Notes — Tomer Cohen on the Full Stack Builder Model, AI-Native Teams, and LinkedIn’s Internal Agents

Four questions [Adler frame]

Q1. What is it about? LinkedIn’s transition from function-siloed product development to a Full Stack Builder model — any person from any function taking an idea from concept to market with AI as the execution layer. The episode covers the why (organisational complexity collapse), the how (three components: platform, tools, culture), and the current state of the pilot.

Q2. How is it argued? Through Tomer’s 14 years as LinkedIn CPO — direct reports of what was built, what failed, and what is showing early results. The argument is empirical rather than theoretical: here are the agents we built, here is how the pod model runs, here is what change management actually required. WEF data (70% of job skills change by 2030) is cited as external grounding for urgency.

Q3. Is it true? The direction is credible. The specific claims — 50% of broken builds fixed by maintenance agents, hours saved per week per team — are consistent with published benchmarks for similar programmes. The model depends on LinkedIn’s scale and resources; the platform investment required (months of data curation, deep customisation) is a real barrier for most organisations. The 70% statistic is from WEF Future of Jobs data; Tomer presents it accurately but it has wide confidence intervals.

Q4. What of it? For product leaders: the cost of waiting is rising faster than the cost of starting. The model works in increments — you do not need the full programme to start a pod. The cultural investment is chronically underestimated. If you ship AI tools without change management, only the top 5% of early adopters will use them.

Glossary

Full Stack Builder — Anyone from any function who can take an idea end-to-end: research, design, code, launch, iterate. Not necessarily alone; small pods enable this collectively.

APB (Associate Product Builder) — LinkedIn’s replacement for the APM programme. Participants receive training in all three disciplines and join pods from day one.

Trust agent — LinkedIn’s internal AI that audits specs for vulnerability and harm vectors specific to LinkedIn’s platform (e.g. Open to Work enabling scams).

Growth agent — Trained on LinkedIn’s historical loops, funnels, and experiments. Critiques growth ideas against past evidence.

Research agent — Trained on member personas, support tickets, and past studies. Evaluates product specs against real user needs.

Analyst agent — Trained to query the LinkedIn graph. Replaces manual SQL requests to data science teams.

Pod model — Small assembled groups of Full Stack Builders, formed around a quarterly mission, then re-assembled differently.

Product jam — LinkedIn’s internal ideation and planning process, now orchestrated via a dedicated agent.

Notes

The complexity collapse

The standard product development lifecycle expanded: research became 10–15 sources, shipping required design reviews + privacy reviews + security reviews. Each step is rational in isolation; together they produce microspecialisation and organisational bloat. “We moved from process complexity to organisational complexity.”

Platform investment is the prerequisite

No third-party tool worked off the shelf. Every agent required customisation. Giving agents full access to a Google Drive “failed miserably and hallucinates like crazy.” Gold examples — curated samples of what good looks like — are the real input. This is analogous to curating training data for an ML model, not just providing access.

Culture is the constraint, not the technology

Tomer’s framing: platform + tools are necessary but not sufficient. The adoption curve follows the desktop-to-mobile transition: top talent moves first (they always seek the cutting edge); the majority needs incentives, visible wins, performance-review alignment, and peer stories. “If we build all those tools, will they use it? Right now the answer is no.”

Key levers:

Performance evaluations incorporating AI fluency/agency
Pod pilots with real results, shared publicly
Creating FOMO via restricted early access
APB programme sending a strong signal about the future

The specialisation question [?]

Tomer says some people do not want to become Full Stack Builders, and that is fine. Specialisation retains a place. He does not quantify what fraction of roles will remain specialist. This is an open question. [?]

Departure context

At time of recording, Tomer had announced he was leaving LinkedIn after 14 years. The episode was recorded as a retrospective on the programme he was building as he prepared to leave. His stated view: “Becoming is better than being” — the journey matters more than the destination.