What is an AI agent inventory?

The scale problem is not hypothetical

Gartner projects the average Fortune 500 firm will run more than 150,000 AI agents by 2028, up from roughly 15 in 2025. That is not a number you track in a spreadsheet a team admin updates when they remember to. Mid-market operators run far fewer agents in absolute terms, but the underlying pattern is identical at any scale: agents accumulate through vendor deployments, internal builds, low-code tools, and individual experimentation, and nobody was assigned to keep the list.

The governance data backs up how wide the gap already is. Only 13% of organizations believe their agent governance is adequate, and only 21% report mature agent governance today, against 74% that expect broad agent use within two years (Deloitte, n=3,235). Inventory is upstream of every other governance capability. You cannot govern, cost-attribute, or produce evidence for an agent you don’t know exists. See agentic sprawl for how this compounds once agents start operating unmonitored.

Declared inventory plus discovered inventory

Centurian builds the inventory two ways at once. Declared: every agent registers through the MCP front door, whether it runs on Salesforce Agentforce, AWS Bedrock, Microsoft Foundry, Google Vertex, Copilot Studio, or custom code, one registry regardless of vendor, which is the structural difference from a hyperscaler's native agent registry that only sees its own platform's agents.

Discovered: activity signals, starting with AWS CloudTrail, are reconciled against the registry. Anything active in the signal but absent from the registry surfaces as a shadow agent, with notification routed to the team admin first and escalation to the org admin after seven days. The inventory is the entry point to agent identity, cost tracking, and the audit trail, every downstream evidence product depends on the list being complete first.

FAQ