67:32
full-stack builder model
LinkedIn CPO Tomer Cohen argues that AI is collapsing the traditional, over-fragmented product-development pipeline (research, design, code, review each split into micro-specialized roles) back into a 'full-stack builder' model, where individuals or small AI-augmented pods take an idea from research through launch themselves — and he is piloting this at LinkedIn as a template he believes other companies will need to adopt, provided they invest as much in culture and change management as in platform and tooling.
Tomer Cohen predicts the skills required to do your current job will change by 70% by 2030, meaning 'whether or not you're looking to change your job, your job is changing.'
LinkedIn's 'full-stack builder model' aims to let any builder take an idea to market regardless of role or team, framed as a fluid human-AI interaction rather than a rigid sequence of handoffs.
The traditional product lifecycle (research → spec → design → code → launch → iterate) became bloated at LinkedIn's scale into dozens of sub-steps, multiple review types, and micro-specialized sub-roles (e.g., design splitting into interaction, animation, content design); Tomer argues the work itself isn't complex, the process was made complex.
LinkedIn scrapped its APM (Associate Product Manager) program and is launching an 'Associate Full-Stack Builder' (APB) program in January that trains recruits in coding, design, and PM before placing them in pods.
The org is shifting to small, cross-trained 'pods' (engineer/designer/PM) that own a problem for roughly a quarter before reassembling, an approach Tomer compares to Navy SEALs' cross-training and small-unit model.
Five human traits Tomer says builders should focus on because AI can't yet replace them: vision, empathy, communication, creativity, and — the single most important — judgment ('test-making') under ambiguity; everything else he is actively trying to automate.
Three-pillar framework for the transformation: platform (re-architecting the codebase/design system so AI can reason over it), tools/agents (building and heavily customizing agents), and culture (incentives, motivation, visible examples) — culture is described as the biggest and most important lever.
LinkedIn's biggest learning: no third-party AI tool works 'off the shelf' on its stack — coding agents, design tools, and knowledge tools all required a custom internal integration layer.
LinkedIn built a suite of narrow, single-purpose internal agents: a trust agent (flags safety/scam vulnerabilities in specs), a growth agent (trained on LinkedIn's own growth loops/funnels/past tests, critiques idea quality), a research agent (trained on member personas plus past research/support tickets), an analyst agent (natural-language querying of LinkedIn's data graph), a coding agent, and a maintenance+QA agent pairing that auto-fixes roughly half of failed builds.
Key technical lesson: giving an agent blanket access to a shared drive/knowledge base performs poorly and causes hallucination; the fix is curating a small 'golden examples' set — a lesson Tomer traces to manually curating 'good LinkedIn post' examples over a decade earlier while rebuilding the feed team.
Progress is measured with a formula: (experimentation volume × quality) ÷ time-to-launch; early wins cited include a few hours saved per team per week and designers/PMs picking up bugs directly from Jira and submitting PRs.
Rollout was deliberately grassroots rather than top-down: small pilot pods, opt-in access conditioned on giving feedback, FOMO/exclusivity tactics, all-hands storytelling, and highlighting unconventional career moves (e.g., a user researcher who became a growth PM).
Org structure is shifting from functional leaders (design/PM/BD heads) to product-area leaders evaluated via cross-functional 360 reviews; a formal 'full-stack builder' job title/career ladder now exists alongside a 'system builders' category and a shrinking but real specialist category.
Top talent — not average or struggling performers — has been the fastest and most enthusiastic adopter of the new tools, which Tomer attributes to elite performers' innate drive to keep improving their craft.
Tomer's advice to other companies: invest in platform and tools but put the most effort into culture; be ambitious/impatient about the goal but patient about how you get there; don't expect fast productivity gains (e.g., not 2x in a week); and don't wait for a formal top-down mandate to start.
The episode closes with personal reflections — book influences (Why Nations Fail, Outlive's 'medicine 3.0', The Beginning of Infinity), in-car AI habits/wishes — and Tomer revealing he is leaving LinkedIn after 14 years.
Full-Stack Builder Model — LinkedIn's model empowering builders to take an idea to market end-to-end regardless of their role or team, replacing rigid role sequences with fluid human-AI collaboration. Apply: Structure product teams around cross-functional builders who can move an idea from research through launch themselves, supported by AI agents handling the automatable steps.
Associate Full-Stack Builder (APB) Program — LinkedIn's new junior training program, replacing the APM program and launching in January, that teaches recruits to code, design, and do PM work before joining pods. Apply: Replace narrow entry-level tracks (e.g., an Associate PM program) with a broader builder-training pipeline that produces generalists from day one.
Pods — Small, mission-focused teams of full-stack builders (e.g., an engineer/designer/PM trio) that tackle a problem for about a quarter before reassembling, loosely modeled on Navy SEALs' cross-trained small units. Apply: Reorganize large standing teams into small, cross-trained pods assigned to a specific problem for a fixed period, then re-form pods around new problems.
