Have you tried to see what things an llm suggests for improving or investigated new features? You can spitball with an agent working on the actual code and see where that goes.