Several approaches to statistical inference are routinely used in practice, yet many focus primarily on latent target functionals of the data-generating process, even when prediction is the ultimate goal. After a brief review, this talk introduces a prediction-centric framework for inference that places observables at the core. The motivation is both theoretical and practical. Indeed, many real-world decisions rely directly on predictive uncertainty about future observations. By aligning inference with these decision-relevant quantities, the framework provides a principled basis for both prediction and uncertainty quantification, yielding induced uncertainty statements for functionals of interest and thereby translating predictive uncertainty into uncertainty over traditional inferential targets. Despite the growing centrality of predictive methods across the sciences, a coherent prediction-centric paradigm for inference remains underdeveloped. This talk presents recent advances in this line of work.