I feel that agent coding is actually giving a second wind of life to solid principles, “proper” software architecture. Now you can nag the llm to follow them and A- it will actually apply them if well directed and does not mind the (small?) extra complexity upfront B- you pretty much immediately see the effects