This technical session examines the ongoing paradigm shift in ML, brought into sharp focus by the advent of Large Language Models (LLMs). Specifically, we explore the transition towards data-centric AI development—a methodology that prioritizes the cultivation of high-fidelity datasets over the traditional, unpredictable hyperparameter tunning and architecture optimisation that have plateaued in performance gains. The discussion will be anchored in quantifiable metrics showcasing how data-centric practices lead to reductions in both time and cost for model training and deployment, while substantially increasing model robustness and interpretability. We will look into workflow enhancements that have the potential to yield 10x productivity gains. Specifically, we give examples of automations that liberate engineering time for high-impact work, leveraging embedding models for multi-modal scenario mining and data quality assessment to uncover nuanced patterns and improve detection capabilities. Concrete case studies will be presented to highlight successful strategies and methodologies that have been employed to expedite development, reduce training and labelling costs, and raise the bar for model performance. Attendees will leave with a clear set of data-centric tactics that promise to catalyze the efficiency and output of their computer vision initiatives.