kind of a part of the performance evaluation piece and we announced it to everybody." [00:45:12] This sends the clearest signal possible: how you're evaluated has changed. "People are really incentivized by how you define expectations for them." [00:41:18]

Celebrate Cross-Functional Career Transitions

LinkedIn highlighted a user researcher who became a growth PM using full-stack builder tools: "She said I feel I can do it and she used all those tools...she's now a growth PM on the team...seeing those openings and then highlighting those people actually people are doing this have been a great example." [00:43:21] These stories demonstrate possibility and create permission for others to explore similar transitions.

6. Overlooked Insights

The Maintenance Agent Revolution (50% Automation Already Achieved)

Buried in the conversation: "We're close to 50% of all those builds being done by the maintenance agent in a QA agent." [00:27:17] This means LinkedIn has already automated half of their failed build fixes—not with future technology, but with current agents. This is a concrete, measurable impact that many organizations could replicate immediately, yet it received minimal emphasis. The implication: the low-hanging fruit in AI-assisted development isn't greenfield coding but maintenance and bug fixing.

The "Becoming vs. Being" Framework for Organizational Change

Cohen casually mentioned his evolving life motto: "Becoming is better than being...it's about continuously growing and evolving without the negativity of it or there's no sense of formal there it's just this continuous thing." [00:43:40] This seemingly simple phrase actually reveals the entire philosophy behind the full-stack builder program. Traditional organizations optimize for "being"—being a PM, being a designer, being an engineer. The future organization optimizes for "becoming"—continuous skill evolution. This mental model shift, mentioned almost in passing, may be more important than any specific tool or process change. It reframes the entire employee value proposition from role stability to growth trajectory.