Agreed but also have you tried to write a feature with LLMs? It will occasionally just add extra files that do the same thing as one library you already have.
You can even catch it and it will be like oh your right then continue without deleting, so next prompt it decides to use its own library again.
The trick is to tell it what to do and how to do it, not just talk to it like it knows what it’s doing.
People get upset when it can’t complete features based on high level descriptions. They were told it could, after all. “Build be a website that…” However, that will always lead to disaster.
Even then, it’s going to make some batshit decisions on stuff you don’t specify and probably screw up some that you did, and much of it will cause subtle issues.
Remember that even with all the shit they add on to it, it’s still based on a text predictor, not some kind of problem solving or programming engine.
All that tracking software is bloated.
Agreed but also have you tried to write a feature with LLMs? It will occasionally just add extra files that do the same thing as one library you already have.
You can even catch it and it will be like oh your right then continue without deleting, so next prompt it decides to use its own library again.
The trick is to tell it what to do and how to do it, not just talk to it like it knows what it’s doing.
People get upset when it can’t complete features based on high level descriptions. They were told it could, after all. “Build be a website that…” However, that will always lead to disaster.
Even then, it’s going to make some batshit decisions on stuff you don’t specify and probably screw up some that you did, and much of it will cause subtle issues.
Remember that even with all the shit they add on to it, it’s still based on a text predictor, not some kind of problem solving or programming engine.