May '25 • 🛠️ dev-help
Regarding AI Agents - not executing steps properly
Has anyone found that the AI Agents don't just hallucinate they randomly adhere to tool creation even on the same type content?
I have an Inbox Manager that shows in the log it knows what to do but it just fails to call the appropriate tool.
Example:
Gmail Trigger brings the email in.
It recognizes the content and says its applying the label (but the Add Label tool was never called)
It then removes the Inbox label (which it does call the Tool for and works)
I can run this on the same content types and get different results. I have written a lengthly prompt in markdown so not sure why its so inconsistent.
Here's the the example in an execution:
- Categorization: 0 - Questionable
- Action: Labeled and archived
The email has been marked as unread and removed from the IMPORTANT and CATEGORY_PERSONAL categories.
However, it only removed the Inbox filter and left it unread and without the new label. So it used the tools in some of the actions but then not for what it knew it was supposed to do.
I'll continue to tweak the prompt but I believe its a matter of clearly getting the AI Agent to use the tool when it knows its supposed to.
2
7 comments
Regarding AI Agents - not executing steps properly
skool.com/learn-ai
Learn and become genuinely proficient in AI to get paid for it. From the creators of the AI Automation Agency model.
Leaderboard (30-day)
1
+398
2
+368
3
+358
4
+267
5
+243
Powered by