When I ask a data scientist to explain an analysis, I am often handed a pull request and sent away to understand the implementation. Twenty minutes later, I may understand how the code is organized and still not know what the data showed. I want to understand the question, what we found, and whether the evidence supports the conclusion. Coding agents such as Claude Code, OpenAI Codex, and Databricks Genie Code can write, execute, and revise code, reducing the effort between having an idea and trying it against data. That gives us an opportunity to reconsider where a data scientist's attention belongs.

Data science has always centered on discovering what we can learn from data. As the effort of writing code declines, we should devote more attention to the inquiry and make that inquiry easier for a colleague to examine. Code is a tool for the scientific work, an extension of a drafting board where a hypothesis takes form we can examine and revise. An analysis in Microsoft Excel can produce useful scientific insight, and its credibility depends on the same considerations as an analysis written in Python.

For coding agents, the calculator comparison is instructive: understanding arithmetic and formulating the problem remain important even when we delegate the calculation. Programming fundamentals help us understand how code transforms data and determine whether an implementation does what the analysis requires. The scientific contribution requires the additional expertise to formulate useful questions and draw defensible conclusions from incomplete observations.

My concern is that code can interrupt the exploration needed to determine what the application should do, particularly when code receives more scrutiny than the analysis or an exploratory adjustment must pass through review before we can investigate its merit. In my experience, larger data science teams often struggle to preserve the discovery process when code review dominates.