We're finding that stuff we thought would be impossibly hard is subject to brute force using simple constructs, and stuff we thought would be simple is proving incredibly difficult.
Funnily the stated definition seems to fall for its own trap. Perception, here named as a particularly hard thing is something AI is starting to get pretty good at. Mobility lags behind a little, but we've seen some incredible demonstrations there too.
Perhaps it's always the things that are on our awareness horizon which seem hard but end up solvable. What's next? Perhaps long term planning, interference and on the fly learning? And then what are the difficult things beyond that horizon, the challenges we can't even see yet?
Moravec's paradox in action: https://en.wikipedia.org/wiki/Moravec%27s_paradox