I was always really good at Grammar Translation exercises in my Spanish and German classes but never lost the “I don't speak X,” feeling at all. I remember using ALM (the Audiolingual Method) just a little for German, mostly the transformations of patterns, which seemed particular meaningless and not at all useful. The tests seemed to be mostly vocabulary (which was hard work to memorize) and this sort of exercise (which said more about puzzle-solving ability than proficiency in the language). We didn't do the full ALM though, just the transformations and a few dialogues. In those classes, I really felt like it was up to me to go out there, find some people to talk to, and become fluent on my own... which I didn't. I learned a lot about grammar, but I'm still pretty monolingual.
In Japan, I think we were trying to be Direct Method / CLT (Communicative Language Teaching) teachers with ALM-slanted textbooks. We also did TPR (Total Physical Response) style activities with kids whenever possible. The energetic methods I was made to use in Japan had some good points and some drawbacks, the biggest drawback probably being the amount of energy and subtle skill those methods took to execute properly. I think real communication in the L2 is key, but the class still needs to have some structure so all the students are clear on what exactly they should be learning. We developed a lot of "question sheets" as we went along, which turned out to be more useful than most of the textbook exercises. These lists of questions were meant to spur real, targeted conversation and could focus either on a particular topic, a single grammatical structure, or a handful of closely related structures. I found the last type of sheet to be the most useful, so that's what I spent the most time developing (after I'd been there long enough to get into the swing of things).
I'd like to use this kind of structured conversation approach (most closely aligned with CLT) about half the time with the other half being devoted to reading and writing. These activities should use genuine content like stories, articles, and websites wherever possible. In a course of a set length, the syllabus should be presented early to give the students an idea of what to expect. Ideally, it should be flexible with several conversation sheets and readings being part of some group from which the students can choose just one. In this way, the students will have more of a personal investment in their learning. If they chose to read a particular story, they might be more likely to pay attention and look for its good points so they don't feel like they made the wrong choice. To make these choices, simple majority voting could work, but I think an application like Choozer is a better way to get really crowd-pleasing results.