|
I find this topic super interesting because I recently created some side-projects using Claude (not in their entirety, but major parts of them) and felt strangely disconnected to them afterwards, even though they accomplished what I was looking for. That made me think about how AI usage for me should look like in my free time. Maybe what the author is talking about, and what I felt to a degree, is the act of deciding and formulating decisions. Looking at the compiler vs. AI example for machine code, one aspect that immediately stands out to me is that a compiler should deterministically produce consistent outcomes for precise instructions. Claude, on the other hand, will take your instructions, which are likely less precise than code, and "reason" about them. It will produce output that varies each time you would ask it and it will bake in behavior you didn't explicitly ask for. AI makes decisions for you. And maybe that's what makes AI so useful in the first place. One way around this may be to be just as strict with your instructions. Giving AI an exact spec to work with, like large Markdown specs or detailed tests to run against. But then again, if you take the time to formulate those, is it much of a jump to write the code yourself? What's clear is that AI is awesome for helping with issues you can't resolve, or explaining concepts in a personalized way. But letting it make decisions for you prioritizes outcome over creation, and for side-projects you build out of the love for creation, that may just not make sense, just like as a consumer I feel disinterested in "AI art". AI is for products, not for art, and that dumb little idea to work on in the afternoon might just be more art than product. |