How AI Is Actually Changing Data Science: Lessons from Leaders

Duygu Dagli & Gorkem Turgut (G.T.) Ozer
Data scientist steering AI agents
Image – Data scientist steering AI agents||

Around this time last year, agentic AI took a major leap. It is no longer just a tool for generating code, rewriting prose, or summarizing documents. In our own work, we are watching agent collectives conduct literature sweeps, extract and clean data, build plausible models, and synthesize insights, all orchestrated under the supervision of a lead agent. While an individual agent remains imperfect, an orchestrated collective is genuinely impressive.

Given these capabilities, what does the future of data science hold, and what will modern data science teams look like? We spoke with five data science leaders across retail and tech to assess the state of the field: the impact of LLMs and agentic workflows on day-to-day work (via automation or augmentation), the limitations that persist, shifts in the market's demand for data science talent, and ways data scientists can stay ahead as the role evolves.

Our key takeaways include the following, starting with the obvious.

AI accelerates research, code development, documentation, and insight generation, with caveats:

  • Code quality and technical debt: AI-generated code tends to be verbose and difficult to review. While it can run a prototype, it rarely scales out of the box and is seldom production-ready. Consequently, data scientists need to upskill in engineering and deployment.
  • Insight validation: AI-generated insights appear polished but lack reliability. Data scientists and analysts need to spend significant time validating outputs, rooting out hallucinations, and filtering for internally consistent yet externally baseless claims.

AI excels at predictive analytics but lags in causal and prescriptive modeling: Autonomous agents struggle where counterfactual reasoning, experimental design, and strategic interventions are required. Turning on advanced reasoning or adding more context does not replace human judgment. In these areas, deep domain and methodological expertise remain the core differentiators; data scientists must scrutinize execution and validate baseline assumptions to guide model development.

The perception gap between technical and non-technical stakeholders creates friction: Non-technical leaders often view agents as a silver bullet capable of end-to-end automation. This mismatched expectation puts undue pressure on technical teams. On the ground, technical teams experience the opposite: they must audit more work and refactor AI code, which often drains productivity.

Organizational design and performance metrics have not kept pace: While data science leaders agree that team structures and responsibilities must evolve, few organizations have established a clear path for how to execute this realignment. This gap represents a major organizational blind spot. On the bright side, more organizations now mandate traceable audit trails that log prompts and outputs, which may provide the foundational data needed to redesign the performance metrics and team workflows.

As of late 2026, AI has not yet replaced data scientists, but it has redefined the role. For junior data scientists, syntax and basic coding are no longer barriers to entry. For senior data scientists, coding now takes up a much smaller share of their time. The data scientist's value proposition is shifting from writing code to applying judgment, validating solutions, and steering AI agents. Data science organizations will inevitably redesign their processes. We think the ones poised to thrive will be those that get the redesign right, pairing AI's speed with human judgment.

Other popular articles

What if parallel trends are not so parallel?

Image
Image courtesy of github.com/asheshrambachan/honestdid Podcast-style summary by NotebookLM Academic's take In difference-in-differences models, parallel trends are often treated as a make-or-break assumption, and the assumption is not even testable. Why not? One misconception is that comparing the treated and control units before treatment is a test of the assumption. In reality, the assumption is not limited to the pretreatment period but for the entire counterfactual. Comparing trends in the pretreatment period is only a plausibility test. That is, "parallel pretreatment trends" and "parallel counterfactual trends" are not the same. We can observe the former, but we need the latter. So, what we have is not what we want, as is usually the case. Even testing for parallel pretreatment trends alone is tricky because of potential power issues (evidence of absence is not the absence of evidence!) and other reasons, including a potential sensitivity to the cho...
Read more

How to (and not to) log transform zero

Image
Image courtesy of the authors: Survey results of the papers with log zero in the American Economic Review Podcast-style summary by NotebookLM Academic's take Log transformation is widely used in linear models for several reasons: Making data "behave" or conform to parametric assumptions, calculating elasticity, etc. The figure above shows that nearly 40% of the empirical papers in a selected journal used a log specification and 36% had the problem of the log of zero. When an outcome variable naturally has zeros, however, log transformation is tricky. In most cases, the solution is to instinctively add a positive constant to each value in the outcome variable. One popular idea is to add 1 so that raw zeros remain as log-transformed zeros. Another is to add a very small constant, especially if the scale of the variable is small. Well, the bad news is that these are all arbitrary choices that bias the resulting estimates. If a model is correlational, a small bias due t...
Read more

Explaining the unexplainable Part II: SHAP and SAGE

Image
Image courtesy of iancovert.com Podcast-style summary by NotebookLM Academic's take This post is an exception to close the loop we opened with our post on LIME:  Explaining the unexplainable Part I: LIME . Starting in 2025, we're changing the scope of Data Duets to focus on business cases (as opposed to methods). Having discovered an excellent write-up  that explains both SHAP (SHapley Additive exPlanations) and SAGE (Shapley Additive Global importancE), I will focus on the why questions, possible links to counterfactuals and causality, and the implications for data centricity. Why do we need SHAP and SAGE? The need for explainability methods stems clearly from the fact that most ensemble methods are black boxes in nature: they minimize prediction error but they obscure how the predictions are made. This is problematic because some predictions are only as good as the underlying conditions are favorable. For example, Zillow's iBuying was a notorio...
Read more