Time Constant of Change vs. Time Constant of Response — Tomer's framework describing that the pace of external change now outstrips organizations' ability to respond, requiring structural change to close that gap. Apply: Diagnose organizational lag by comparing how fast your market/skills are changing to how fast your team can adapt, then redesign structure and process to shrink that gap.
Platform / Tools / Culture (Three Pillars) — Tomer's framework for at-scale AI transformation: re-architecting the platform for AI, building and customizing tools/agents, and investing heavily in culture and change management. Apply: Sequence an AI-adoption rollout by first making the codebase/platform AI-legible, then building customized (not off-the-shelf) tools, then investing the most effort in cultural change management.
Trust Agent — An internal LinkedIn agent, built by the head of trust, that reviews specs and ideas for vulnerabilities and harm vectors (e.g., scam exposure) before launch. Apply: Build domain-specific review agents staffed with your organization's specialist knowledge (e.g., trust/safety) to automatically flag risks in early-stage specs.
Growth Agent — An agent trained on LinkedIn's own growth loops, funnels, and past experiment history that critiques the quality of an idea, not just its execution. Apply: Feed an agent your team's historical experiment/test data so it can evaluate new idea quality against what has and hasn't worked before.
Research Agent — An agent trained on member personas and LinkedIn's past research and support-ticket corpus, used to critique specs from a specific user's point of view. Apply: Encode your user personas and past research findings into an agent so teams can pressure-test specs against a persona without waiting on the research team.
Analyst Agent — An agent that lets people query LinkedIn's full data graph in natural language, reducing dependence on SQL/data-science teams for some questions. Apply: Build a natural-language query layer over your core data warehouse so non-analysts can self-serve data questions.
Maintenance Agent + QA Agent — Paired agents that automatically diagnose and fix roughly half of failed builds without an engineer intervening immediately. Apply: Automate build-failure triage and remediation so engineers aren't pulled off other work for routine build breaks.
Design Agent(s) — A category of vendor tools (Figma, Subframe, Magic Patterns) being evaluated in parallel rather than settled on a single winner, since different teams gravitate to different tools. Apply: Pilot multiple narrow, single-purpose tools per function and expect to converge to a small set rather than picking one vendor up front.
Orchestrator Layer — A planned coordination layer letting narrow, single-purpose agents (trust, growth, research, etc.) interact with each other instead of being used only in isolation or sequence. Apply: Build and validate narrow single-job agents first, then add an orchestration layer once each agent's individual quality is proven.
Product Jammer Agent / Product Jam Engine — An agent wrapping LinkedIn's internal 'product jam' process that can invisibly call other underlying agents (trust, growth, research) on the user's behalf. Apply: Wrap an existing internal ritual or process in a single user-facing agent that orchestrates specialist agents behind the scenes.
Golden Examples Curation — The practice of hand-curating a small, high-quality set of training examples for an agent instead of giving it blanket access to an entire knowledge base or drive, which causes poor weighting and hallucination. Apply: Before deploying a knowledge agent, manually filter a 'golden' example set of the specific documents that matter rather than pointing it at an entire shared drive.
Benefit Formula (Experimentation Volume × Quality / Time-to-Launch) — Tomer's metric for measuring whether the full-stack builder transformation is working, combining how much and how well teams experiment against how fast they can ship. Apply: Track experimentation volume and quality alongside time-to-launch as a combined KPI for AI-tooling ROI rather than measuring speed alone.
Product-Area Leadership + 360 Review — Replacing functional leaders (design, PM, BD, etc.) with product-area leaders evaluated via 360 reviews across all functions to confirm they can deliver full-stack. Apply: Evaluate leadership candidates for cross-functional roles using peer reviews from every function they'd need to oversee, not just their home discipline.
FOMO / Exclusivity Rollout Tactic — Deliberately limiting early access to new tools and programs, requiring sign-up, so scarcity drives demand and word-of-mouth adoption. Apply: Roll out new internal tools to a limited early-access group first, requiring active feedback, rather than releasing broadly at once.
AI Agency — A trait describing an individual's proactive drive to explore, adopt, and improve AI tools, which Tomer treats as something to encourage and eventually formalize in reviews. Apply: Recognize and reward employees who proactively test and improve new AI tools, and consider building this into hiring and performance criteria.
System Builders vs. Full-Stack Builders vs. Specialists — A talent framework distinguishing people who build underlying systems/platforms (system builders), generalist full-stack builders, and remaining specialists, acknowledging not everyone needs to become full-stack. Apply: When redesigning roles around AI, create explicit buckets (system builders, full-stack builders, specialists) instead of forcing a single universal role change.
Idea-to-Design vs. Code-to-Launch — Tomer's framing of the product lifecycle into two macro stages, noting that code-to-launch had already received more agent investment (coding/maintenance agents) than idea-to-design. Apply: Audit where your AI tooling investment is concentrated across the full lifecycle and deliberately fund the earlier, under-invested stages too.
