Data scientist in the age of LLMs: is it still a good career?

Yes, data science is still a good career, but a narrower one. LLMs now draft the SQL, the boilerplate code and the first chart, so the generalist whose output is mostly reports and dashboards is squeezed. Data scientists who own a decision (pricing, experiments, measurement) or who measure AI itself are still hired, including by AI companies.

That is our answer on the future of a data scientist career. Below: the market data, and what the 43 data scientist ads on Tokn asked for as of 14 October 2026. For other paths, see our guide to careers in AI.

In short

• Still a good career if you own a decision or measure AI. The report-and-dashboard generalist is the most exposed.
• Mixed signals, not doom: the US BLS projects +35% for data scientists over 2025–35, while data and analytics postings sit well below 2020 levels in the UK and Germany.
• Of 38 data scientist ads we read in full, 23 had no AI in the role. Only 4 asked the data scientist to build GenAI tools, and 5 to evaluate AI models or agents.
• All 8 junior data scientist ads on Tokn were at three companies: Shift Technology, Artefact and Ekimetrics.

The data scientist career future: what the market says

The long-term US outlook is strong; short-term hiring for data and analytics roles is weak. Both are true because they measure different things.

The US Bureau of Labor Statistics projects data scientist employment to grow 35% from 2025 to 2035, from 275,600 jobs, with a median pay of $120,230 (May 2025). It names AI as a reason: data scientists "will be needed to help businesses leverage AI". Europe has no equivalent series.

In Indeed's Job Postings Index (1 February 2020 = 100), the broad "data & analytics" sector stood at 51 in the UK, 75 in Germany and 63 in the US on 2 October 2026. The UK and German values were down on a year earlier (53 and 87); the US one was up from 59 but still the lowest of 47 US sectors. The category also covers analysts and data engineers, and it tracks a hiring cycle, not the size of the profession.

Read together: fewer openings now, a role that is not disappearing, and pickier employers (see the European AI job market in October 2026).

What Tokn's data scientist ads ask for

Mostly classic data science. Of the 38 ads we read in full, 23 had no AI work in the role, and where AI did appear it was more often about measuring it than building it.

First, what Tokn covers: practically every data scientist title at the AI and tech companies it tracks, but few analyst roles and nothing from the banks, insurers and retailers that employ most data scientists. As of 14 October 2026 that gave 43 data scientist ads (39 after removing multi-city duplicates) at 15 companies, about 5% of the roughly 800 jobs on Tokn. Two consultancies, Artefact and Ekimetrics, posted 15 of them (35%). So this is data science at AI companies, not the whole market.

We coded each ad by what the role itself does with AI, ignoring company boilerplate:

The new niche is measuring AI, not prompting it. Only one ad, Lovable's Data Scientist, Agent, explicitly asked for LLM evaluation: "building evals, scoring outputs, tracing agent behavior, and catching regressions."

The skills are less exotic. Share of the 38 ads mentioning each theme:

* Broad keywords: read as direction. Shares are higher than in the AI skills employers ask for most, which used shorter text and included analysts.

Will AI replace data scientists?

No, it replaces tasks: SQL drafts, boilerplate code, first-pass exploration, charts. On hard, multi-step analysis, agents still fail most of the time unless someone has defined what the business metrics mean.

DABstep, a benchmark Adyen and Hugging Face built from real payment-operations tasks, found that the best agent at launch (June 2025) solved about 14% of its hard tasks. A September 2026 preprint on DABstep, single-author and not peer-reviewed, tested what helps. Adding governed business definitions raised hard-task accuracy from 14–37% to 55–77% across four models, while retrieval scaffolding alone added 0 to 15 points. If that holds, data agents get good once someone encodes the semantics: data scientist and analytics engineer work.

Employers now write this into ads. Legora described "an AI-first data team", and MongoDB expected its data scientists to "review, understand, and be able to redesign any code that ships regardless of whether a human or an AI wrote the first draft."

Which kinds of data science hold up

Work that owns a decision or measures AI. The 38 ads by type of work:

Experimentation and causal inference appeared in 50% and 21% of ads. An LLM can run the regression; someone still has to design the holdout and judge whether the effect is causal. Measuring AI is the other strong line: Wayve's role was to work out "when data-quality, annotation and evaluation signals can be trusted".

