← All sections
§15

The central AI registry

Single source of truth for every AI asset in the company — built or bought, internally orchestrated or vendor-embedded.

You cannot govern what you cannot see. Every source article makes this point.

15.1 Registry schema

FieldTypeNotes
Agent nametextHuman-readable
Agent IDtextStable slug
SourceenumInternally built / Vendor-embedded / Hybrid
VendortextIf vendor-embedded — which SaaS product
OwnerpersonDepartment Champion
DepartmentenumSales / Finance / HR / Ops / etc.
StatusenumIdea / Intake / Approved / Build / Pilot / Production / Retired
Risk tierenumLow / Medium / High
Risk driver — PIIboolean
Risk driver — Consequential decisionboolean
Risk driver — Autonomousboolean
Jurisdictional exposuremulti-enumEU / US state / Canada / UK / etc.
Autonomy stageenumAssistive / Validated / Autonomous
Approved byperson + dateLast approval
Platformenumn8n / LangGraph / vendor / etc.
LLM(s) usedtextWith versions
Monthly cost (est.)numberAPI + infra + observability
ROI metrictext"20 hrs/week saved" / "12% conversion lift" / etc.
Last revieweddateQuarterly minimum
Source repo linkURLFor internally built
Monitoring linkURLDashboard
Audit log locationURLFor Medium/High
NotestextFree-form

15.2 Vendor-embedded AI is in scope

AI features inside the SaaS tools we already use (CRM, ERP, ticketing, doc tools, email, code copilots, recording tools, HR systems) are the largest single source of shadow AI. They ship via auto-update, often silently.

Operating rules:

  • Catalog every SaaS app with AI features quarterly.
  • Tag each one on the three risk drivers (§10.2).
  • Treat AI inside vendor tools the same as internally built agents — same Agent Card shape, same risk tier, same approval path before "turn on the AI feature" is approved for a department.
  • Refresh quarterly. Vendors ship new AI features regularly without telling customers.