I really like this structured approach, which shouldn't really be that different if you didn't use AI to assist you. If you can't understand or articulate what you want, you should not trust an LLM to do it for for, because you're right, that would lead to chaos. I wish we didn't have to forcefully explain that we want basics like error handling added - I think that should just be the expected default - but I have seen some success using Claude for larger scale problem solving and analysis in a recent project.
Excellent analisis, I really connect with your observation that 'no clear goal, no feedback loop, no idea what done looks like' often leads to chaos, though sometimes I wonder if a little bit of undirected exploration is almost inevitable before a clear prototype emerges, even with AI.
The Four R's framework is a refreshingly practical take on AI-assisted development. What resonates most is the emphesis on failing fast - too often devs hide errors with fallbacks, loosing critical feedback loops in the process. The prompt templates section alone could save weeks of frustration for anyone building with AI right now.
I really like this structured approach, which shouldn't really be that different if you didn't use AI to assist you. If you can't understand or articulate what you want, you should not trust an LLM to do it for for, because you're right, that would lead to chaos. I wish we didn't have to forcefully explain that we want basics like error handling added - I think that should just be the expected default - but I have seen some success using Claude for larger scale problem solving and analysis in a recent project.
Great write up!
Excellent analisis, I really connect with your observation that 'no clear goal, no feedback loop, no idea what done looks like' often leads to chaos, though sometimes I wonder if a little bit of undirected exploration is almost inevitable before a clear prototype emerges, even with AI.
The Four R's framework is a refreshingly practical take on AI-assisted development. What resonates most is the emphesis on failing fast - too often devs hide errors with fallbacks, loosing critical feedback loops in the process. The prompt templates section alone could save weeks of frustration for anyone building with AI right now.