In groupby._transform, I think we can always use _wrap_transform_fast_result in the case of a string aggregator instead of falling back to _transform_general for certain cases. The only thing stopping this now is it removes the warnings added here. Digging into it, I'm finding some inconsistencies:

df = DataFrame(
    {
        "A": ["foo", "bar", "foo", "bar", "foo", "bar", "foo", "foo"],
        "B": ["one", "one", "two", "three", "two", "two", "one", "three"],
        "C": np.random.randn(8),
        "D": np.random.randn(8),
    }
)
df.groupby("A").mean()
df.groupby("A").agg("mean")
df.groupby("A").agg(np.mean)
df.groupby("A").transform("mean")  # <--- warns
df.groupby("A").transform(np.mean)  # <--- warns
df.groupby("A").apply("mean")
df.groupby("A").apply(np.mean)  # <--- warns

Should all, just the np.mean, or none of these warn? From the whatsnew note, it seemed to me that all of these should warn.

Should all, just the np.mean, or none of these warn? From the whatsnew note, it seemed to me that all of these should warn.

I think you're right. IRC getting consistency across all these cases was something we punted on.

Thanks, I'll take a deeper look into emitting the warning in the right places.

I missed that the default of numeric_only is still currently True, so many of the above cases are currently correct. With #46072, all of the .mean() and "mean" forms will warn.

The np.mean should warn in all cases, but doesn't for agg. I plan to be putting up a PR for this.
I'm no longer certain on this. agg intercepts np.mean and replaces it with 'mean', which is why it doesn't warn. transform tries to do the same, but ends up having to call transform_item_by_item instead. apply also intercepts and replaces with 'mean', but then uses DataFrame._reduce which warns.