It may just be terminology.
'AUTOINCREMENT' implies that that record '103' will get created between records '102' and '104'. In clustered environments, that isn't necessarily the case for sequences. One node may insert '100','101','102' while the other node is inserting '110','111','112', so the records are 'out of order'. [Of course, the term 'sequence' has the same implication.]
If you choose not to follow the sequence model, then you introduce locking and serialization issues. Do you force an insert to wait for the commit/rollback of another insert before determining what the next value is, or do you accept that, if a transaction rolls back, you get gaps in the keys.
Then there's the issue about what you do if someone wants to insert a row into the table with a specific value for that field (ie is it allowed, or does it work like a DEFAULT) or if someone tries to update it. If someone inserts '101', does the autoincrement 'jump' to '102' or do you risk attempted duplicate values.
It can have implications for their IMP utilities and direct path writes and backwards compatibility.
I'm not saying it couldn't be done. But I suspect in the end someone has looked at it and decided that they can spend the development time better elsewhere.