I'd like to see LRMs be much faster, but not lock on. You could still
target them from behind hills and such, but they wouldn't do the whole MW2+ homing bees thing. This is of course a pipe dream at present, but possibly it may happen in the future when more resources are available.
I partially disagree with the "no skill" accusation, however. Properly employed LRMs do require skill in the form of positioning and map knowledge - same as direct-fire weapons. All those people who just sit back behind everything and play "lock-on box warrior" are doing it wrong. So there
is skill that should be applied to LRM usage, even if many LurmWarriors don't actually... use it. Similar to how so many players using direct-fire weapons don't torso twist much.
That being said, I think the problem with LRMs is inadequate counterplay on the part of the targeted player - particularly in the PuG environment. In random matches, you can bring AMS and possibly ECM, but other than that you kinda just... hope. Does your team have enough ECM to stop indirect fire? Hope so, or it's a bad day. Do you have team assets willing/able/observant enough to keep spotters off your back? Hope so. Does
not your entire team consider UAVs to be
Somebody Else's Problem? Hope so... Did the puggles vaccinate their dang 'mechs with AMS like thinking humans? Hope so, but probably not. The only active counterplay you really have is to hide behind cover, potentially ceding vast portions of the map to the enemy team. Thankfully, LRMs don't do enough damage (focused and overall) to really be worthwhile against skilled players, even when skillfully employed. If LRMs go back to doing enough damage (particularly with a high enough missile speed) to be viable as weapons in their own right, they'll be unstoppable - you know, again.
Combine this with the amazing variability in how LRM-friendly the maps are -from Caustic Valley easy mode to Crimson Strait's air raid shelter - and you have a weapon system that's nearly impossible to balance, which brings it to its current state of high-level nonpresence. I'm increasingly sure that LRMs will need a mechanics overhaul like the one I described in order to be both balanced and viable.