The A320 crash near Strassbourg comes to my mind, the pilots completely failed to notice the high descent rate they had programmed (I think it was 1800 ft/min in a non-precision approach). I wondered that nobody has mentioned that yet.
They've since re-designed the interface to that particular system to avoid that kind of mode-confusion as I understand it. Though what the designers were thinking when they allowed a push/pull system on a single knob to switch between descent angle and descent rate I'll never know...
My lecturer on Software Engineering & Reliability was Peter Mellor, who's quite a well-know talking head in aviation circles, as I understand it. The Habsheim and Strasbourg accidents were used as examples of how software can be doing exactly what it has been told to do, but that the system is not communicating to the user exactly what that is.
As I understand it, the AA 757 at Cali was at least as much down to an out-of-date Jeppesen chart as it was the automation. Offered a straight-in approach rather than the pre-programmed flight plan, they were told to head to Rozo beacon, which was indicated on the chart as 'R'. When 'R' was punched into the FMS, it deleted the previous route completely, including the reference to the 'Tulua' beacon mentioned earlier. The mistake was that as far as the FMS was concerned, 'R' was not Rozo, but the Romeo beacon near Bogota - and unfortunately the mountains lay inbetween the two locations and the rest is history.
I'm a programmer by trade, but I have always felt that automation and manual handling skills are complementary rather than exclusive in nature. The computer should be there to help, *not* to replace. I guess one of the major problems was Airbus's (and to a slightly lesser extent Boeing's) marketing department aiming their sales pitch over the heads of the pilots, hoping that the cost savings of the automation would sell the plane alone.