AI Agents Inventory and Attributes

The AI Agent is one of the recognized Noun Types in the IF4IT Enterprise Inventory Management taxonomy — the master catalog of every enterprise inventory type the framework governs. For the full taxonomy, the governance principles that apply to every inventory, and the Inventory of Inventories that catalogs them, refer to the IF4IT Enterprise Inventory Management Best Practices document.

The AI Agent is among the most connected Noun Types in the taxonomy, and its closest neighbors fall into three bands. Its core relationships are five: it is built on and driven by an AI or machine-learning model governed in the AI and Machine Learning Models Inventory; it is embedded in, exposed by, or acts upon software governed in the Applications Inventory; it acts through the tools, APIs, and connections governed in the Integrations Inventory; it accesses, reads, and writes data governed in the Data and Information Inventory; and it is owned by, accountable to, and attested by people governed in the People and Personnel Inventory.

A second band of relationships is nearly as defining: an agent runs on frameworks and runtimes governed in the Software Technologies Inventory; it automates or delivers business capabilities governed in the Capabilities Inventory; it is deployed to and runs in the operating environments governed in the Environments Inventory; it is supplied through three vendor roles — model provider, platform provider, and builder — governed in the Vendors Inventory, where the builder may resolve to the enterprise itself; and it is constrained by the laws, regulations, and directives governed in the Regulations Inventory.

Beyond these, the AI Agent has direct secondary relationships to the MCP Servers Inventory, which governs the servers an agent connects to as first-class assets; to the Regulatory Agencies and Regulatory Obligations inventories, reached through the Regulations it must satisfy; to the Security Assets and Controls Inventory, which governs the guardrails, credentials, and controls that bound it; and to the Data Sensitivity Types Inventory, which classifies the sensitivity of the data it handles. Where any related inventory is not yet published, its relationship to the AI Agent is defined in the IF4IT Enterprise Inventory Management Best Practices document.

The AI Agent’s place in the taxonomy is best understood through its dual nature. It is a specialized, application-like software actor — designed, deployed, versioned, owned, and operated much like an application — and it is simultaneously person-like, carrying an individual identity, a role, access, training and attestation, and a named accountability, much like an onboarded worker. Its neighbor set was derived from that dual nature rather than inherited from any single parent Noun Type: the application-like facet gives it the deployment, technology, integration, and environment relationships; the person-like facet gives it the identity, accountability, attestation, and regulated-conduct relationships that an application does not carry. This is why the AI Agent is its own Noun Type and not a subtype of the Application — it is neither the model it runs on, nor the application that hosts it, nor the integrations it uses, nor a bare non-human identity; it is the governed actor that ties them together and acts.