Rethinking How We Build: What It Takes to Become AI‑Native at Scale

AI is rewriting the rules of software development. Every day, on my Linkedin Feed and at work, I see engineers racing ahead with new AI tools, shipping products faster than ever. It’s an exciting time to be a software engineer, as the pace of innovation is unprecedented. I can’t recall another time when we’ve delivered this much, this fast, with such quality.

And these tools aren’t just powerful, they’re accessible. Anyone can use them. For startups, possibilities are endless. For a company like LinkedIn, the opportunities are just as big, but it does come with added complexity regarding scale, safety and reliability. 

These AI tools open massive possibilities so ignoring them isn’t an option. We’ve quickly brought them into our business and tested and iterated, before ultimately, embedding AI into every part of our product experience, transforming how our members and customers learn, grow, hire and connect to opportunity. 

EMPOWERING BUILDERS TO INVENT 

This shift started with our culture. It’s fundamental to how we’ve been able to rapidly use this technology and shift how our teams build. We see our culture as our force multiplier and by giving the teams the space, autonomy and right tools, they don’t just build, they invent. It feels like almost every week, we're releasing new product features that would have never been possible to build so quickly just 18 months ago. 

To fuel this innovation, we’ve started to organize cross-functional teams into new pod structures focused on  key initiatives. Pods create shared ownership across multiple disciplines, from product management to design, to engineering and data science, and bring them together into small teams with a clearly defined outcome and timeline to support a business priority. 

One stand out example is our video pod which ran 100+ experiments across three months, from the testing of the video tab, to subtle UX tweaks and even entirely new formats. As we’ve iterated on those, we’re seeing acceleration in members finding value. As an example, the video tab (ramped in US), members are now watching 40% more videos than a year ago — and they go deeper into those videos too: members are now spending 2.2x more time watching videos in the tab. This model has unlocked a new scale of experimentation and faster iteration cycles. Teams can form a hypothesis, test it, and see members engage in a matter of a few hours and then spin up the next few iterations of their feature based on what they saw. 

We’re also elevating a talent archetype: the Full Stack Builder. A Full Stack Builder owns the product experience end-to-end, working fluidly with AI as a teammate across the lifecycle. This removes strict functional boundaries so builders can move from insight to impact faster. 

REBUILDING FOR THE AI ERA 

But culture alone isn’t enough. Our engineers need the right foundations to build. Modernizing how we build isn’t just about adding AI features, it’s about rethinking how engineers collaborate to scale AI-powered tools fast.  

Like many mature platforms, our apps and service infrastructure weren’t built for today’s AI. Complexity accumulates, codebases tangle, and the “map” is such that AI tools that work for smaller companies don’t work out of the box for us. To fix this problem we’re adapting our technology stack to meet the opportunities with AI to build products that we previously thought were unimaginable. 

For example, we’re rewriting our user facing applications based on simple, composable UX components—clean interfaces that both engineers and LLMs can understand. This both raises code clarity and amplifies AI usefulness, because LLMs thrive on well‑structured building blocks. Early signs are promising: teams are validating and iterating features much faster than before, with the capacity to run up to 7 ramps a week, compared to 1-2 previously. 

The centerpiece of this modernization is a server-driven UI framework (SDUI), which decouples UI logic from client applications and shifts it to the server. This enables dynamic updates and faster iteration and deployment across many platforms. This allows LinkedIn to build once and deploy everywhere, reducing duplicate work and delivering a consistent, high quality experience for our members and customers.   

Better tools empower our engineers to iterate and develop at speed and scale, unlocking the freedom to focus on what matters most - experimentation. To help with this, we’re also arming our engineers with the right tools.  Internally we’ve introduced coding assistants and agents across the dev lifecycle. One of our engineer assistants now auto‑resolves ~35% of build failures, reducing toil and keeping momentum high. While our engineers are also using external tools such as Copilot to accelerate development. 

I've been building products for over 20 years, 16 of those at LinkedIn, and I've never seen us build with such purpose and speed. Even when LinkedIn was still a relatively small startup, hustling every day to improve our product day by day, nothing compares to what I am seeing now at 100x the scale. What an exciting time to be building!