Why Nations Fail (Extractive vs. Inclusive Institutions) — A book framework Tomer cites arguing that inclusive versus extractive institutions, not culture or resources, explain why nations succeed or fail. Apply: Tomer draws an implicit analogy to organizational building, favoring inclusive, opportunity-oriented systems over extractive ones.
The Beginning of Infinity (Explanatory Knowledge) — David Deutsch's framework, cited by Tomer, that clear causal explanations of why things happen — not just correlation — unlock iterative, near-infinite progress. Apply: Tomer uses it as a personal lens for building: understand root causes deeply before iterating on top of them.
Medicine 3.0 — A concept from Peter Attia's 'Outlive' describing a shift to personalized, preventive medicine, which Tomer expects AI to accelerate. Apply: Cited by Tomer as an area he's optimistic AI will transform, though not something LinkedIn is directly building.
Becoming Is Better Than Being — A personal growth-mindset mantra Tomer applies at home, mapping onto continuous-improvement thinking at work — progress as an ongoing process rather than a fixed end state. Apply: Run an annual self-check comparing this year's version of yourself to last year's to measure personal skill growth rather than treating any state as 'arrived.'
LinkedIn's UXR (user research) team started using the growth agent to prioritize which member-facing ideas had the biggest growth opportunity — an adoption pattern the team hadn't originally designed for.
Retroactively running an old feature spec (the 'Open to Work' green badge) through the newly built trust agent surfaced trust/scam vulnerabilities that weren't caught until well after the feature originally shipped, suggesting the tooling has real retrospective analytical value, not just theoretical promise.
Tomer frames the core problem not as needing to break the existing model but as the model already being broken — the pace of external change is simply what's exposing pre-existing dysfunction in specialized, multi-team, multi-sprint workflows.
Design is described as the hardest craft for AI agents to replicate well, which paradoxically gives designers a 'leg up' cross-training into coding and PM faster than people cross-training into design.
A user researcher applied for a long-open growth-PM role, said 'I feel I can do it,' used the new AI tools to make the leap, and got the job — Tomer now cites this as a career-mobility story used to sell the full-stack model internally.
The single most effective agent-building technique wasn't a smarter model but stricter curation: pointing an agent at a hand-picked 'golden' example set instead of an entire knowledge base or shared drive, since blanket access causes poor weighting of importance and heavy hallucination.
Only a 'cutting-edge ~5%' of staff adopt new tools/behaviors readily by disposition; the remaining ~95% require deliberate incentive design, visible success stories, and structural nudges (FOMO, exclusivity) rather than tool access alone.
The interview doubles as a departure announcement: Tomer reveals mid-episode that he is leaving LinkedIn after 14 years, reframing the entire full-stack-builder initiative as part of his final major push there.
«When we look at the skills required to do your job, by 2030, they will change by 70%.»
— 00:00
«So, whether or not you're looking to change your job, your job is changing.»
— 00:06
«We call it the full-stack builder model. The goal itself is to empower great builders to take their idea and to take it to market, regardless of their role in the stack and which team they're on.»
— 00:24
«It's really a fluid interaction between human and machine.»
— 00:34
«I see a lot of companies roll out their agents and just expecting companies to adopt. Doesn't work this way.»
— 00:51
«Technology has always been about empowerment. It's not about what it does for us, it's about what it enables us to do.»
— 05:37
«You never see success in the launch itself.»
— 08:20
«The work itself is not complex, but the process we made very complex.»
— 08:23
«we haven't seen anybody be able to work off the shelf immediately on our code base, design systems, and unique context we have.»
— 18:22
«I don't think there's going to be a winner takes all.»
— 25:17
«it's not great to just give it access to your drive and say reason all over this knowledge base.»
— 28:26
«the first and most important part was fitting in the right data, not all the data.»
— 29:53
«I could care less about your title. I care about how you work.»
— 45:49
«If you're looking for a formal re-org or declaration to start building differently, you're waiting too long.»
— 46:07
«Failed miserably and hallucinates like crazy.»
— 48:47
«Some people do not want to be full stack builders. And that's completely okay.»
— 49:20
«If you don't invest in your platform, I just don't see how this could be a successful outcome.»
— 54:27
«Becoming is better than being.»
— 63:27
«I will always think of LinkedIn as as one of the babies I helped grow.»
— 66:48
Reception
Most commenters used the video as a springboard to vent frustration at LinkedIn as a product and at AI-driven overwork/hype, with only scattered praise for the content itself.
As a source it's an unusually concrete internal playbook — specific on agent architecture, data-curation practice, and change-management tactics rather than abstract AI hype — but it is delivered by an executive narrating his own company's initiative (and, it turns out, his own departure), so the details map LinkedIn's aspirations and self-reported early wins more than externally verified results.
67:32