Systems Thinkers Over Specialists: Netflix CPTO Elizabeth Stone on Talent, AI and Excellence as an Operating System

Lenny Rachitsky opens by noting that AI lets PMs ship code, designers write PRDs and engineers do product work, which leaves many people wondering what their job is. Stone says she hears the same confusion inside Netflix. Her view is that any transformative technology brings a storming phase before the forming phase, and that the answer is not to put AI back in the box.

She welcomes experimentation, such as prototyping an idea quickly or writing initial code to test it, but doubts that everyone should be shipping code to production. When the business problem is clear, she says, some fluidity is healthy, since product and design can move before engineering is free. They should still work with engineering on how to productize, scale and put guardrails around the idea.

To get the benefits rather than the costs, Stone points to organizational prerequisites: clear source-of-truth data, guardrails or testing before large changes ship, a view on where AI output can be trusted and where review is needed, and a reminder that humans remain responsible for what they create. An agent may have written the code, she says, but accountability stays with the person.

Over the past two and a half years, Stone says, PMs, designers and data scientists can get much further in the product cycle before engineering needs to step in. She adds a caution: prototypes should aim at an important business problem and engineering partners should know about them, rather than throwing spaghetti at the wall.

A second shift is access to Netflix's decades of experiments, consumer insights and stakeholder input. AI is good at distilling and analyzing that material, and she can now find what research was done and what was tested almost instantly instead of interrupting a colleague. She treats this as a head start rather than something to rely on exclusively, and says business stakeholders in finance, content and advertising can now bring initial hypotheses to data scientists too.

Asked whether specialist functions will survive, Stone says yes. Data scientists remain experts on whether data can be trusted and interpreted correctly, product managers on framing the right problem, and engineers on how things scale and what quality looks like. People now speak more languages thanks to AI tools, she says, but she still finds great engineering, data science and creativity scarce.

Asked what Netflix is hiring more of, Stone says it needs more systems thinkers, which looks different by function. In core infrastructure, Netflix's past success came from local teams building the stack they needed, but in a world of agents working across systems and wanting source-of-truth data, common paved paths and solving problems once become more important.

That means engineering profiles skew more toward distributed systems and infrastructure, while still keeping deep expertise in areas like personalization, advertising and content delivery. Her phrase is that what got Netflix here doesn't get Netflix there. The same pattern shows in design, where the experience team builds templates and systems so non-designers can build coherent products and Netflix avoids shipping "Frankensteins."

Stone says the reasons are velocity and scaffolding. Platforms give most teams an 80% head start, and Netflix is making more bets across the business. With more people doing more types of work, questions about access, identity, security and quality can't be left to each builder, and an organization of thousands can no longer rely on tribal knowledge. Her vision is many agents contributing, with humans guiding the problem and judging the output.

Lenny cites a guest's thesis that the design process is dead because there's no time for it. Stone has mixed feelings. Infrastructure and systems thinking should let more people do well-designed work, but for the most important priorities design is critical, and she thinks squeezing out design expertise just because code and analysis are faster would be a mistake.

For a large consumer product like Netflix, she says, product, technology and design make complexity invisible and produce a seamless experience, which is a design mindset that has to stay core. The work may look different, and designers can show more options and iterate faster, but the mindset remains.

On skills trending down, she names very narrow, deep specialization. She still sees a need for rare experts, such as people who understand Netflix's encoding or playback systems, but compared to five or ten years ago she expects fewer specialists and more people who adapt across back end and front end, functions or infrastructure. A specialist is fine, she says, as long as they're willing to grow and question whether the tool is still the right one.

Asked how people develop systems thinking, Stone offers a small trick: for each problem, step out one click and ask what you are assuming about the broader space. If building a feature, think about the larger consumer problem, whether it will scale across content types, or whether it should become a platform capability. You don't need to solve company strategy, she says, and shouldn't stay in the questioning state so long that you stall.

Another route is to look at your work from your manager's view, thinking about finance, content and other parts of the business. She adds a related principle from Netflix's engineering culture: do what is right for the broader organization rather than only what is right locally, and leave systems better for colleagues and future innovations.

Rather than rewriting each career-ladder level for AI, Netflix put an overlay of AI fluency across everyone. It varies by function and career stage, and the definition evolves almost by the quarter. Stone says it isn't using tech for tech's sake but having good judgment about where it helps and an open mind to try, including for senior leaders, and hiring now explores how candidates use AI and allows AI tools even in coding interviews.

