It feels like it’s not as reliable as it used to be. Tasks that should take a couple of hours are now dragging into days because it keeps making unnecessary changes to the code. Instead of streamlining the process, it’s actually slowing things down.
One of the issues is that Cursor seems to misunderstand or even invent rules. For example, I’ve had it flag certain files as “relevant” and claim they came from my Cursor rules, but they didn’t. It seems like it’s making assumptions that aren’t based on the actual context. When I try to dig into its reasoning by asking Composer to explain its decisions, it often outputs something confusing or unrelated. It looks like Cursor isn’t directly using coding LLMs anymore but relies on some kind of intermediary memory system, probably to manage rate limits or optimize context usage. Unfortunately, this memory system seems to be the root of a lot of the issues—it pulls in unrelated files as “relevant” and uses them in the context, which ends up derailing the code it generates.
There’s also a big problem with how Cursor handles context. Once it picks up a piece of information, it doesn’t seem to let go of it. Even if I update or change the code, it keeps referring back to the old version like it’s stuck. Refreshing the session or attaching new files doesn’t seem to help—it still clings to the outdated context in its memory. Even when I attach updated files to reset the context, Cursor seems to ignore them entirely and defaults to the old information it already has stored. This makes it really difficult to work on projects that evolve over time because no matter how much you try to guide it with new inputs, it sticks to its original assumptions. The only way I’ve been able to resolve this is by restarting Cursor completely, as even starting a new session doesn’t seem to properly reset the context.
It’s almost like Cursor’s reliance on this intermediary memory system has introduced a lot of issues. It doesn’t seem to handle context length or updates properly, and it often ends up hallucinating rules or making incorrect assumptions about the files you’re working with. Instead of helping streamline coding tasks, it’s now introducing confusion and making the process harder. Has anyone else noticed these issues, and is there a better workaround than just restarting Cursor constantly?