AIATS Insights · Topic
ServiceNow platform guidance: AI agents, ITSM, CMDB, SDK and release playbooks.
Australia's Change Data Quality AI Agent now persists scores in ai_change_quality_score. A UK CAB production gate: policy documents, supervised vs autonomous versions, approval thresholds, and Platform Analytics — without letting agents rewrite RFC fields in PROD.
Code-first ServiceNow only works if the guardrails are as deliberate as the tooling. This part sets out environment tiers, an AI agent access model where agents work in LAB, PDI and DEV but never touch PROD without human approval, secret handling, audit evidence, and a balanced guide to when Fluent is right and when native ServiceNow development is the better tool.
What code-first ServiceNow development looks like day to day: a Fluent project layout, short TypeScript examples that compile with SDK 4.13.0, the build and deploy loop, branching and pull requests, CI quality gates, ATF, and a promotion path from DEV to TEST to PROD that keeps the SDK away from production.
Code-first ServiceNow development needs a workspace you can rebuild in minutes and trust with credentials. Here is how I set one up: an Ubuntu 24.04 host, Docker, a Coder workspace template, Node 24, a pinned ServiceNow SDK and instance authentication that never touches Git.
Building ServiceNow apps by clicking through forms works, but it is hard to review, hard to diff and hard to hand to an AI agent safely. This is the architecture I'm using instead: Coder workspaces, Node 24, the ServiceNow SDK with Fluent, Git, pull requests, ATF and gated promotion from DEV to TEST to PROD.
Zurich pairs Developer Sandboxes with ReleaseOps pipelines so UK teams can build in parallel and promote with CAB-grade evidence. A platform production gate covering MIM, analyzer checks, ATF hooks, and upgrade/clone sandbox hygiene.
AI Gateway GA (10 Sep 2026) inside AI Control Tower enforces MCP connections with OAuth 2.1, tool-level governance, and block-not-mask sensitive-data controls. A UK Steward CAB checklist for production.
AI Control Tower Aug/Sep 2026 adds runtime observability and LLM-as-a-judge evaluations (including Traceloop for external agents). UK Steward playbook: Zurich P11 / Australia P4+, managed inventory, Trace Collectors vs SDK, Monitor dashboard — not Kill Switch.
AI Control Tower Aug/Sep 2026 ships a Kill Switch that revokes agent credentials across Okta, GCP and AWS Bedrock and stops agents at runtime — cutting mean time to contain from ~30 minutes to seconds. UK CISO playbook before you treat it as production IR.
Australia / Zurich P11+ Dynamic IRE auto-scores Hardware CI identity and can arrive committed on zBoot. UK playbook: simulate on non-prod, review cmdb_ire_output_comparison_item parity, exclude custom classes, then CAB-commit — with Change to IRE rollback.
From 9 Jul 2026, OOTB Now Assist / Otto skills switch to third-party model defaults with each store app update. UK estates need an AI Control Tower allowed-provider gate, CAB dual-run evidence and a Now LLM retirement plan — before Assists look fine and skills are blocked.
September 2026 reimagines AI Agent Studio with guided build, Advisor and side-by-side test. UK gate: Zurich P13+/Otto 9.0.8+, CAB, GDPR, sn_aia.admin, Prime/Pro Plus and human-in-the-loop.