Agent Readiness Glossary: 50 Terms Every Business Should Know

Agent-to-Agent protocol (v0.3) that defines how AI agents discover and delegate tasks to other AI agents. Uses agent-card.json for discovery. Different from MCP, which is agent-to-tool.

A protocol for structured message passing between AI agents, enabling multi-turn conversations and task negotiations beyond simple tool calls.

A JSON file (agent-card.json) placed at /.well-known/agent-card.json that describes an AI agent's or business's capabilities, supported protocols, and interaction methods. Zero of 500 businesses scanned have one.

The emerging economic layer where AI agents discover, evaluate, and transact with businesses on behalf of humans. Estimated at $3-5 trillion by 2030. Businesses invisible to agents are excluded from this economy.

The ninth scoring dimension (10% weight) measuring how pleasant an API is for agents to use. Includes request IDs, structured errors, response envelopes, rate-limit headers, cursor pagination, idempotency keys, and OpenAPI examples.

The 6-step path an AI agent follows when interacting with a business: Find, Understand, Sign Up, Connect, Use, Pay. Most businesses fail at step 1.

A business or API designed from the ground up for AI agent interaction, not retrofitted from a human-first interface. Agent-native businesses typically score 60+ on the Agent Readiness Score.

A 7-level classification system (ARL-0 through ARL-6) measuring how prepared a business is for AI agent interaction. ARL-0 is Dark (invisible). ARL-6 is Interoperable (full agent ecosystem participant).

A 0-100 composite score measuring how well a business can be discovered, understood, and used by AI agents. Calculated across 9 weighted dimensions. Average score across 500 businesses: 43/100.

A markdown file placed at the root of a repository or website that describes the project's capabilities, tools, and workflows in a format optimized for AI agent consumption. The README.md equivalent for agents.

The second scoring dimension and highest-weighted at 15%. Measures OpenAPI spec availability, endpoint structure, response format consistency, and API documentation quality.

An authentication mechanism where an API key or access token is passed in the Authorization header (Authorization: Bearer <token>). The preferred auth method for AI agents because it is stateless and programmatic.

Agent Readiness Score of 40-59. The business has basic digital infrastructure — HTTPS, some structured data, possibly an API — but lacks agent-specific features like agent cards or MCP servers.

An OAuth 2.0 grant type where an application authenticates using its own credentials (client_id + client_secret) rather than on behalf of a user. The correct OAuth flow for machine-to-machine agent authentication.

HTTP headers that control which domains can make API requests. Misconfigured CORS blocks agents from different origins. Agent-ready CORS allows credentialed cross-origin requests and exposes rate-limit headers.

A pagination method using opaque tokens (cursors) instead of page numbers. More stable than offset pagination because it is not affected by data changes between pages. The agent-preferred pagination pattern.

The nine scoring dimensions of the Agent Readiness Score: D1 Discoverability (0.12), D2 API Quality (0.15), D3 Onboarding (0.08), D4 Pricing (0.05), D5 Payment (0.08), D6 Data Quality (0.10), D7 Security (0.12), D8 Reliability (0.13), D9 Agent Experience (0.10).

The lowest Agent Readiness Level. The business is completely invisible to AI agents — no API, no structured data, no machine-readable content. Score: 0-19. 40% of businesses scanned fall here.

The sixth scoring dimension (10% weight). Measures structured response formats (JSON vs HTML), consistent error envelopes, JSON-LD schema markup, and machine-readable content.

The first scoring dimension (12% weight). Can an agent find your business at all? Checks DNS, robots.txt allowing GPTBot, sitemap.xml, agent-card.json, llms.txt, and OpenGraph tags.

A standardized JSON format for API error responses containing an error message, machine-readable code, HTTP status, and request ID. Example: { "error": "Invalid amount", "code": "invalid_amount", "status": 422, "request_id": "req_abc123" }.

Agent Readiness Score of 75-89. The business has comprehensive agent infrastructure including MCP server, agent card, structured errors, and self-service onboarding. Only 1 of 500 businesses scanned (Resend, 75) achieved Gold.

An API endpoint (typically /health or /status) that returns the current operational status of a service. Agents check this before making requests to avoid wasting calls on a down service.

A cryptographic method for verifying webhook authenticity. The sender signs the payload with a shared secret; the receiver verifies it. Prevents webhook spoofing. Used by Stripe, GitHub, and other agent-ready platforms.

A unique identifier sent with API requests that prevents duplicate operations on retry. If an agent sends the same request twice with the same idempotency key, the API returns the cached result instead of processing it again.

JavaScript Object Notation for Linked Data. A method of encoding structured data (Schema.org markup) in JSON format within HTML pages. Agents extract business identity, pricing, hours, and services from JSON-LD blocks.

