Claude code seems to be all that an engineer would need. I was wondering is it still worth building a new harness which is lightweight and have inbuilt router and memory to switch between models? Optimised for task resolution with multiple labs working together?