Stone says the use she personally relies on most is data analysis and distillation: asking about past experiments, metrics to watch and consumer research, with much higher velocity, provided results are validated with a local data scientist and source-of-truth data. This goes beyond prototyping to translating data into action and insight.

The second area is content creation. ML was already used in production tools, promotional assets, subtitles and dubs, and generative AI is a step function for creative ideation such as pre-visualization and for post-production. She mentions Netflix's acquisition of a company started by Ben Affleck whose models can relight, reframe or change dialogue after shooting, always led by the filmmaker.

Lenny reminds listeners of the Netflix Prize as an example of how early the company was to ML. Stone says personalization has long been central, and with a growing catalog that includes games, live and podcasts it becomes more important. She notes that AI is a tool Netflix is comfortable using even though it isn't branded an AI company.

Lenny observes that Netflix's early culture of high agency, autonomy, talent density and top-of-market pay resembles how top AI labs operate. Stone describes the goal behind these elements as excellence as an operating system: giving people agency and accountability, pushing decisions deep, and hiring people with good judgment, which she says drives outcomes and motivation.

She calls talent density the non-negotiable. Beyond that, Netflix is comfortable with risk, recovering quickly from failures rather than avoiding them, and she cites the move into live events as an example of accepting imperfection to learn fast. There is also a selflessness: decisions are about Netflix and its members, not personal preference.

Stone says some of this is unnatural. She often sees decisions she would make differently, but her job is not to overrule unless the stakes are material, and she asks for reflections afterward. When things go badly she also resists the instinct to add process, since in her experience more planning and leveling process has not produced better outcomes. She prefers blameless retrospectives and people who ask how they can do the work differently next time.

Stone says the keeper test is often thought of as the moment you let someone go, but it's also used to say how extraordinary someone is. A direct report can ask how they are doing on it, and most of the time her answer is that she would fight hard to keep them, followed by strengths and areas to improve.

Sometimes the answer is hard: the person is not passing right now, with either milestones to get there or no visible path. She sees it as good feedback hygiene. Lenny summarizes the idea as asking whether you would hire the person today, and Stone adds the corollary of whether you would fight to keep them if they said they were leaving. She also lists context over control and being highly aligned but loosely coupled as things organizations tend not to do at scale.

On competing with frontier labs for talent, Stone says she doesn't feel other companies are vacuuming up the best people, because many are at Netflix. But the company must be explicit about the people who thrive there: those passionate about applying technology to entertainment and consumer products at global scale. The frontier labs, she says, draw a different persona more interested in the foundational work.

Asked whether junior people can still become seniors if AI does the work, Stone says Netflix still hires them. It runs an intern program and a new-grad program that began a few years ago, because younger people tend to be more open-minded, native to new ways of working and fluent in how entertainment and consumer behavior are changing.

She says mastery of the craft remains important: people are responsible for the quality of the code they submit and the products they build, so mentorship on what good looks like matters as much as ever. Training will need to change because juniors use different tools, and she expects early-career talent to teach older colleagues new things too.

On engineering's future, Stone distinguishes writing code in a language from understanding how code, systems and products work, and says the latter isn't going away. If agents write all the code, she argues, you can't judge whether a product is good or fix it when it breaks. She admits that some agent-written code is hard to follow and unsettling, but expects engineers to become fluent in guiding it, and notes that engineering has shifted dramatically before.

Stone says entertainment is already not one thing, which is why Netflix is going beyond film and TV into mobile, cloud games, live content and podcasts. She describes a seamless journey, such as from a Bill Simmons podcast to the *Quarterback* series to playing a FIFA cloud game on TV or phone, and says the product challenge is making discovery and engagement easier across so much content.

On AI in Hollywood, she says there is a wide array of views. Some creators want no AI at all, and Netflix works with them; a growing number want to explore what generative tools make possible; and Netflix supports everyone in between. Its role is creator enablement rather than a prescriptive single way of doing things.

Asked whether whole shows might be AI-made and loved, Stone says she has a hard time picturing entertainment without humans at the heart of it, since storytelling and humanity are one and the same. AI will play a material part in how some productions look and feel, but she doesn't see a version without the human as the backbone, and she closes by saying it is an exciting time to build consumer products people love.