Appendix C
Glossary
| Term | Meaning |
|---|---|
| A2A | Agent2Agent protocol: how agents discover each other and hand off work. |
| Agent 365 | Microsoft’s agent control plane for its own estate, with Microsoft Entra Agent ID for agent identities. |
| Agent Broker | MuleSoft Agent Fabric capability that routes multi-step work across agents and tools, with guided determinism for fixed hand-off rules. |
| Agent Kill Switch | MuleSoft Agent Fabric capability that stops agent activity quickly. |
| Agent lifecycle | The stages an agent passes through: proposal, registration and onboarding, evaluation, promotion, monitoring, recertification and retirement. |
| Agent Registry and Scanners | The single source of truth for agents, APIs and MCP tools, with automatic discovery across ecosystems. |
| Agent sponsor | The named person accountable for one agent from registration to retirement, including recertification and the first call in an incident. |
| Agent Visualizer | MuleSoft Agent Fabric capability that shows how agents, tools and systems interact, with confidence, bottleneck and hallucination-risk signals. |
| AI Gateway | The one policy point for LLM and MCP traffic: routing, token-based rate limits, masking, spend attribution and logging. |
| Amazon Bedrock AgentCore | AWS services for running agents: a runtime, memory, a gateway that connects MCP servers, identity and observability. |
| Audit mode | Running a new policy so that it records what it would block before it enforces anything. |
| Autonomy tier | The level of authority an agent has (assisted, supervised, autonomous, multi-agent) and the controls each level requires. |
| Bound approval | An approval tied to one exact action and valid for a limited time. The gateway refuses the action if either has changed. |
| Control plane | The governed layer between everything that consumes intelligence and everything that provides it. |
| Cumulative value | The sum of each milestone’s run rate over the months it was in force. |
| Data contract | The schema, meaning and quality thresholds that an MCP tool guarantees. |
| FOCUS | The FinOps Open Cost and Usage Specification: a common format for cloud and AI billing data. |
| Golden record | The trusted, de-duplicated master record for an entity such as a client, worker or asset. |
| Informatica | The data foundation for AI agents: master data, data quality, catalog and governance, with data services exposed to agents through MCP servers. |
| Leading indicator | An early signal, defined the same way by every team, that moves within weeks and predicts a scorecard metric. |
| LLM Governance | MuleSoft Agent Fabric capability that gives central visibility into token usage, cost and data flows for third-party models. |
| MCP | Model Context Protocol: how agents reach tools and data. |
| MCP Bridge | MuleSoft capability that turns existing APIs into governed MCP tools. |
| MuleSoft Agent Fabric | The control plane that registers, governs, orchestrates and observes agents and MCP endpoints, whatever platform built them. |
| Multidomain MDM | Informatica master data management for domains such as customers, suppliers, products and assets. |
| North star | The vendor-neutral target reference architecture that every decision is measured against. |
| Omni Gateway | MuleSoft gateway that applies policies to APIs, A2A and agent traffic, including custom policies built with the Policy Development Kit. |
| On-behalf-of token | A delegated token that lets an agent act for a specific person with narrowed scopes. |
| OWASP Top 10 for Agentic Applications | The OWASP list of ten agentic security risks, ASI01 to ASI10, from agent goal hijack to rogue agents. |
| Policy Development Kit (PDK) | The kit for writing custom Omni Gateway policies in Rust, compiled to WebAssembly. |
| Quick win | A first slice of a use case that shows measurable value within about 90 days, runs read-only or propose-only, and leaves a reusable foundation behind. |
| Recertification | A scheduled review in which an agent’s sponsor confirms that it is still needed, in the right tier and with the right permissions. |
| Run rate | Annualized value being delivered at a milestone. |
| Runtime Fabric | Customer-managed MuleSoft runtime for running gateways and MCP servers inside the client’s own environment. CloudHub is the MuleSoft-managed alternative. |
| Signed agent card | An A2A agent card signed by its issuer, so another agent can verify who it is dealing with. |
| System of record | The authoritative application for a business object, such as SAP S/4HANA for projects and billing. |
| Takeover runbook | The plan for who takes over an agent’s open work when it is stopped, with what information, and how the process continues. |
| Trajectory evaluation | Judging an agent by the tools it chose and the order it used them in, not only by its final answer. |
| Trusted Agent Identity | MuleSoft Agent Fabric capability that lets agents act with a specific user’s identity and authorization, with mobile approval for high-risk steps. |
| Work mode | Whether a step is human-led, AI-assisted or AI-led, decided by asking whether AI could do it and whether it should. |
| Zero trust | A security model in which every actor is authenticated and every call is authorized for the specific task. Nothing is trusted because of where it sits on the network. |