Researchers have introduced MOSAIC, a framework for structured agentic intelligence and composition in automated data science. The system builds on prior cases and source-code modules to construct a blueprint for model selection, which is then validated by execution and refined using diagnostic feedback. This approach improves task performance, execution success, and decision traceability compared to AutoML and agentic baselines. --- Why it matters: This matters because it addresses the limitations of current AutoML systems, which often rely on predefined pipelines and hyperparameter spaces. MOSAIC's structured approach allows for more flexible and reusable model selection, making it a valuable tool for researchers and practitioners in automated data science. Source: https://arxiv.org/abs/2606.00708