Thanks again for the launch of Appfarm AI 
Some initial feedback:
It does a lot of good stuff, but if the prompt is rather big, it failed. I assumed it was Claude API being overloaded since this was on the day after the event, but the error persisted on retries the following days, so eventually I created a new thread with smaller prompts, and then it got going. It would be nice to see the technical error when the agent fails, or some error that lets us know what is going on so we know if we need to wait or change our prompts. Just in general it helps seeing it.
Also, in the demo we saw Thursday it seems to do big picture tasks, but when we get more specific, it seems to not always do what it says it did.
Here’s one example:
In the action “Send Prompt to AI”, we are catching all exceptions. Where we catch all exceptions, add result mapping to that catch, so that we get the technical messages from the web request and stored it in the Exceptions data source, before we persist the exception.
With the response:
However it didnt do this. It reasons well, but it did not make any changes (via actions to save). Also tried refreshing.
Let me know if I should have worded my prompt differently. I know prompting is a mix of art and science 