Which kinds are squeezed

The generalist whose output is a dashboard, a one-off SQL ticket or a notebook nobody else runs. This is our reading of ads and benchmarks, not proof of job losses.

Indeed's UK mid-year report (August 2026) found that data & analytics had the highest share of UK postings mentioning AI (48.8%), in a sector whose postings index sat at about half its 2020 level. Ads were blunt. Shift Technology: "writing scripts for models and notebooks is not enough". Legora: "Insights are useful; impact is what we hire for."

Data scientist vs ML engineer, and the other new titles

Part of what used to be called data science now sits under other titles. On Tokn, ML and AI engineer titles (102 ads) outnumbered data scientist titles (43) about 2.4 to 1 as of 14 October 2026.

For production systems, read how to become a machine learning engineer or what an LLM engineer does; for research, research scientist vs research engineer.

Who hires data scientists in Europe

On Tokn: consultancies, AI product companies and consumer tech. Paris (14 ads), London (11) and Stockholm (6) led as of 14 October 2026.

Eight of the 43 ads were over a year old, mostly standing pipeline ads at consultancies. The US labs kept data science at home: OpenAI (18 data scientist roles), Stripe (8), Anthropic (6), Databricks (6) and Figma (4) listed none in Europe, while Mistral AI, Lovable, ElevenLabs, Legora and Wayve did.

Pay, examples only: Tokn read a salary on 5 of the 43 ads, all at consultancies: €60–85k in Paris and £60–90k in London, senior to manager. More in AI salaries in Europe.

Can you still start as a junior data scientist?

Yes, but the doors are few. On Tokn, 8 of 43 data scientist ads were junior (19%, against 9% board-wide), all at Shift Technology, Artefact and Ekimetrics.

The wider signals are cautious. Indeed counted UK graduate postings down about 7% year on year in July 2026, the lowest for the time of year since 2020. In the US, Stanford's "Canaries in the Coal Mine?" paper (revised November 2025) found a 16% relative employment decline for 22 to 25-year-olds in the most AI-exposed occupations, not data scientists specifically.

What to do if you're a data scientist today

1. Own one decision area end to end: pricing, growth or experimentation.
2. Get serious about experiment design and causal inference, the skills an LLM cannot vouch for.
3. Learn to evaluate AI: build evals, score outputs, trace agents.
4. Model the semantics: dbt, metrics layers, business definitions that AI tools can consume.
5. Ship code others can run, and review AI-written code as hard as your own (MongoDB, Shift).
6. Use AI tools daily for the boring parts, and say so in interviews.
7. Pick a direction: deeper data science (decision science, evals), sideways to analytics engineering, or over to ML or AI engineering.

Find data science jobs on Tokn

Browse data science jobs and set up a job alert to get new data scientist roles by email.

FAQ

Is data science dying?

No. The BLS still projects 35% growth for US data scientists over 2025–35. Data and analytics postings are weak in the UK and Germany, and the report-and-dashboard generalist is exposed, but employers still hire data scientists who own decisions or measure AI.

Will AI replace data scientists?

It replaces tasks: SQL drafts, boilerplate code, first-pass analysis. On the DABstep benchmark, the best agent solved about 14% of hard tasks at launch (2025), and a 2026 preprint suggests agents improve sharply once humans define the business logic. That logic is data scientist work.

Data scientist vs ML engineer: which is safer?

On Tokn, ML and AI engineer titles outnumbered data scientist titles about 2.4 to 1 as of 14 October 2026. Data scientist roles lean on statistics, experimentation and stakeholders; ML engineer roles on production systems. No title is safe on its own: pick the work that fits your strengths.

What skills should a data scientist learn in 2026?

Experimentation and causal inference, dbt-style data modelling, evaluating AI and LLM outputs, and enough engineering to ship. Statistics, Python and SQL remain the base: each appeared in 68% to 82% of the ads we read.

Do AI companies hire data scientists in Europe?

Yes. Lovable, Legora, ElevenLabs, Wayve and Mistral AI listed data scientist roles in Europe as of 14 October 2026. OpenAI and Anthropic listed theirs only in the US.