Robots Are Better Than People home
Subscribe

Will AI Replace Data scientists?

Microsoft's research ranks data scientists among the occupations most exposed to AI, and vendors now ship agents that write and run analysis code from a single prompt. The same government data that tracks this work projects one of the fastest growth rates of any job.

Our readGrowing; agents take the grunt workWe think agents are taking over setup, boilerplate code and routine queries, while BLS projects strong growth and benchmarks show agents still fail most hard multi-step analysis.

Data scientists face two signals that look like they should cancel out. A Microsoft Copilot study, as summarized by Extern, ranks data scientists 19th of 785 occupations for AI applicability. Over the same period, the U.S. Bureau of Labor Statistics projects employment rising from 275,600 jobs in 2025 to 371,000 in 2035. That is a +34.6% change against 3.5% for all occupations.

Both are true. AI is aimed squarely at the daily tasks of this job, and employers still expect to hire many more people to do it. The BLS goes further and names AI as a driver. It writes that "Adoption of AI technologies, including generative AI tools, is expected to fuel strong job growth among computer and mathematical occupations".

The practical question for anyone in the field is which parts of the work are moving to agents and which parts are becoming more valuable.

The numbers behind data science jobs

The BLS projections show 24,800 openings a year on average, a $120,230 median annual wage in 2025, and a bachelor's degree as the typical entry education. A recruiting guide from Pin cites the same $120,230 BLS median.

In its earlier 2024 to 2034 projections, the BLS used data scientists as its showcase example, projecting a 33.5-percent increase in employment over that decade.

The postings picture is less tidy. Extern reports that live data scientist postings rose from July to October 2026, while a broad postings index for math and data roles remains well below 2020.

Where data science agents are already running notebooks and queries

This is not a forecast. Several large companies already ship or run agents that do recognizable data science tasks.

  • Google, Colab. Google's Developers Blog says that in December it shared how its Data Science Agent in Colab creates notebooks, then opened it to Colab users age 18+ in select countries and languages. Google's own page warns: "Data Science Agent may make mistakes." ITPro reported that Google cited a scientist at Lawrence Berkeley National Laboratory, working on a global tropical wetland methane emissions project, who estimated time savings from the agent. That is a vendor-supplied example, not an independent test.
  • Databricks. In 2025 Databricks introduced a Data Science Agent for Databricks Assistant, calling it "an autonomous partner that plans, executes, and self-corrects" workflows. Its community announcement says it "can explore data, generate and run code, and fix errors, all from a single prompt. This can cut hours of work to minutes." Forrester analyst Charlie Dai told InfoWorld that the agent upgrades the Assistant beyond a code-generation helper. Databricks later made the Assistant available in AWS GovCloud, pitching it as a replacement for "spending days stitching together experiments, tuning models, and iterating on code".
  • OpenAI, internally. OpenAI built an internal data agent to serve over 3,500 employees across Engineering and other functions, according to a ZenML case-study summary. VentureBeat called it "one of the most aggressive deployments of an AI data agent" and reported that OpenAI has no plans to sell it. A Medium post describing it says "Employees across functions access nuanced analysis that previously required data team support."
  • Product managers at Microsoft. In a Microsoft study of 885 PMs, 12% said they used GenAI for prototyping and coding and could now do tasks they used to rely on data science colleagues to complete.

What we did not find: a documented layoff of data scientists specifically attributed to AI. The closest case is company-wide. Interview Query reports that in early March 2026, Block CEO Jack Dorsey announced cuts of nearly 4,000 employees, about 40% of the global workforce, citing AI.

Data science tasks, split between agent and analyst

TaskWhat AI does todayWhat still needs a person
Notebook setup and boilerplateColab's agent creates notebooks from a promptChecking that the right data was loaded and joined
Exploratory analysisDatabricks' agent explores data and runs codeDeciding which patterns matter for the business question
Debugging analysis codeAgents fix errors and self-correct inside notebooksCatching errors that run cleanly but give wrong answers
Ad hoc questions from colleaguesOpenAI's internal agent answers queries for over 3,500 employeesDefining metrics and data that the agent can trust
Single-step lookupso4-mini reached 76.39% accuracy on easier benchmark tasksSpot-checking the remaining misses
Multi-step reasoning over messy dataTop agents scored only 14.55% on Hard tasksMost of the work, for now
Ethics and trustworthiness reviewNo documented agent roleApplying ASA principles on trustworthy AI

What to do next, by who you are

Skills that gain value as agents write the code

Multi-step problem framing. The DABstep authors write that "tasks requiring multiple steps of reasoning remain largely unsolved." Google Research found its DS-STAR agent needs more refinement on hard work: "Hard tasks average 5.6 rounds versus 3.0 for easy tasks". Breaking a vague business question into checkable steps is where people still lead.

Deep statistical specialization. Microsoft's report states that "deep specialization remains critical: data scientists can't do marketers' work, and marketers can't do data scientists' work."

Validation and ethics. The American Statistical Association issued its Statement on Ethical AI Principles for Statistical Practitioners, 10/8/24, which names trustworthiness as the overarching objective. Its ethical guidelines apply to everyone doing statistical practice regardless of job title, which covers people now running analyses through agents.

Data platform design. Internal agents like OpenAI's depend on well-built datasets and definitions. Someone has to build and maintain them.

The messy middle

Exposure scores disagree. Extern cites an Anthropic observed exposure score of 0.46, while another analysis lists data scientists at about 38%. We could not confirm either figure in Anthropic's own pages, so treat both with caution.

