I don't see us tinkering with Pi in 5 years.

Some standard solution will emerge, which will be amplified by models being trained specifically to work with it.