I’ve been experimenting with Agentic AI to deploy a custom project on ServiceNow using strong, structured prompts — exactly as ServiceNow recommends. The AI did generate multiple code files and attempted several deployment approaches automatically, but none of them successfully deployed or ran inside the platform.
Finally, the AI suggested copying all the generated HTML code into a UI Page manually. But as we all know, ServiceNow UI Pages require XML Jelly formatting, and the JavaScript must be separated appropriately. That part still needed full human intervention.
In the process, I ended up using 9 out of 10 available prompts just to troubleshoot a single deployment attempt.
This made me wonder:
Is this challenge common in the community?
Is anyone else facing issues with Agentic AI deployment accuracy on ServiceNow?
Have you found effective ways to improve prompt limits?
Right now, with only 10 prompt attempts, it becomes difficult to iterate and troubleshoot efficiently.
Any suggestions on how to get more prompt capacity or handle complex deployments better?
My learning from this:
Agentic AI is powerful, but for now, human expertise is still essential, especially for XML Jelly, UI Builder, and platform-specific deployment logic.
Would love to hear insights, tips, or similar experiences from the ServiceNow community!
