From this I start thinking that folders will be eventually just replaced with labels but labels should become nested.
Leaving terminology aside nested label becomes just a folder that can be assigned to multiple nodes. Which in Linux filesystems is called hard linking folders.
I just noticed that gmail has moved from simple Labels to nested Labels! Now you can create sublabels with seemingly unlimited depth.
Nesting structural aspects/sublabeling has become inevitable for gmail so hopefully UX aspects of working with nested labels becomes more standard.
In practice, tagging systems seem to be limited by relatively immature interaction models relative to the far better explored nested structures. For example: for a given folder in the hierarchy there are often few enough children to allow easy recognition of a forgotten name whereas tags for non-empty subsets of a given set of files are hard to suggest meaningfully; nested folders demand more care beginning with the earliest placement of files but tagging makes wholly unsustainable conventions perfectly reasonable early on; sections of nested structures are easy to relocate to a new parent but changing the related set of tagged files to uniformly substitute comparable target labels for precisely those obsolete labels is difficult or impossible.
In short, there's a theoretical basis for replacing folders with tags, but the real-world adoption is severely hampered by grossly incomplete UIs for many needs previously met by nested folders.