Applied / Research Scientist, ML Engineer: What’s the Difference?
Eugene Yan
Data science splintered into four jobs: data scientist, applied scientist, research scientist, ML engineer. They sound similar but do very different work, and titles are getting inflated to win hiring wars.
Based on reporting by Eugene Yan — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Remember when "data scientist" was crowned the sexiest job of the 21st century? That was 2012. By 2020, the label had fractured into a whole taxonomy of roles, and nobody outside the field can keep them straight anymore.
Eugene Yan traces how this happened, and the story is less about new skills emerging and more about titles chasing prestige. Facebook's product-focused data scientists were basically doing analyst work, so the company spun up a separate Core Data Science group for research. Lyft went further in 2018, flipping the labels entirely: their old data analysts became data scientists, and their old data scientists became research scientists. Why? Because Lyft was losing analytics candidates to competitors who dangled the shinier "data scientist" title. Zero cost, real hiring edge.
Once you strip away the branding, the actual work breaks into four lanes. Data scientists run analysis and A/B tests to guide business decisions, mostly SQL and Python, with dashboards as the deliverable. Applied scientists build ML systems, like fraud detection or recommendation engines, that ship into production. Research scientists chase fundamental problems, model compression, speech-to-text, image segmentation, publishing papers on benchmarks like ImageNet or GLUE, usually only at companies with the budget for multi-year bets. And machine learning engineers build the infrastructure, think Uber's Michelangelo or Netflix's Metaflow, using more enterprise-grade languages like Java and Go, since they're really software engineers with an ML lean.
But Yan flags a real cost to all this specialization: when scientists prototype in Jupyter notebooks and hand off to MLEs for productionizing, context gets lost and nobody owns the whole pipeline end to end. He also worries about a title arms race, more companies inventing grander names like "AI engineer" just to win talent, which only muddies what people actually do all day. This isn't unprecedented, though. Statisticians got rebranded as data scientists a decade ago, and software engineers already splintered into frontend, backend, mobile, devops. Titles chase status; the underlying work evolves more slowly.
My take — AI-written commentary, not fact-checked reporting
Titles are a hiring gimmick dressed up as an org chart, and Lyft admitting they rebranded analysts as scientists just to win candidates should embarrass the whole industry more than it does. I'd rather see companies post the actual deliverables and skip the cosplay, because the AI hiring market is already frothy enough without everyone inventing "AI engineer" as this decade's costume change.
Read more about this at: Eugene Yan