[GitHub Copilot] Improve AI adherence to user plans and codebase context

Still obsessively working on my strategy tile/turned based game using VSC + GitHub Copilot + Unity. In my last update, the unit recruitment system was implemented which allows the player to recruit up to ten units at a time. In the screenshot here, units are organized in the UI in groups which are represented by group icon prefabs on the tilemap. In the exploration/early part of the game, groups move to undiscovered path tiles (grey tiles) which trigger an event based on the tile type. If the player completes the event challenge (combat, quests, puzzles, narrative choices), they take ownership of the tile (blue tiles). Groups can be stacked on player owned tiles and units can transfer between groups. There's also a fast travel option that allows any group to use their movement turn to move to any player owned tile even if not adjacent. So far, I've got a basic tile type and event system implemented. For example, in the first turn, the player moves a group icon to the first adjacent tile next to player's base. The group discovers a peaceful, neutral village who decides to join the player thus awarding the player a free tile and the resources that come with it. One of the next major phases of development will be to create a myriad of tile types and events. I have been set back a few times because of imprecise communication with Gemini. This is because my code base is becoming more complex and it can be a challenge remembering how everything works. Additionally, AI sometimes wants to implement a feature or mechanic in their own way, diverging from my request. By not being on top of what AI was doing (looking at you agent mode) or overly relying on AI to understand my code base, unneeded or unnecessary changes made to the game caused quite a few cascading bugs. I spent many hours helping AI fix these bugs by reading through the code, finding probable mistakes or issues, discussing them with AI, formulating a plan of action, and finally addressing the bugs, one by one. I have to say though, one amazing thing about working with AI to write code is its willingness to implement debugging logic for testing when problems occur and it can't figure out the solution. Going forward, all major implementations will be better documented and include a plan of action that can be agreed upon with AI to ensure that we aren't breaking parts of the game that are already working. A little bit more planning on my part will go a long way in reducing the amount of time I'm spending on bugs!