LinkedIn CPO on Building AI Agents for Every Team

How LinkedIn Is Building AI Agents for Product Teams

This is Tomer Cohen - longtime LinkedIn CPO - explaining a radical experiment in how product gets built. They've scrapped their APM program, created a "full-stack builder" career path, and built custom internal agents that are already changing how teams operate.

"70% of job skills will change by 2030." LinkedIn has unique labor market data. The fastest growing jobs are 70%+ different from last year's list. The pace of change now exceeds the pace of response. Traditional product orgs - with their 15 review steps, multiple specialized functions, and complex processes - simply can't keep up.

The full-stack builder model: Take one builder from idea to launch, regardless of their original function. Emphasize five human traits: vision, empathy, communication, creativity, and judgment. Automate everything else. The analogy: Navy SEALs cross-trained across areas, specialized on the mission, operating in small nimble pods.

Off-the-shelf AI tools don't work. You can't just bring Cursor or Figma into LinkedIn's codebase and have it reason well. They had to build a platform layer, rearchitect components for AI to understand, and work in "alpha mode" with tool companies. Each team gravitated to different tools - some prefer Figma, some Subframe, some Magic Patterns - creating convergence challenges.

The internal agents are impressive. Trust Agent runs specs through harm vectors; found all original issues plus holes they missed for years on "Open to Work" feature. Growth Agent contains all their unique loops, funnels, and past tests; now UX Research uses it to prioritize opportunities. Research Agent is trained on personas plus all historical research and support tickets. Analyst Agent lets anyone query the LinkedIn graph without SQL.

Head of craft builds their own agent. Each domain expert is building the agent for their function. The trust team built the trust agent. This ensures context, not just generic capabilities.

10 Lessons From LinkedIn's Full-Stack Builder Model

• 70% skill change by 2030 - LinkedIn's labor data; job is changing whether you want it to or not
• Full-stack builder - Idea to launch regardless of function; new career path at LinkedIn
• Five human traits - Vision, empathy, communication, creativity, judgment; automate rest
• Navy Seal pods - Cross-trained, mission-specialized, small and nimble
• Off-the-shelf fails - Must build platform layer for AI to reason over codebase
• Trust Agent - Runs specs through harm vectors; found old holes in "Open to Work"
• Growth Agent - All loops, funnels, tests; UXR now uses it for prioritization
• Research Agent - Trained on personas + all historical research + support tickets
• Analyst Agent - Query LinkedIn graph without SQL
• Head of craft builds agent - Domain experts create their function's agent

What LinkedIn's AI Transformation Means for Product Orgs

LinkedIn has unique labor market data showing 70% of job skills will change by 2030. Their response: scrap the traditional product org, create "full-stack builders" who go from idea to launch regardless of function, and have each domain expert build the AI agent for their discipline. The specialized function is dissolving.