The Data Scientist Role Is Splitting Into Three Jobs. Here's Which One to Become.

For most of the last decade, "data scientist" was three jobs wearing one coat. You cleaned and modelled the data, you built the thing that predicted something, and then you were somehow also responsible for getting it to run in production. One title, one salary band, three genuinely different skill sets.

That coat is coming apart. And if you are deciding what to learn this year, the most useful thing you can know is which of the pieces you are actually training for, because the market has stopped hiring the generalist and started hiring the specialists.

Why the coat is coming apart now

The trigger is foundation models. Before 2023, if a company wanted a model that did something specific, someone had to build it: gather the data, engineer the features, train and tune. That someone was the data scientist, and the bundle made sense because the whole pipeline was one person's problem.

Foundation models collapsed the custom-model path for a large share of use cases. If a capable model already exists behind an API, the work is no longer "train a model." It splits into two very different jobs: the person who figures out what the business should do with data, and the person who wires a model into a product and keeps it alive. Those two people want opposite things. One lives in questions and statistics. The other lives in latency, endpoints, and uptime. Asking one human to be excellent at both was always a stretch, and the market has decided it no longer needs to.

The three roles the bundle split into

Taxonomies vary, and anyone who tells you the boundaries are crisp is selling a bootcamp. But across the 2026 job market, the split lands roughly here.

The analyst (and analytics engineer). This is the "what should we do" half. Business questions, SQL, experiment design, decision support, the dashboards and models that change what a team does on Monday. It is the most underrated of the three and, in most organisations, the one closest to the money.

The ML engineer. This is the "ship it and keep it shipped" half. Model serving, retraining pipelines, monitoring, cost ceilings, the infrastructure that turns a notebook into a system. The Stanford AI Index 2026 found that the fastest long-term growth in AI job postings came from deployment-flavoured skills: cloud platforms, scalability, and workflow management. Deployment is where the durable demand is.

The AI engineer. This is the new one, the job foundation models created. Building products on top of LLMs: retrieval pipelines, agents, evaluation harnesses. LinkedIn's most common listed skills for the role are LangChain, retrieval-augmented generation, and PyTorch. It is a software-engineering job first and a modelling job second.

And what is left of "data scientist" proper? It narrows to what only it can do: rigorous experimentation, causal inference, statistical modelling, the work that tells you whether an effect is real. That is a smaller, deeper role than the old bundle, and a genuinely valuable one.

The data looks contradictory. It is not.

Here is where you have to read two numbers at once.

The US Bureau of Labor Statistics projects employment of data scientists to grow 34% between 2024 and 2034, from 245,900 jobs to 328,300, roughly 23,400 openings a year, making it the fourth fastest-growing occupation in the entire economy. On that number, the title is thriving.

But LinkedIn's 2026 Jobs on the Rise report names AI Engineer the single fastest-growing role in the United States, with postings up around 143% year over year, and four of its top five fastest-growing jobs tied to AI. Meanwhile the Stanford AI Index reports that employment for developers aged 22 to 25 has fallen nearly 20% since 2024, and US computer-science enrolment is down about 11%.

So the growth is real, and the re-sorting is real, and the thing getting squeezed is the exact profile most beginners are told to aim for: a bit of everything, specialist in nothing.

What to actually do

Pick your one. Not forever, but for now.

If you like the questions more than the code, become the analyst who can prove what works, and learn enough experiment design and causal reasoning that people trust your answers. If you like systems, become the ML engineer who can deploy, because the postings keep asking for deployment and will keep asking. If you like building products, become the AI engineer, and go deep on retrieval, agents, and evaluation rather than collecting another model architecture.

What you should not do is spend another year being a generalist because a 2019 roadmap told you to. That roadmap was written for the coat, and the coat is gone.

What would make me wrong

The titles are a mess, and I would rather say so than pretend otherwise. LinkedIn explicitly groups "AI engineer" and "machine learning engineer" together, so that #1 ranking blurs two of my three roles into one. Different sources draw the analyst/scientist/engineer boundaries in different places, and a role called "data scientist" at one company is an analyst at the next and an ML engineer at a third. Postings are demand signals, not hires. And the BLS projection runs to 2034 on a category defined before this split accelerated, so treat it as evidence the field is growing, not as proof of where inside the field the jobs land.

None of that changes the advice. Whichever way the labels settle, the market is paying for depth in one direction, not shallowness in all of them.

Key takeaways

1. The data-scientist bundle is unbundling into the analyst, the ML engineer, and the AI engineer, with a narrower experimentation-focused "data scientist" alongside.
2. Foundation models caused it by collapsing the custom-model path, splitting the "what should we do" work from the "ship it" work.
3. BLS projects 34% growth for data scientists to 2034, but that broad bucket hides the re-sorting underneath.
4. The generalist entry seat is the one shrinking, not data work overall: junior developer employment is down about 20% since 2024.
5. Pick your one, unless your market is too thin to specialise, in which case being the whole coat is your advantage.

Which of the three are you closest to right now, and which one do you actually want? Those are often not the same answer, and the gap between them is your next year of learning.

---

Sources: US Bureau of Labor Statistics, Occupational Outlook Handbook, Data Scientists (bls.gov/ooh/math/data-scientists.htm), 2024 to 2034 projections. LinkedIn Jobs on the Rise 2026 (fastest-growing US roles). Stanford HAI, 2026 AI Index Report, Chapter 4 labour-market data (hai.stanford.edu/ai-index/2026-ai-index-report).