Good software practices ensure high-quality, maintainable, and scalable code. They reduce technical debt, improve collaboration, and make production deployments far less painful. In data science, the focus often sits on model accuracy. But deploying a model means the entire pipeline, including data I/O, cleaning, and monitoring, must meet production standards. Designing with best practices from the start avoids costly rewrites later. Kedro is an open-source Python framework that embeds these principles into data science projects from day one. This talk covers modularity, separation of concerns, testability, and reproducibility, showing how Kedro structures projects for easier deployment and long-term maintainability. Beyond traditional ML, Kedro's architecture extends naturally to modern AI techniques like Graph RAG and agentic reflection. Managing complexity and reproducibility becomes even more critical in non-deterministic agentic systems, and Kedro provides a solid foundation for both. Attendees will leave with practical insights applicable to any data project, regardless of their software engineering background.