Hey everyone. For the booking assistant here, https://discord.com/channels/1211482211119796234/1211483571454218261/1246263371368497202, here's the exact prompt I used. For some reason, after transitioning from GPT-4 to GPT-4o, a previous prompt I was using was no longer as effective, so I had to go back and rewrite to finally this. I had to capitalize specific statements so that GPT4-o would more reliably infuse those portions into the conversation. Also, you'll note that there is no phrase "range" in the prompt as noted that GPT-4/GPT-4o struggled with that concept since if we say "2 pm to 4 pm", we mentally understand it to be inclusive to 4 pm, however, the natural tendency for these models noted during testing is for them to assume 4 pm is not inclusive; so removing "range", capitalizing key statements, and providing those bracket examples made noticeable improvements to the AI logic. Also, though I did not have to provide a specific guard clause going from Task Step 6 to Step 7 and the webhook response only says "Here are the available times for X", was refreshing to see the AI was smart enough to prevent illegal times from going through (most of the time). Ultimately, you'll need guard clauses in your booking appt webhook to catch those exception instances and prevent booking overlaps or illegal times since if you keep going back and forth trying to force a failure it will finally give up and deviate from it's prompt logic 😄 . I easily spent weeks refining this prompt to this so hopefully it helps anyone struggling with a strong baseline prompt to use. Hopefully others will provide their refined prompts and associated reasonings here as well to help others along the way!