If you can build a reliable agent just with prompting and tools then I'd recommend sticking to that. The use case for blocks/workflows is for agents with multiple complex multi-step instructions for implementations where there's no room for error