AI coding assistance

Stating the data-science step increased acceptance of AI code suggestions, while alternatives alone did not

Dhivyabharathi Ramasamy, Cristina Sarasua, Abraham Bernstein

Empirical Software Engineering

An experiment with 28 data scientists tested a notebook assistant that could include workflow context and offer alternative code recommendations. Naming the data-science step increased acceptance of suggestions, while alternatives alone showed no significant effect. Predictive tasks were easier for the assistant-supported workflow than descriptive analysis. The small study informs interface design; accepting more generated code does not by itself demonstrate better scientific reasoning, reliable findings or improved learner competence.