Clustering resolution for scRNA-seq

This is an open-ended question, but maybe you could share your heuristics. What’s your approach when clustering and annotating clusters in scRNA-seq data? Do you prefer to start with a low number of clusters(e.g. CD4, CD8, monocytes) and then re-cluster or look for low-level cell types from the get-go(e.g. Treg, Th, etc)? I’ve seen both approaches in the literature. Are there any pros and cons?