In our latest Data Day livestream, my guest Ray Mi and I explored a concept that's becoming central to how we think about AI adoption: what it means to be "AI native."

Ray, a data scientist with experience at DataRobot and Snorkel AI, drew a distinction between people and companies who grew up with AI as a given versus those adapting to it mid-stream.

Born Into It

"Kids are growing up in this environment. They know how to deal with it," Ray said. "It's similar to companies. Companies born from 2022 onwards are much more AI native, cloud native, digital native. They don't need to figure out how to pivot from traditional ML to an AI-native business model. They were born and raised in that era."

For established enterprises, the shift is harder. Ray compared them to heavy ships that struggle to change direction. They carry legacy infrastructure, existing workflows, and organizational habits that make transformation slow and costly.

The numbers bear this out: Ray cited industry research showing that while 95% of companies are adopting generative AI or agentic systems, only 13-14% report seeing tangible returns. That gap between adoption and value creation is where AI-native companies have an advantage. They're building with AI as a foundation rather than retrofitting it onto existing systems.

The Education Question

Our conversation touched on what AI-native learning might look like. Ray referenced a recent keynote by Fei-Fei Li at AI4, where she discussed how AI could transform education by reducing reliance on rote memorization. Instead of spending years memorizing facts, students could focus on higher-order thinking, using AI to handle recall and retrieval.

I offered a counterpoint from my own household: my daughter uses Chrome's built-in Gemini to solve math problems without working through them herself. The tool is so seamlessly integrated that she doesn't even leave the browser where she's doing homework.

This raised a question neither of us were able to fully answer: if AI-native students don't learn to do things the hard way first, does that matter? Or is "the hard way" just an artifact of pre-AI constraints?

Humans as Orchestrators

Ray's view on the future landed somewhere optimistic. Even as AI handles more execution, humans remain essential as the initiators and orchestrators.

"I am the agent," he said. "The other tools are there, but I'm the one that drives everything and orchestrates all those steps together. The motivation is the key. What to prompt, what you ask the AI to do is still the most human part."

For data scientists specifically, Ray sees the role evolving toward architecture and evaluation. Agentic systems involve many components working together: tool calls, user interactions, task routing. Each step's accuracy affects the final output. Designing those systems and measuring their performance still requires statistical thinking and scientific rigor.

"How those tasks are stitched together still requires some scientific mind to architect this process," Ray said. "You need a system that can work not only in the testing environment but at scale in production."

The implication: being AI native doesn't mean knowing less. It means knowing different things, with a focus on orchestration, evaluation, and knowing which tools to call for which problems.

Data Day is Zerve's livestream series featuring conversations with practitioners and leaders in data science and AI. Watch previous episodes here.