The agentic-AI glossary.

An open protocol that connects AI applications to external tools and data through one standard interface — so integrations are written once and used by any capable client.

An open protocol for AI agents to discover each other, exchange tasks over JSON-RPC, and return results — across frameworks and vendors.

A machine-readable JSON file at a well-known URL describing what an agent can do and how to reach it — a business card other agents can parse.

Markdown instruction packages served at a well-known URL that teach visiting AI agents how to use a site — documentation agents load exactly when they need it.

A markdown file at a site's root that indexes its content for AI models — robots.txt's cousin, written for machines that read rather than follow links.

How well a public website serves AI agents as visitors — structured facts, machine-readable signals, and the emerging protocols agents can act through.

Commerce where AI agents search, compare, and buy on people's behalf — and where a business has to be legible to a buyer that never sees the page.

Designing what an AI model actually sees — retrieval, memory, tools, user state — instead of writing ever-longer instructions. The discipline that succeeded prompt engineering.

Directing each piece of work to the AI model that fits it — by capability, cost, and speed — with explicit rules for when work escalates.

Deliberately keeping a person at the decision points of automated work — approvals, escalations, and final accountability — while agents do the labor around them.

Systematic tests for AI systems — fixed inputs, graded outputs — that measure whether a change made things better or worse. The safety net under everything else in this glossary.

The mechanism by which an AI model calls external functions — declared with typed schemas — instead of only producing text. The difference between an assistant and an agent.

The standing instruction layer that shapes every response a model gives — role, boundaries, format. Where preferences belong; where information does not.

Employees using AI tools their organization never approved, with data it would never deliberately send anywhere — the most common finding in AI security assessments.

An emerging web standard that lets AI agents identify themselves cryptographically — HTTP requests signed with published keys, so "which agent is this?" stops being guesswork.