This glossary collects the terms used across AI Hive and defines each in plain language. It is arranged alphabetically so you can find a word quickly, and it grows as new terms enter the site.

​

A2A (Agent-to-Agent protocol). An open standard, introduced by Google and now developed under the Linux Foundation, that lets agents from different makers discover one another, describe what they can do, and collaborate on a task.

​

Agent card. A machine-readable description an agent publishes so that others can learn what it does, how to reach it, and what it needs, much like a business card that other software can read.

​

Agent economy. The emerging world in which AI agents find each other, agree on work, and pay one another, often with no person directing each exchange.

Agentic. A general word for systems that act with autonomy toward a goal, as opposed to systems that simply respond when prompted.

​

AI agent. Software that can pursue a goal on its own, make decisions, use tools, and take actions, rather than only answering a single request.

​

AP2 (Agent Payments Protocol). A standard from Google focused on how a person or business authorizes an agent to spend money on their behalf, and within what limits.

Autonomous. Acting without a human directing each step. In this context it usually means acting within set limits and oversight, rather than being entirely unchecked.

Delegated authority. A grant of permission that lets an agent act, or spend, on behalf of a person or business, bounded by explicit limits. It is the difference between handing an agent a task and handing it a blank cheque.

​

ERC-8004. An emerging Ethereum-based standard for giving agents verifiable identities and reputations, so that one agent can prove who it is to another.

​

Guardrails. The limits and checks placed around an agent's behaviour, such as spending caps, permission boundaries, and points where a human must approve an action.

Human in the loop. An arrangement in which a person must review or approve certain actions before an agent carries them out, reserved for steps where the stakes are high enough to warrant it.

​

Large language model (LLM). The kind of AI system, trained on vast amounts of text, that underlies most current agents and gives them their ability to interpret instructions and generate language.

​

Machine-to-machine. Describing activity, especially payments, that happens directly between two pieces of software with no person involved.

​

MCP (Model Context Protocol). An open standard, introduced by Anthropic, that gives an agent a consistent way to connect to tools and data sources.

​

Multi-agent system. A setup in which several agents work together, dividing a task among themselves, which increases what is possible and also multiplies the ways things can go wrong.

​

Orchestration. The coordination of multiple agents or steps toward a single goal, deciding which agent does what and in what order.

​

Prompt injection. An attack in which hidden or malicious instructions trick an agent into doing something it should not, a leading security concern for agents.

Reputation. A verifiable record of how an agent has behaved over time, used by others to decide whether to trust it.

​

Stablecoin. Digital money designed to hold a steady value, usually pegged to the dollar, well suited to the fast, tiny, borderless payments that pass between agents.

Tool use. An agent's ability to reach beyond generating text and actually do things, by calling external services such as a search engine, a database, or another program.