How AI is Changing Organizations: What is the 'Full Stack Builder' Model Reinvented by LinkedIn?
"By 2030, 70% of the skills required for current jobs will change."
Tomer Cohen, former CPO of LinkedIn, says this at the beginning of the interview. The point is that regardless of whether you change jobs, the work itself is becoming something else entirely.
Now that the speed of change has begun to exceed the speed of human learning and adaptation, both companies and individuals are forced to rethink the very "common sense of product development" they have relied on. The answer LinkedIn has presented is the Full Stack Builder (FSB) model.
This is not about "abolishing the PM role." Rather, it can be said to be an attempt to dissolve the boundaries between PM, design, and engineering, and create a "new builder image" where AI and humans work as one.
1. Why did LinkedIn rebuild its product development?
1-1. The bloating of processes and organizations has reached its limit
The first problem Cohen identified was not the "difficulty of the work itself," but the "complexity of the processes and organizational structures surrounding it."
In large companies, numerous steps and reviews pile up for a single feature addition. Looking at the research stage alone, it has reached a point where you cannot say you have "investigated enough" without checking 10 to 15 types of information sources, such as user interviews, log data, feedback tickets, and reactions on social media. Furthermore, interactions with specialized teams follow, such as design reviews, privacy reviews, and security reviews.
There are rational reasons for each of these steps, but when viewed as a whole, it has become a massive mechanism that requires multiple teams and multiple sprints just to release a small feature. As a result, a structural problem was born where "it takes an abnormally long time to reach the 'improvement phase' where actual user value is created."
Cohen reflects, "The work itself is not that complicated, but the processes and organizations we have built have become too complex."
1-2. The "shelf life" of skills and job titles has shortened
As a platform with data on job titles and skills from around the world, LinkedIn can detect changes in the labor market faster than anyone else. The fact that "70% of necessary skills will be replaced by 2030" has become visible from that data.
This is not at a level where you can "just update your current skills a little." Even for job titles like engineers, marketers, and recruiters, the content of their daily work is changing fundamentally. Furthermore, data is introduced showing that 70% of jobs currently classified as "fast-growing" were not even on the list last year.
In other words, just thinking about "how to improve efficiency" based on existing job titles and division of labor will not keep up with the speed of change. The awareness of the problem behind the introduction of the FSB model is that it is necessary to redesign the way job functions are divided itself.
2. What is a "Full Stack Builder" in the AI era?
2-1. A builder who handles everything from idea to launch alone (+ AI)
The core of the Full Stack Builder is the idea that "excellent builders should be able to bring ideas to market regardless of their job title." Conventionally, roles passed the baton, with PMs summarizing specifications, designers creating UIs, and engineers implementing them.
In the FSB model, it is assumed that a single builder will use a group of AI agents to carry out research, conceptualization, design, implementation, and launch in an integrated manner. Of course, it is not "complete solo development," but rather working in small pods (small teams), but each member is expected to reach beyond their functional boundaries and touch the entire product creation process.
At this time, Cohen emphasizes the following five core skills that humans should take on: Vision (what kind of future to realize), Empathy (what kind of problems who is facing), Communication (the power to involve others), Creativity (thinking outside the box), and above all, Judgment (the power to decide in ambiguous and complex situations). The philosophy is to leave other tasks to AI as much as possible.
2-2. Shifting to a Navy SEALs-style "platoon organization"
The organizational structure has also been redesigned to match the FSB model. Instead of having "PM teams," "design teams," and "engineering teams" existing in silos as before, small pods formed by mission units become the basic unit.
A single pod is a collection of builders capable of working in a full-stack manner. Every member can write code to some extent, do some design work, handle user research, and use AI agents as a matter of course. Like Navy SEALs, they flexibly change roles according to the situation, complete missions in a short period, and are reorganized into different pods once finished.
Cohen explains that this is not so much about 'eliminating the PM as a profession' as it is about shifting the center of gravity toward increasing the number of 'builders who are responsible for the mission'.
3. The Reality of the 'AI Agent Ecosystem' Built by LinkedIn
3-1. First, Rebuilding into a 'Foundation AI Can Handle'
LinkedIn did not think from the beginning that 'introducing off-the-shelf AI tools as they are would solve everything.' When they actually tried it, it became clear that Copilot and external agent tools could not easily understand LinkedIn's massive and specialized codebase or design system as they were.
Therefore, the company started with 'foundation-side renovations,' such as reviewing code structures and UI component designs so that AI could easily understand the context. Figma and the design system were also reorganized into formats that are easier for AI to interact with.
In other words, a major feature is that they started the transformation not by 'buying tools,' but by rebuilding their own software to be 'AI-friendly'.
3-2. LinkedIn's Proprietary Agents
With the foundation in place, LinkedIn is creating agents one after another that reflect its own unique insights.
