This warning was introduced in pandas 1.0.0, following a discussion on GitHub. So best use what was suggested there:
df.groupby([0, 1])[[1, 2]].apply(sum)
It’s also possible to move the slicing operation to the end, but that is not as efficient:
df.groupby([0, 1]).apply(sum).loc[:, 1:]
Thanks @ALollz and @cmosig for helpful comments.