I don't think your example fit this scenario: when you search things in google, you know you are using google. When you ask Siri to do something and the answer

I don't think your example fit this scenario: when you search things in google, you know you are using google. When you ask Siri to do something and the answer is garbage, where the issue lies? It's Siri not giving the right context? It's Siri that is not providing the correct tools to the LLM? Or it's the LLM that botched it? We already see these kind of issue with LLM and harness, and we are talking with developers that "should" know how to move in this scenario ;)