I'll second this.

Doing asm on the AVR is beautiful, you can count clock cycles easily and then observe them on the scope.

I wrote a bit banging serial interface for an AVR once and had a mystery when I was testing it from a PC just with a basic echo. Every Nth character would be wrong. Was able to figure out a timing problem by counting clock cycles and found the bug in my code.

Was cool to see it align with what I was seeing on my oscilloscope.