The data science community is right to question which parts of their work will survive automation. The answer, as this Reddit discussion highlights, is that the core of data science, understanding business context and matching it to available data, resists automation far more than the coding that surrounds it. This distinction matters because it reframes the entire conversation about AI's impact on your workflow.
Consider the practical reality behind that observation. Writing code to pull data, clean it, or train a standard model is already being handled by AI tools, and that trend will accelerate. But defining the problem itself, what do you actually need to predict, and why, remains stubbornly human. A forecasting model for retail inventory requires knowing whether the goal is to minimize stockouts, reduce waste, or balance both depending on seasonality. An optimization algorithm for delivery routes needs to account for driver union rules, weather patterns, and customer preferences that no dataset captures cleanly. These are not technical gaps; they are contextual ones.
The subreddit user's curiosity about specific fields reveals a useful hierarchy. Anomaly detection and vision tasks, where the business objective is relatively standardized (find the outlier, classify the object), will likely see more automation because the outcomes are easier to define. Causal inference and optimization, by contrast, demand that you articulate a counterfactual or a trade-off, a judgment call that requires intimate knowledge of how the business operates. A/B testing sits somewhere in between: the statistical engine can be automated, but deciding what to test and how to interpret the results in light of real-world constraints cannot.
What this means for you as a practitioner is straightforward. The coding you do will become less central to your value, not because it disappears but because it becomes a commodity. Your real leverage shifts toward the questions you ask, the constraints you surface, and the interpretation you bring to the numbers. The field that resists automation most is the one that depends on understanding a specific problem within a specific organization. That understanding is not something AI can infer from data alone.