Yes, Bulldozer was clearly the biggest mistake AMD ever made. If the company does end up failing, this will be remembered as the point where everything really started to go wrong.
AMD would have been better off just biting the bullet and eating the R&D costs in 2011 when it became clear that Bulldozer wasn't going to be competitive. They should have fired the leadership most responsible for this fiasco, and reassigned their best engineers to incremental improvements on K10 (IPC increases, adding AVX, die-shrinks, etc.) Sure, this is easier to say with the benefit of hindsight, but even at the time it was clear from the benchmarks that Bulldozer was going to be flat-out inferior to Thuban in many regards (despite the fact that Bulldozer had the benefit of a node shrink!) Releasing a product like that is a classic example of the sunk cost fallacy. It should have been written off; that would have done far less damage to AMD.
The idea of the APU was sound enough. After all, in all product segments except servers, the vast majority of CPUs today from all companies have GPUs on the same die.
There were several execution problems that kept APUs from being a sales success. The bandwidth limitations of DDR3 meant that standard PC APUs could never really compete effectively with even a $99 discrete GPU. As a result, the die size turned out to be much larger than it needed to be; AMD probably shouldn't have put more than 256 GCN shaders (four computing units) in any DDR3 APU, but they packed double that figure in Kaveri, and it was mostly a waste. And the desirability of APUs was further worsened by the fact that they were bogged down with the terrible Bulldozer-derived CPU architecture.
That said, the PS4 demonstrates that an APU can work just fine for consoles, when it can take advantage of soldered GDDR5. In the long run, HBM2 will work even better for this. But AMD tried to integrate too much too soon, and ended up with a mediocre product that few people wanted.