The problem in this case is that LLMs are bad with golang, I don't write go, I am guessing from my experience with kotlin. I mainly use kotlin (rest apis) and LLMs are often bad at writing it. They e.g. confuse mockk and mockito functions and then agent spiral into a never ending loop of guessing what's wrong and trying to fix it in 5 different ways. Instead I use only chat, validate every output and point out errors they introduce.

On the other hand colleagues working with react and next have better experience with agents.