just curious if anyone knows or has a link to studies regarding comments in codebases and agentic dev. I recently went through and chopped 50%+ of the comments the model had left on initial build and I'm wondering if this improves parsing (obviously it will improve token usage) and reasoning -- hopefully not the inverse? anyone has looked into this?