I don't mind prime-stat ability scores slowly increasing as you level up, as a side effect of lots of practice and training at doing what you do, but this advancement IMO shouldn't be nearly as predictable as the WotC editions have it.
1e's Unearthed Arcana introduced the Cavalier class. Unremarkable in itself, it brought with it a wonderful mechanic for unpredictable but reasonably consistent ability score advancement by level called "percentile increments". This system can easily be tacked on to all classes in any edition.
How it works, in short:
--- At 1st level your prime stat gets a d% roll attached; thus a Mage with starting Int of 15 who rolls 87% has that Int become 15.87.
--- Each time that class levels up, some dice* are rolled and added to the percent number. If, say, the dice roll is 9 then that 15.87 becomes 15.96. If the dice roll, however, is 16 then that 15.87 becomes 15.103, which becomes
16.03: the stat advances.
--- Repeat each level. Next level that 16.03 might add 12 and become 16.15. That's it. Simple as pie.
The huge benefit of this in my eyes is the unpredictability of it - one character might advance a stat right away while another might wait several levels or more to advance a stat; but the law of averages pretty much dictates everyone will advance a point over a certain amount of levels. You can (and we have) expand this to a player-chosen secondary stat as well, which advances more slowly by rolling a smaller set of increment dice per level.
* - the incremental dice can vary from campaign to campaign depending how often the DM wants the characters, on average, to advance their prime stats. I used to use 2d10 per level but found it a bit too slow, now I use 3d8 per level which seems to work OK. To match WotC's advance speeds (a point per four levels), it would probably have to be more like 6d6 per level, or 4d6+8, or something similar.
For secondary stats I use 2d6 per level and they tend not to advance very often.
