AIs can query the web and parse the results, so it is possible for Claude to understand the question, search the web for movie schedule of a cinema, parse the website and give an accurate answer. It’s also very possible for it to do it incorrectly or just pretend that it did that and give you a random time instead so that you are happy.
I can’t imagine trusting an AI with something like this, where it’s so easy to just check it yourself
Going even further, we had this capability already, for example the original Google Assistant was able to do that task just fine.
Yet Gemini similarly seems to struggle with this on mobile.
In the same example, I don’t understand why, that when the agent protocol started happening they didn’t offer the original Google Assistant pathway as an agent of mcp for Gemini to call and add its response to the context.
AIs can query the web and parse the results, so it is possible for Claude to understand the question, search the web for movie schedule of a cinema, parse the website and give an accurate answer. It’s also very possible for it to do it incorrectly or just pretend that it did that and give you a random time instead so that you are happy.
I can’t imagine trusting an AI with something like this, where it’s so easy to just check it yourself
Going even further, we had this capability already, for example the original Google Assistant was able to do that task just fine.
Yet Gemini similarly seems to struggle with this on mobile.
In the same example, I don’t understand why, that when the agent protocol started happening they didn’t offer the original Google Assistant pathway as an agent of mcp for Gemini to call and add its response to the context.