4 hours - think about a problem, come up with a new algorithm or variation of algorithm, or read papers to find an algorithm that will solve your problem. This is SO FUCKING FUN and what keeps pulling me back in.
8 hours - do initial test implementation to prove concept. It works! Yay!
And now the fun part is over.
50 hours - do more careful implementation that handles all the annoying corner cases; integrate with the rest of your library; handle failures and so on. Provide lots of options for slightly different use cases that massively increase the complexity.
50 hours - adding features and fixing rare bugs; spread out over the next year
20 hours - have to install new SDKs to test it; inevitably they've broken a bunch of APIs and changed how you package builds so waste a bunch of time on that
10 hours - some stupid problem with Win8 loading the wrong drivers; or the linux dir my test is writing to is no longer writeble by my user account; whatever stupid computer problem; chase that around for a while
10 hours - the p4 server is down / vpn is down / MSVC has an internal compiler error / my laptop is overheating / my hard disk is full, whatever stupid problem always holds you up.
10 hours - someone checks in a breakage to the shared library; it would take a minute just to fix it, but you can't do that because it would break them so you have to do meetings to agree on how it should be fixed
10 hours - some OS API you were using doesn't actually behave the way you expected, or has a bug; some weird corner case or undocumented interaction in the OS API that you have to chase down
40 hours - writing docs and marketing materials, teaching other members of your team, internal or external evangelizing
30 hours - some customer sees a bug on some specific OS or SDK version that I no longer have installed; try to remote debug it, that doesn't work, try to find a repro, that doesn't work, give up and install their case; in the end it turns out they had bad RAM or something silly.
The reality is that as a working coder, the amount of time you actually get to spend working on the good stuff (new algorithms, hacking, clever implementations) is vanishingly small.