If yours is, I totally get where you’re coming from, AI-generated code is pretty bad right now, albeit it does often work.
The way I often do things is have an agent create a rough first implementation according to my own architecture, so I have done all that high-level thinking that it currently struggles with; then I have a dedicated improvement agent come and clear that up substantially, and then I review whatever is left.
Programming is an art and using LLMs makes you as much of a hack as a visual artist resorting to Stable Diffusion.
Everything matters. From high-level architecture to little details of implementation.
Last time I checked us programmers had preferences for details as small as tabs vs. spaces, or curly bracket placement.
You think I’m going to let a fucking machine make decisions about implementation for me?
Let alone high-level architectual ones?
Everything about your software should have intent, and its impossible for LLMs to actually intend to do something.
So you can fuck right off with that, and while you’re at it, stop poisoning the worlds codebases (and drinking supply.)
Trust me my code aint worthy of being “art”.
If yours is, I totally get where you’re coming from, AI-generated code is pretty bad right now, albeit it does often work.
The way I often do things is have an agent create a rough first implementation according to my own architecture, so I have done all that high-level thinking that it currently struggles with; then I have a dedicated improvement agent come and clear that up substantially, and then I review whatever is left.