An identity verification framework for AI agents interacting with business APIs. Defines agent types (autonomous, supervised, delegated) and trust levels for different operations.

A markdown file served at the root of a website (/llms.txt) that provides a concise, AI-readable description of the business, its API, and how to interact with it. The robots.txt equivalent for AI models. Fewer than 5% of businesses have one.

An open standard created by Anthropic that defines how AI agents discover, connect to, and interact with external services. Exposes tools, resources, and prompts. The HTTP of the agent economy.

A server implementing the Model Context Protocol that exposes tools (callable functions), resources (readable data), and prompts (interaction templates) for AI agents. The equivalent of a website but for agents instead of humans.

A callable function exposed by an MCP server. Each tool has a name, description, and typed input/output schema. Examples: get_menu(), check_availability(), create_booking().

Agent Readiness Score below 40. The business has insufficient digital infrastructure for meaningful agent interaction. Used in user-facing text instead of "Failed" or "Unaudited."

Natural Language Web — a protocol that allows agents to query websites using natural language and receive structured responses. AgentHermes supports NLWeb queries at /api/nlweb.

An authorization framework that enables third-party applications (including AI agents) to obtain limited access to APIs. The client_credentials grant is the agent-preferred flow because it requires no human in the loop.

The third scoring dimension (8% weight). Can an AI agent sign up for API access, get credentials, and start making calls without a human? "Contact sales" is a D3 score of zero.

A standard format (formerly Swagger) for describing REST APIs in YAML or JSON. Includes endpoints, parameters, request/response schemas, and authentication methods. The single biggest factor in Agent Readiness — D2 is weighted 15%.

The fifth scoring dimension (8% weight). Can an AI agent complete a purchase end-to-end via API? Requires programmatic payment methods, not hosted checkout redirects.

Agent Readiness Score of 90-100. Full agent ecosystem participation with MCP server, A2A protocol, x402 micropayments, and cross-agent interoperability. Zero of 500 businesses scanned have achieved Platinum.

The fourth scoring dimension (5% weight, lowest). Whether pricing is machine-readable — structured JSON/JSON-LD, not a PDF or "contact for quote." 30% of businesses fail D4 completely.

Throttling API requests to prevent abuse. Agent-ready rate limiting includes machine-readable headers (X-RateLimit-Limit, X-RateLimit-Remaining, X-RateLimit-Reset) and 429 responses with Retry-After.

The eighth scoring dimension (13% weight, second-highest). Measures status pages, health endpoints, uptime history, incident tracking, and SLA documentation. Agents automate repeat actions, so unreliable endpoints kill adoption.

A unique identifier (X-Request-ID header) returned with every API response. Enables agents to correlate requests with responses, debug failures, and reference specific transactions in error reports.

A test version of an API with fake data where agents can learn endpoints without risking real money or data. Stripe's test mode (sk_test_*) is the gold standard. Most businesses do not offer one.

A collaborative vocabulary for structured data markup on web pages. JSON-LD Schema.org markup (Organization, Product, Service, Offer) lets agents extract business identity and offerings without API calls.

The seventh scoring dimension (12% weight). Measures authentication method (Bearer preferred), HTTPS enforcement, OAuth support, CORS configuration, and credential management.

Agent Readiness Score of 60-74. The business has strong API infrastructure, documentation, and security. Developer-focused companies dominate Silver. 22 of the top 30 Silver scorers are dev tools.

A transport protocol used by MCP servers to stream real-time data to agents over HTTP. Allows long-running tool calls and progress updates. The standard MCP transport alongside stdio.

A public page (typically status.domain.com) showing current service health, incident history, and component status. Directly impacts D8 Reliability (13% of score). Agents check status before making API calls.

The encryption protocol behind HTTPS. A hard cap in the Agent Readiness scoring model: no TLS means the score cannot exceed 39. Without HTTPS, all data between agent and API is interceptable.

An emerging standard for sharing context (user preferences, session state, conversation history) between AI agents and services. Detected by AgentHermes but not yet widely adopted.

Industry-specific adjustments to dimension weights in the Agent Readiness Score. A restaurant weighs D4 Pricing higher than a SaaS company. AgentHermes supports 27 vertical profiles.

An HTTP callback that pushes event data from a service to a registered URL when something happens (order placed, payment received, status changed). Agents prefer webhooks over polling because polling wastes compute budget.

A micropayment protocol that uses the HTTP 402 Payment Required status code. Enables AI agents to pay for API calls per-request using USDC with sub-second settlement. No signup or credit card needed. The missing piece for ARL-4 Automated.