Microsoft's applicability paper gives data scientists a score row of 0.77, 0.86, 0.51 and 0.36 across its measures, on an employment base of 192,710, as reproduced by Cloud Wars. The researchers themselves wrote, "Our data do not indicate that AI is performing all of the work activities of any one occupation." R&D World notes that data analysis tasks received among the lowest user satisfaction scores in that study.

Reliability is the open problem. Google warns its agent may make mistakes, and the VentureBeat reporter who tested it described a bumpy first experience. An analysis that runs without errors can still be wrong, and nobody in this evidence set has shown how often agents produce confident, plausible, incorrect results in production.

The quiet shift may matter more than layoffs. When 12% of surveyed PMs say they no longer need data scientists for some tasks, demand moves without anyone being fired.

Our verdict: the job is shifting toward judgment, not disappearing

We think AI will not replace data scientists in the coming decade, based on this evidence. Agents from Google, Databricks and OpenAI already handle setup, routine code and simple questions, and that work will keep shrinking. The hard, multi-step analysis that benchmarks show agents mostly fail is the core of the senior job, and the BLS projects strong growth with AI cited as a cause. Our main worry is the entry rung, where Microsoft's findings point to slower junior hiring.

The bottom line: In our read, data scientists who can frame hard problems, validate agent output and own data quality should see rising demand, while roles built mostly on writing routine queries face real pressure.

Questions people ask

Will AI replace data scientists?

We think not in the coming decade. The BLS projects data scientist jobs growing from 275,600 in 2025 to 371,000 in 2035, and agents still achieve only 14.55% accuracy on hard multi-step benchmark tasks.

Which data science tasks can AI already do?

Google's Colab agent creates notebooks, and Databricks says its agent can explore data, generate and run code, and fix errors from a single prompt. OpenAI runs an internal data agent for over 3,500 employees.

Is data science still a good career?

The BLS projects a +34.6% change from 2025 to 2035, against 3.5% for all occupations, with 24,800 openings a year and a $120,230 median wage in 2025. Microsoft's research suggests entry-level hiring can slow in exposed occupations, so new graduates should expect competition.

Have data scientists been laid off because of AI?

We found no documented layoff of data scientists specifically attributed to AI. Block cut nearly 4,000 employees in early March 2026 citing AI, but that was company-wide.

How exposed are data scientists to AI?

A Microsoft Copilot study ranks data scientists 19th of 785 occupations for AI applicability. The Microsoft researchers also wrote that their data do not show AI performing all the work activities of any one occupation.

Sources

Every number, quote and example on this page was checked by script against the sources below before publishing: quotes must match the source word for word, and figures must appear in a source or in BLS data. Our verdict is editorial judgment. Spotted an error? Tell us.

  1. U.S. Bureau of Labor Statistics: Employment Projections, Data scientists (15-2051)
  2. Google launches free Gemini-powered Data Science Agent on its Colab Python platform (venturebeat.com)
  3. Data Science Agent in Colab: The future of data analysis with Gemini- Google Developers Blog (developers.googleblog.com)
  4. Google just released a new AI agent for data scientists on Colab, and it’s free to use (itpro.com)
  5. Data Scientist Recruitment: The Essential 2026 Guide - Pin (pin.com)
  6. Will Data Scientists Be Replaced by AI? 2026 Outlook | Extern (extern.com)
  7. Industry and occupational employment projections overview and highlights, 2024–34 : Monthly Labor Review : U.S. Bureau of Labor Statistics (bls.gov)
  8. Artificial intelligence, information technology, and employment, 2024–34 : The Economics Daily : U.S. Bureau of Labor Statistics (bls.gov)
  9. New Future of Work Report 2025 (microsoft.com)
  10. Microsoft study: AI could transform desk work and data science more than lab bench (rdworldonline.com)
  11. AI and the Future of Work: Microsoft Identifies Jobs Most Vulnerable to GenAI (cloudwars.com)
  12. Microsoft study highlights jobs most at risk from AI | Outsource Accelerator (news.outsourceaccelerator.com)
  13. Databricks (@databricks) on X (x.com)
  14. Introducing Databricks Assistant Data Science Agen... - Databricks Community - 130961 (community.databricks.com)
  15. AI/BI Genie, Foundational Model API, and Databricks Assistant Now Generally Available in AWS GovCloud (databricks.com)
  16. Databricks adds Data Science Agent to automate analytics tasks (infoworld.com)
  17. OpenAI Internal Data Agent Handles Complex Queries Without Human Oversight (medium.com)
  18. OpenAI: Building an AI Data Agent for Enterprise-Scale Analytics - ZenML LLMOps Database (zenml.io)
  19. OpenAI's AI data agent, built by two engineers, now serves thousands of employees — and the company says anyone can replicate it (venturebeat.com)
  20. DABstep: Data Agent Benchmark for Multi-step Reasoning (arxiv.org)
  21. DABstep: Data Agent Benchmark for Multi-step Reasoning (arxiv.org)
  22. DS-STAR: A state-of-the-art versatile data science agent (research.google)
  23. ASA Board Statements (amstat.org)
  24. American Statistical Association Statement on Ethical AI (amstat.org)
  25. ASA Ethical Guidelines for Statistical Practice — CASRAI (casrai.org)
  26. AI Layoffs in 2026: What Block’s 40% Cut Means for Data Scientists (interviewquery.com)
  27. Theoretical Capability vs Observed AI Usage by Occupation (pooyagolchian.com)

The White-Hat Brief

Weekly: what the machines learned, and how to stay ahead

One email a week. Verified stories, plain-language defenses, no hype. Unsubscribe anytime.