Comment by CityOfThrowaway

Situationally, the original post claims that LLM Agents cannot do the tasks well. But they only tested one agent and swapped out models.

The conclusion here is that the very specific Agent that Salesforce built cannot do these tasks.

Which frankly, is not a very interesting conclusion.