I’m building a small agent around the Grok API and trying to decide how much recovery logic belongs in my app. If a tool call fails—for example, because a service times out or returns an error payload—does Grok generally recognize the failure and make a sensible next move, or does the flow need to be managed almost entirely by the caller?

I’m especially interested in what happens after a failed first attempt: does it explain the issue, retry with adjusted arguments, or keep acting as if the tool succeeded? Experiences with tool call errors in real integrations would be helpful.