Clustering is one of the most classical problems in data analysis, yet it continues to inspire new ideas and methods. In this talk I will revisit the Density Peaks framework and some of its recent developments, including Advanced Density Peaks and strategies to improve its computational efficiency. Building on these ideas, I will introduce a new perspective that connects Density Peaks with other well-known approaches such as Mean Shift and Spectral Clustering, leading to a method we call Spectral Density Peaks. Finally, I will briefly explore a different direction: whether representation learning tools such as Variational Autoencoders can help simplify clustering problems by reshaping the geometry of the data. I will conclude with a few speculative ideas and open questions for those interested in pushing these approaches further.