I'm currently doing some tv work that has lots of animation of objects into place, you know the stuff, fly 1000 glass panels into this sky scraper, or 1000 stone blocks into an ancient structure. I'm generally an all in "one geo node" bloke, and my core technique for this specific task is to have my geo in both a final position and a start position and user a linear interpolate to blend between the 2, and use proximity to some animated object to drive the blend, and up till now this has been fine, but I am now running into the overhead of doing everything as a shape animation rather than a straight up transform at object level.

Thinking about how I would have done this in softimage it would have been done by building a massive hierarchy using a copy and transform, then doing the same simple animated transform to each piece, then using "sequence animation" to offset the animation. It would work, but could be annoying to set up, but it feels like I would have had more control over things, not least 3 keyframes rather than the 2 of my lerp setup.

Seems like there are no tools whatsoever for doing stuff at object level, am I missing something, or have you all written a load of custom scripts for doing this type of thing?

you don't want to do it on object level, 1000's or even 100's of objects would slow down your scene to crawl

SOP level with packed geo is the most efficient way and you have full control over animation if you want
just animate the points first and instance packed feo on them later
you can make a single animation as trs curves and sample them using chf() expression at different times per point to acheive your offsets etc... (similarly you can use Ramp parameters, but animation curves may be more intuitive)

And to add to Tomas's answer, if you somehow need to export the animation as separate objects for somebody else to use, you can easily generate unique names for your packed geos and recreate the needed hierarchy from those names at export time, all the while staying at SOP level.