Using Update Sets for AI Agents Development

[Max Dore]

Overview

When configuring AI Agents in ServiceNow, their definitions and related artifacts can be captured in update sets just like other platform records. This article explains what to expect in an update set when working with AI Agents, which elements are captured, which are not, and what to check during migration.

This guide will use the HRSD product for examples to help explain some of the records and their dependencies. Depending on your business requirements some record types may not be part of your update sets; of course, there could be more too, such as a requirement for an integration.

 

Record Types

• Use Case — sn_aia_usecase
  These are the Agentic Workflows you see when you navigate to AI Agent Studio -> Create and manage. They're the top-level wrapper for the business use case that groups agents, tools, and triggers.
  Example: "Resolve noncritical HR cases","Resolve policy for Performance Reward".

• AI Agent — sn_aia_agent 
  These are the core agent records. They hold the agent’s name, role and and logic such as what the agent is designed to do with markdown-like instructions showing the step by step process and possible error handling. Expect these XMLs to have large payloads, tool references, and user interaction flows.
  Examples: "Performance Reward HR case closure evaluator", "Performance Reward – case details manager".

• AI Agent Config — sn_aia_agent_config 
  These are activation/config records that tie an agent definition into the platform. They mark whether an agent is active active=true and link the agent sys_id. Without these, agents exists but aren't active.

• Tool — sn_aia_tool
  Callable tools that agents can use (for retrieval, validation, printing final conclusions, etc.).
  Examples: "fetch approval tasks", "Notify Manager", "Get the Performance Reward Policy".

• Agent Tool — sn_aia_agent_tool_m2m
  Relationships declaring which tools an agent can call.

• AIA Trigger Configuration — sn_aia_trigger_configuration
  Determines where/when the use case is invoked (e.g.,table=sn_hr_core_case_total_rewards, state=ready).

• Trigger M2M — sn_aia_trigger_agent_usecase_m2m
  Associates trigger configurations to specific Agentic Workflows.

• Team — sn_aia_team
  A group for orchestrating a set of agents that collaborate on the Agentic Workflows.

• Team Member — sn_aia_team_member
  Individual agents linked to the Team.

• AI Agent Skill MetaData — sn_aia_skill_metadata
  Metadata referencing a Document table and Document ID.

• Generative AI Skill — sys_gen_ai_skill 
  Defines a Generative AI Skill (Now Assist) that can be surfaced in the UI. One per Agentic Workflow.

• Generative AI Skill Config — sys_gen_ai_skill_config
  Related to the Generative AI Skill record, the Config piece contains parameters such as display and active.

• Generative AI Skill Applicability — sys_gen_ai_skill_applicability 
  Another record related to the Generative AI Skill record. This one contains conditions that determine when the skill is offered and the roles (e.g., now_assist_panel_user) that can access it.

• Flow — sys_hub_flow 
  Supporting Flows tied to the Agentic Workflow. These match the sn_aia_trigger_configuration records and go hand in hand.

• Client Script — sys_script_client 
  Client-side UI logic (e.g., "Show AIA notification" on HR case forms). These records, like many others, are created automatically and developers are likely to not need to make changes.

 

What Does Not Get Captured

• Transactional/runtime data (tests, HR cases, attachments, approvals, user-generated content).

• External dependencies that were not changed in this scope (e.g., other app-scoped tools or parameters untouched by this work).

 

Migration Procedure

1. Pre-checks: Confirm the HR Service Delivery AI Agent collection (and any required plugins) is installed in the target instance and on the same version.

2. Preview and import: Import the update set, run preview, and resolve collisions or missing dependencies.

3. Post-import validation:

   • Ensure intended agents are Active.

   • Verify that the Agentic Workflow and Trigger Configuration are present and pointing to the intended tables.

   • Spot check Agent to Tool relationships to confirm all required tools are present and linked.

   • Open the relevant HR Case (or Incident, Case, etc..) form and validate Client Scripts (e.g., notification banners) and any additional functionality behave as expected.

   • Execute the use case end-to-end (e.g., create a Total Rewards HR case and confirm the trigger offers the correct AI Agentic Workflow or Skill).

 

Best Practices

• Build and capture all AI Agent artifacts in the same application scope and update set to minimize risks and conflicts.

• Keep the use case cohesive: agents, tools, triggers, team, skills, etc... should align to one functional package.

• Review the update set contents prior to promotion; remove incidental records that are not part of the release

• After import, validate:

  • Agent activation

  • Tool availability and M2M links

  • Trigger applicability on target tables

  • UI behavior such as Client Scripts

  • Flow execution paths

Summary

AI Agent work promotes cleanly via update sets, but it involves a broader package than just the agents themselves: tools and their relationships, Agentic Workflows (aka use cases) and trigger configurations, team structure, supporting UI, flows, and more. Reviewing these record types before promotion and validating them after import will make migrations predictable and low risk.

I hope this article has been useful to you and please let me know if you'd like me to add something you feel is missing.

Happy developing!