For example, the Trust Agent is an agent that has learned LinkedIn's specific 'trust and safety' rules, past incident cases, and policy documents. When you feed it the specifications for a new feature, it points out what kind of abuse risks might exist and which points are vulnerable. It is said that when they actually ran past specifications for the 'Open to Work' feature through the trust agent, it was able to uncover issues that the human team only noticed later.
Growth Agent has learned past experiment logs, funnel data, and the structure of growth loops. When you input an idea or specification, it critiques it from perspectives such as 'what kind of growth impact can be expected' and 'which metrics is it likely to affect.' It is said that the user research team also uses this agent for prioritization when deciding 'which features to focus on'.
Furthermore, they have prepared a group of agents that reflect 'tacit knowledge unique to each job function,' such as a Research Agent that integrates user personas, support tickets, and past research, and an Analyst Agent that can query LinkedIn's massive graph without writing SQL.
Ultimately, they are also advancing a concept for a 'Product Jam Agent' that bundles these together, orchestrating multiple agents behind the scenes from a single entry point.
4. Pilot Results: Who Does AI Empower the Most?
4-1. Reductions of 'Several Hours a Week' and Quality Improvements Are Already Visible
Although the FSB model and the agent group are still in the pre-company-wide rollout stage, pilot teams have already reported 'work reductions of several hours per week'.
PMs have been able to leave initial tasks like research and requirements organization to agents, allowing them to focus on extracting deeper insights. Designers have begun to reach beyond just prototyping into simple code fixes and PR creation. In the engineering domain, a 'maintenance agent' responsible for handling fixes when builds fail has been introduced, and data shows it is automatically repairing about half of the failed builds.
Cohen presents the framework 'Number of experiments × Quality of experiments ÷ Time from idea to launch' as a metric for measuring impact. The current assessment is that with the introduction of agents, both quantity (amount of experiments) and quality (depth of insights) are improving, while the time taken is decreasing.
4-2. Those Who Are Growing the Most Are 'Already Talented People'
What is interesting is that the ones using AI the most effectively are the original top talent. When asked, 'Does AI make average people excellent, or does it make excellent people even more excellent?', Cohen clearly expressed an impression leaning toward the latter.
Those with a 'growth mindset'—who actively try new tools, build their own workflows, and provide feedback to development teams—are the ones reaping the benefits of AI adoption first.
At the same time, this shows the reality that simply introducing AI does not automatically increase the 'average productivity of the entire organization'.
5. The Key to Transformation is 'Culture'—Tool Adoption Alone Will Fail
5-1. Rewriting 'Culture' Through Expectations, Evaluations, and Success Stories
What Cohen repeatedly emphasizes is that 'platforms and tools alone are insufficient; investment in culture is what is decisive.'
At LinkedIn, they first required the management layer to adopt the FSB mindset. Leaders who originally came from PM backgrounds are also receiving 360-degree reviews from the design and engineering sides to measure 'whether they are actually acting in a full-stack manner.' Furthermore, they have incorporated AI literacy and agent utilization skills into hiring and evaluation systems, clearly sending the message that 'mastering AI is a plus for your career.'
At the same time, they are visualizing success stories from limited pods across the entire company, repeatedly sharing stories of 'how this team used agents to increase speed and quality to this extent.' By doing so, they are evoking a healthy form of FOMO (fear of missing out) that makes others think, 'I want to try that too.'
Furthermore, they have ended the traditional APM (Associate Product Manager) program and newly established the APB (Associate Product Builder) program. The policy is to have new graduates and young employees learn coding, design, and PM skills across the board from the moment they join, training them with an FSB-based curriculum to increase the number of 'generations without boundaries from the start.'
6. How Should Companies Adopt It? Three Practical Points
Cohen lists the following three points for companies considering similar transformations.
First is investment in the Platform (foundation). Unless existing codebases, design systems, and data foundations are organized in a way that is easy for AI to handle, no agent can demonstrate its true power.
Second is customization of Tools (agents). It is known that simply giving general-purpose agents full access to Drive or internal wikis does not allow them to understand differences in importance or context, and hallucinations increase. Humans carefully selecting which knowledge to teach as 'gold samples' is the key to quality.
Third, and most importantly, is continuous engagement with Culture. It is necessary to continue sending messages that support 'people who use AI to act full-stack' at every touchpoint, including evaluation systems, awards, training programs, and internal communication.
Conclusion: PMs are not 'ending,' they are 'evolving'
LinkedIn's Full Stack Builder model by no means signifies the 'death of the PM profession.' Rather, it is an attempt to rethink the meaning of job labels like PM, Designer, and Engineer, and to redefine careers based on 'what you can do as a builder.'
AI reintegrates tasks that were divided for the sake of division of labor and expands the definition of 'the power to create.' Within that, what continues to hold value is the human ability to envision, empathize with users, and continue making judgments in complex situations.
To borrow the words Cohen quoted, 'Becoming is better than being'—those who can enjoy the act of changing themselves, rather than remaining in a fixed role, are the ones who can become builders in the AI era.
It is not the end of the product manager, but the beginning of the product builder. LinkedIn's experiment may be anticipating that future image.