Because LLM's can do amazing things, but they can also make huge blunders. Claude constantly makes incorrect assumptions, or implements the wrong thing, or claims there is a bug when there is not.....

On the other hand its VERY fast at finding and fixing some bugs, and finding problems before they hit production. Very good at checking things like what API changes an upgrade brings or finding the right answer about a badly documented library, etc.