Why LinkedIn is turning PMs into AI-powered "full stack builders” | Tomer Cohen (LinkedIn CPO)

By the end of this decade, the skills inside most jobs will shift dramatically, so even if you’re not changing roles, your role is changing, which means we have to go back to first principles and rethink how we build.

You’re piloting a very different way of shipping product at LinkedIn that leans fully into AI, which you call the full stack builder model, and it feels like a blueprint for how companies will build going forward.

The aim is to let great builders take an idea all the way to market regardless of where they sit, with a fluid handoff between people and machines.

My guest today is Tomer Cohen, longtime CPO at LinkedIn, who’s replacing the old APM path with an associate full stack builder track, adding a formal full stack builder role, and wiring in agents and workflows to create human plus AI teams that move fast, adapt quickly, and do more with less; welcome back, Tomer.

Technology should empower, and the pace of change now outstrips how fast organizations can respond, which our data shows in how fast skills and jobs are evolving, so we have to collapse the bloated process and organizational layers we created and return to craftsmanship.

How confident are you in that skills shift data?

Change has always been there, but this wave is steep across functions like engineering, marketing, sales, and recruiting, with some roles seeing modest impact and others reshaped almost entirely.

Give us the gist of the program today and where you want it to go.

Builders own vision, empathy, communication, creativity, and above all judgment, while we automate the rest, so we can form small mission-driven pods that reassemble quickly, like elite teams trained to flex across skills.

So one person carries the idea from insight to prototype to ship, with a tight pod around them.

Yes, it’s team-first but smaller and more fluid, with pods spinning up for a quarter and then regrouping based on the next mission.

Of those traits you want builders owning, which matters most?

Judgment is the crown skill, because high-quality decisions in ambiguity determine outcomes.

Where have you had the most success automating, and what did it take?

We’re investing across three layers: platform, tools and agents, and culture, starting with re-architecting code, components, and design systems so AI can reason over them, which meant working closely with vendors because nothing worked off the shelf at our scale.

So you’re essentially refactoring your stack to be AI-ready.

Exactly, and we also adapt external tools to our stack so they actually perform well in our context.

Walk us through the agents you’ve built.

We created a trust agent that flags harm vectors early using LinkedIn-specific patterns, a growth agent that encodes our loops and experiments to critique ideas, a research agent trained on member personas and past research plus support signals, and an analyst agent that helps query our graph; they’re strong MVPs we’re rolling out internally.

What tech are you building on?

We use enterprise-grade LLM platforms like Copilot and ChatGPT with a lot of our own customization and orchestration so agents collaborate rather than run in a straight line, and we’re testing multiple design tools since teams gravitate toward different workflows.

It sounds like you’re creating focused agents and planning to hide that complexity behind a single flow.

Yes, we’ll mask specialty agents behind a product jam experience while we benchmark each one, then let the orchestrator route work among them.

You’ve also put a lot of effort into the early product phases, not just code.

Code-to-launch has been boosted by coding, maintenance, and QA agents, with roughly half of broken builds now fixed automatically, but the idea-to-design stage is where quality is set, so we’re doubling down there to broaden who can build.

How long did it take to stand this up?

We announced late last year, had early agents trained within a handful of months, and learned quickly that you must curate golden examples and tight context rather than dump the entire knowledge base, or you get noise and hallucinations.

Where is the pilot today and how are you measuring progress?

We have sizable pockets using it, seeing hours saved each week and better conversations, and we look at experimentation volume times quality divided by time to launch, with internal GA coming as the pieces mature.

How are you running the pilot inside teams?

A central team builds the track while pods across product areas opt in and commit to feedback, and we’re seeing PMs self-serve dashboards, designers opening PRs, and teams like Semantic Search using the tools to ship faster.

You also replaced the APM program with an associate builder path.

We’re launching an Associate Product Builder program that teaches coding, design, and PM together, then places folks into pods, and we’ll reuse that training to uplevel the whole org.

Who’s getting the most out of this so far?

Top performers are leaning in hardest, giving great feedback, and seeing strong gains, and we expect broader numbers as access expands over the next two quarters.

There’s always the question of whether AI lifts everyone or supercharges the best, and it sounds like it’s the latter.

Tools alone won’t change behavior, so we’re backing them with incentives, examples, and change management, much like the desktop-to-mobile shift, because only a small slice will adopt without cultural support.

What culture moves are working?

We set clear expectations for AI fluency, spotlight wins, run pod pilots that show results, launch the associate builder program, celebrate cross-role moves like a researcher becoming a growth PM, and keep access and feedback loops open so people see it’s worth the effort.

Have you changed performance criteria to reflect this?

We updated hiring and evaluations to include AI agency and full stack behaviors, and rather than a mass reorg, we’re pushing for a mindset shift and giving explicit permission to start now and bring in tools that work.

How do you nudge people to try the tools on their own?

We demo in all-hands, invite people to share what’s working, and encourage folks to pick a few tools that click and go deep, then pull others along.

Any negative surprises?

Off-the-shelf integrations rarely worked at our scale, naive data access caused hallucinations, teams favored different tools which complicated convergence, and not everyone wants to be a full stack builder, which is fine because we’ll still need specialists, just fewer than before.

Is full stack builder now a real title and path?

Yes, there’s a formal ladder, and people are coming from design, engineering, product, and even BD, with mindset and growth appetite mattering more than labels.

This also lowers the friction for moving between product roles.

Exactly, your career incentives align with what the organization needs, so you have a tailwind to learn new skills and operate across the stack.

For leaders who want to start, what’s your advice?

Design your platform and tools, but invest heavily in culture, build in the open instead of in a silo, be ambitious yet patient, customize agents to your context, allocate real resources, and don’t expect instant two times productivity without upfront work.

Anything to emphasize before we sprint into the lightning round?

Don’t wait, set the ambition, measure and share progress, overcommunicate updates, and if your company won’t move, start anyway or join one that will so you stay at the edge of how products get built.

Lightning round time; what books are you recommending?

Why Nations Fail for how inclusive systems create opportunity, Outlive for personalized longevity thinking, and The Beginning of Infinity for the power of good explanations driving progress.

A recent show or podcast you loved?

A Hebrew podcast called One Song that unpacks the story and craft behind a single track, which I adore as a music fan.

A product you’re into or wish existed?

I want a steering wheel button that instantly invokes my AI copilot so I can talk naturally in the car without friction.

A motto you live by?

Becoming is better than being, which keeps me focused on growth and the craft, not a fixed end state.

You’re leaving LinkedIn after fourteen years; how are you feeling and what’s next?

I’m proud and grateful for the mission and the hard-earned lessons, I have a bias for change and learning, and I’m excited to go deep on new problem sets in this incredible era for builders, though LinkedIn will always feel like one of my babies.

Tomer, thanks for coming on and sharing so much; if you enjoyed this, please follow the show and leave a rating, and you can find past episodes at Lenny’sPodcast dot com; see you next time.