{ "title": "The Agentic Web Lexicon", "description": "Canonical, concise definitions of the terms that make up the agentic web (June 2026). Written to be quoted: each term has a one-line short_def for citation and a longer long_def for context. This enriched edition adds full EAV depth per term — etymology/origin, related terms, nearest-neighbour contrast, a dated example, an authoritative source, status, why-it-matters, sameAs links, the Almanac bridge entity, last_verified date and a markdown-twin path.", "updated": "2026-06-15", "fields": [ "id", "term", "category", "short_def", "long_def", "see_also", "etymology_origin", "related_to", "contrast_with", "example", "source", "status", "why_it_matters", "sameAs", "bridge_entity", "last_verified", "md_twin" ], "records": [ { "id": "ai-agent", "term": "AI Agent", "category": "core", "short_def": "A software system that uses a language model to pursue a goal by reasoning, planning and taking actions through tools.", "long_def": "Unlike a single prompt-and-response, an agent runs a loop: it observes, decides on an action (often a tool call), executes it, observes the result, and repeats until the goal is met. Autonomy and tool use are the distinguishing features.", "see_also": [ "tool-use", "agentic-loop", "mcp" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/Intelligent_agent — 'agent' in AI predates LLMs (Russell & Norvig, AIMA, 1995); the modern LLM-agent sense has no single coiner" }, "related_to": [ "tool-use", "agentic-loop", "mcp", "agentic-web" ], "contrast_with": "Unlike a chatbot, which returns a single response to a prompt, an AI agent runs an autonomous observe-decide-act loop and takes actions through tools to reach a goal.", "example": "Claude Code and OpenAI's Operator (launched January 2025) are AI agents: they plan, call tools and iterate toward a goal rather than answering once.", "source": "https://en.wikipedia.org/wiki/Intelligent_agent", "status": "active", "why_it_matters": "The AI agent is the first-class visitor the agentic web is built for; understanding it is the precondition for every readiness decision a site owner makes.", "sameAs": [ "https://en.wikipedia.org/wiki/Intelligent_agent" ], "bridge_entity": "agentic-web", "last_verified": "2026-06-15", "md_twin": "/glossary/ai-agent.md" }, { "id": "agentic-web", "term": "Agentic Web", "category": "core", "short_def": "The web reimagined for AI agents as first-class visitors — machine-readable content, callable tools, agent identity and agent-native payments.", "long_def": "Where the traditional web optimized pages for human eyes and search-engine crawlers, the agentic web adds a parallel layer agents can read and act on: markdown twins, structured data, MCP/WebMCP tools, signed agent identity, and inline payment protocols.", "see_also": [ "agent-experience", "mcp", "x402" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/Semantic_Web — 'agentic web' is an emerging umbrella term with no single coiner/standards body; the 2001 'Semantic Web' (Berners-Lee) is its conceptual ancestor" }, "related_to": [ "ai-agent", "agent-experience", "mcp", "x402", "ai-crawler" ], "contrast_with": "Unlike the Semantic Web, which made content machine-readable for inference, the agentic web makes the web machine-actionable — agents not only read it but call tools, prove identity and pay.", "example": "By May 2026, Cloudflare Radar reported AI crawlers such as GPTBot (11.48% crawl share) actively traversing the agentic web's content layer.", "source": "https://radar.cloudflare.com/", "status": "emerging", "why_it_matters": "The agentic web is the central entity of this Almanac — the whole readiness, audit and certification path exists to help a site become a first-class participant in it.", "sameAs": [], "bridge_entity": "agentic-web", "last_verified": "2026-06-15", "md_twin": "/glossary/agentic-web.md" }, { "id": "agent-experience", "term": "Agent Experience (AX)", "category": "optimization", "short_def": "The discipline of designing products and websites so that AI agents can use them effectively — the agent-era counterpart to UX.", "long_def": "Coined by Netlify CEO Matt Biilmann in January 2025. AX asks: when an agent (not a human) is the user, can it discover what your service does, understand its options, take action, and have its principal trust the result? Biilmann's framework has four pillars — Access, Context, Tools and Orchestration.", "see_also": [ "agentic-web", "intent-preview", "action-audit" ], "etymology_origin": "Coined by Matt (Mathias) Biilmann, co-founder and CEO of Netlify, in his essay 'Introducing AX: Why Agent