If DBDs are upgraded they can be made to work with both (new) 1.x DBI versions and with 2.x DBI.
DBI trace level will be a bitmask instead of just a number. 8 bits will be reserved for drivers. There'll also be "trace keywords" with levels independent from the general trace level.
dTHR will be mandated in the drivers for DBI 2 which will make DBI slightly faster in embedded and threaded eenvironments (mod_perl 2).
$sth->reset (name to be determined) method to reset the statement state (call finish, reset attributes etc). This is to allow passing in handler attributes when you run prepare and to be able to run prepare again on a $sth.
unicode: maybe there'll be some utility functions for the drivers to use, but essentially it'll still be up to the drivers to do the appropriate thing for the database. Matt suggested writing up strong guidelines for how the DBDs should work. He was then volunteered to do it. :-)
Preparse magic: makes it easier to move SQL from one database to another.
Improving the test suite so the same test suite can be used by all the different drivers. So if you start writing a new driver you'll have an instant test suite. It'll also help making sure the driver behaviors don't divert in subtle ways.
There's also the DBI TODO (username and password is guest).