
I’ve been trying different models that it offers. The most recent one I used that made me so fucking angry is Claude Sonnet 4.5. Just now I asked it “where is the JSON output being created in this codebase” and it told me a file that DOESN’T FUCKING EXIST even though I gave it all the necessary context
I didn’t know what method it was or what it would have been called and searching for relevant search terms returned like 200 results. I am aware of all the tools available to find what I was looking for. Anyway I found it by setting Copilot to Agent mode instead of Ask mode so that when I provided the file structure as context, it could independently look through the actual contents of the files (without me having to provide ALL the files’ contents as context and overload its context window).
I guess the real problem I was running into was that I never use Agent mode because I don’t ever want it making any edits to my files ever, but I guess it was actually necessary in this case for it to be able to find the context it needed to answer the question. But yeah the C codebase was too complicated and unfamiliar for “just search for method uses” to be helpful. Like yeah obviously I had tried that, I’m not an idiot, and I still used that after the AI gave me a place to first start lookin
I wasn’t passing them all through as context, just the folder structure. But I guess that doesn’t actually let the AI look through the actual file contents, so I needed to use agent mode to have it actually independently decide which files’ context it needs. And yeah I have no prior C experience (some C++ tho) and this is a mature codebase I’m new to. But this example really was a me problem, oopsies 🫠