One is the dungeon floor generator. Originally, I wanted to make really large maps, with spread-out rooms, similar to, say, Pokémon Mystery Dungeon. They would have required a tile engine to make the pieces all fit together properly, and players would spend quite a bit of time on each floor. Now, though, each floor will have only one or two locations of interest, perhaps only a single enemy, and a time limit that keeps players rushing around. That way, even the smallest obstacles may be interesting challenges. This also makes the purely turn-based battle system more exciting.
Speaking of time, the original plan was to have these dungeon dives as only one part of a larger time management scheme. As time passed, events would be scheduled on the fly, including special character conversations, class sessions, and main story points. They could happen in different orders due to player choices, sometimes multiple in a single time frame. However, I've now thought of ways to untie the character conversations and skill learning system (classes) from the schedule. Only the story events will still be part of the schedule, and since those were at specific points in the first place, it won't be so messy.
These are the main things I've seen fit to change about the game design itself. Next time, I'll talk a bit about changing the way I work on development.