Maintaining AI Agent Context Across Incident Updates

[craigrowlson]

I am currently designing an L1 Service Desk AI agent in ServiceNow that communicates asynchronously through Incident additional comment field rather than through a live Teams or through the Now Assist agent on the Service Portal.

The current design works like this:

The agent reviews the Incident, asks the caller for any missing information, and posts the questions back to the ticket. Once the caller responds, a workflow triggers the agent again so it can review the new response, pick up where it left off, and continue working the issue until it is either resolved or ready to be escalated.

This allows the agent to gather information over time without requiring a live conversation. At this point, I do not have the agent communicating through Teams or a Service Portal live stream/chat as we would like to pilot this for after-hours support. The interaction is more of an async ticket-based conversation using the Incident record as the communication thread.

My question is around Now Assist/AI Agent cost and session handling.

Is there a way to keep these async interactions tied to the same conversation/session so the agent can maintain context without starting a new assist/session every time the workflow re-triggers? Or is each workflow-triggered review of the updated Incident treated as a separate assist interaction from a consumption/cost perspective?

I am trying to understand the best design pattern for reducing assist cost while still allowing the agent to:

• Ask the caller for missing details
• Review the caller’s response later
• Continue from the prior step
• Avoid repeating questions
• Work the issue through completion

Has anyone designed a similar async AI agent workflow using Incident comments/work notes instead of live chat? If so, how did you handle session continuity, context persistence, and cost optimization\consumption?