A CMDB answer is only as useful as your ability to prove where it came from. Health scores, ownership, dependencies, and findings usually live in separate views, so the moment someone needs to act on what the CMDB is telling them, the investigation turns manual and hard to audit.
Agent Maestro, built by Adam Nugroho for KeenStack’s internal StackUp hackathon, is a working answer to that problem. It’s an AI agent that investigates configuration items, explains what’s wrong using real CI facts and relationship data, and routes a fix to a person for approval, all without ever writing to the CMDB itself.
How it works
Agent Maestro runs on five read-only tools built in ServiceNow’s AI Agent Studio, grounded entirely in the CI facts, health findings, and relationships already sitting in the CMDB. The workflow moves through four steps: find configuration items that need attention, explain the finding using cited relationship context, prove every claim against the exact ServiceNow record it came from, and route the fix as a scoped remediation task for a person to review.
Nothing gets fixed automatically. A reviewer sees the recommendation, the score impact, and the evidence behind it, and has to explicitly approve it before a remediation task is created. The scoring itself is deterministic rather than model-generated, so the same CI produces the same score every time, and the AI’s role is limited to explaining the finding and drafting the recommendation around it.
Why it matters
Most CMDB investigations today mean stitching together several views by hand: one for health, another for ownership, another for dependencies. Agent Maestro replaces that with one guided investigation, which is where the time savings actually come from.
The bigger win is trust. Every cited CI is checked against its exact sys_id and record marker before it ever reaches a reviewer, so nothing gets surfaced that can’t be verified in a click. Reviewed evidence turns directly into an assigned, auditable task instead of a note someone has to remember to follow up on. And because the pattern never writes to a CI and scores deterministically, it holds up the same way whether it’s running against 30 configuration items or a much larger production CMDB.
What the pilot showed
Run against a scoped set of 30 configuration items and 40 relationships, Agent Maestro surfaced six items needing attention, mostly missing owners and unverified dependencies that don’t show up cleanly in a standard health view. Thirty regression tests passed before the tool touched a live instance, and the live deployment does an exact read-back against the source system with no CI write path at all.
Agent Maestro is one way to bring evidence-backed investigation to a CMDB. If yours could use the same solution, reach out to KeenStack.
Built by:
Adam Nugroho
ServiceNow Developer