The advantage of the visible light approach is that it doesn't reach through walls.
It's inherently very dense, and can provide users in its footprint nearly unlimited bandwidth. If it reaches only 50% of users in a classroom or office building, then those 50% of users will have very fast connectivity, and wifi will be twice as fast for everyone else.
There are other potential approaches but WiFi+LiFi hybrids could be neat. LiFi would also have the regulatory advantage of not needing to worry about typical licensing or bandwidth issues beyond basic emissions like anything with a power supply has, since those spectrums are of course fully public :). Although the industry could in principle muck the whole thing up with burdensome patents I suppose, but then again the basic principle are similar to fiber and free space optical transmission (including extremely high speed) has been done for ages so that might limit the scope of any potential claims to some degree.
I think an uplink with visible light might be too much to ask, but if we could just glue these two networks together, it becomes a lot more useful.
I wish more technologies could do one-to-many broadcast. for example a live video stream that many were watching, or tv/movies/presentations where many people were listening to the same audio with headsets or earbuds.
This could solve the problem of getting reliable speeds to devices without Ethernet, that are either hard to wire or don't want to plug in.
Higher density is more costly too of course, but even so depending on the performance that might well still be cheaper/easier then fiber drops everywhere, plus it'd avoid some of the minor but sticky end user physical interface issues of fiber connections. Whether the value is there remains to be seen and will no doubt vary, but as others have said a lot of typical office plans (with drop ceilings, ceiling rails or other designs meant to enable more modular ceiling use) on the surface seem pretty amenable to such an approach.
Ultimately all depends on what the actual implementations look like though, and at what cost.