My thoughts on When I reject AI code even if it works: When I reject AI code even if it works
Commentary
- Vinicius’ Blog: When I reject AI code even if it works
- Yes, exactly. When YOU write the code, you have the context, you have the details, you know why exactly there were 2 parameters and not one, why it was abstracted from the module, why the code was written the way it was. But when we offload that to an agent, it has the context. No matter how much you read the ask mode answer, you still have lost the connection to the code, you have connection but scattered or not fully formed.
- This is might not seem like a issue to have confidence in the code at first when merging, but when things start to break over time, as they do, it becomes harder and harder to reason through it. The compounding effect of your lose connection to the code and the pile of changes from the Agent which might be unwanted and trash. We software developers have to pay that cognition debt to resolve it, its a trade off people are thinking coding is solved. It might not be just yet. The compounding will take some time, some of them have already realised it, some are about to hit the storm.
- This paragpraph sums it so wellYes, coding agents can help you with this task with more than just writing code, but that doesn’t mean they can do it autonomously in a sustainable manner yet.
- They cannot just yet.