Experience Matters', published January 2025.", "related_to": [ "agentic-web", "intent-preview", "action-audit", "accessibility-tree", "geo" ], "contrast_with": "Unlike UX (user experience), which optimizes for a human at the screen, AX (agent experience) optimizes for an AI agent as the primary user class — discovery, machine-readable context and callable tools rather than visual layout.", "example": "Netlify positioned AX as its 'North Star for the next decade', following the essay Biilmann published in January 2025.", "source": "https://biilmann.blog/articles/introducing-ax/", "status": "emerging", "why_it_matters": "AX is the umbrella discipline under which agent-readiness engineering sits; it reframes site quality from 'how it looks to a human' to 'how usable it is to an agent'.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/agent-experience.md" }, { "id": "mcp", "term": "Model Context Protocol (MCP)", "category": "protocols", "short_def": "An open standard from Anthropic that connects AI agents to tools and data through a single JSON-RPC interface — the de facto agent-to-tool standard.", "long_def": "MCP standardizes how an agent discovers and calls tools, reads resources and uses prompts, so any compatible agent can talk to any compatible server without bespoke integration. Introduced by Anthropic in November 2024 and now governed under the Linux Foundation via the AAIF (formed December 2025).", "see_also": [ "webmcp", "tool-use", "ai-agent" ], "etymology_origin": "Introduced and open-sourced by Anthropic on 25 November 2024; governance moved to the Linux Foundation's Agentic AI Infrastructure Foundation (AAIF), formed December 2025.", "related_to": [ "webmcp", "tool-use", "ai-agent", "a2a", "nlweb" ], "contrast_with": "Unlike A2A, which connects agents to other agents, MCP connects a single agent to tools and data — they complement rather than compete, the most common confusion in the vocabulary.", "example": "Within a year of its November 2024 launch, MCP was adopted across the ecosystem and is housed alongside A2A under the AAIF (Linux Foundation, formed December 2025).", "source": "https://en.wikipedia.org/wiki/Model_Context_Protocol", "status": "active", "why_it_matters": "MCP is the dominant way a website exposes callable tools to agents; implementing or speaking MCP is a core agent-readiness capability.", "sameAs": [ "https://en.wikipedia.org/wiki/Model_Context_Protocol", "https://modelcontextprotocol.io/" ], "bridge_entity": "protocols/capability/mcp", "last_verified": "2026-06-15", "md_twin": "/glossary/mcp.md" }, { "id": "webmcp", "term": "WebMCP", "category": "protocols", "short_def": "A W3C-draft standard that lets a web page expose callable tools to a visiting agent via the navigator.modelContext browser API, bringing MCP into the browser.", "long_def": "Instead of an agent screenshotting a page and guessing where to click, a WebMCP-enabled page declares its capabilities as typed tools the agent can invoke directly through navigator.modelContext. Published as a W3C Draft Community Group Report on 10 February 2026 (developed in the Web Machine Learning Community Group) and available as an early preview in Chrome 146.", "see_also": [ "mcp", "accessibility-tree" ], "etymology_origin": "Proposed jointly by Google and Microsoft engineers (unified proposal August 2025); accepted into the W3C Web Machine Learning Community Group September 2025; published as a Draft Community Group Report on 10 February 2026. Note: the browser API is navigator.modelContext (the prior document.modelContext naming is superseded).", "related_to": [ "mcp", "accessibility-tree", "agentic-web" ], "contrast_with": "Unlike MCP, which connects an agent to a remote server, WebMCP exposes tools in-browser via navigator.modelContext — it is the in-browser sibling of MCP, not a server protocol.", "example": "WebMCP shipped as an early preview in Chrome 146 (February 2026), following its W3C Draft Community Group Report dated 10 February 2026.", "source": "https://www.w3.org/community/webml/", "status": "draft", "why_it_matters": "WebMCP lets any website turn its existing UI into agent-callable tools without a separate server, lowering the bar for in-browser agent-readiness.", "sameAs": [], "bridge_entity": "protocols/capability/webmcp", "last_verified": "2026-06-15", "md_twin": "/glossary/webmcp.md" }, { "id": "llms-txt", "term": "llms.txt", "category": "protocols", "short_def": "A markdown file at a domain's root that gives language models a curated index of the site's most important content.", "long_def": "Proposed by Jeremy Howard (Answer.AI) on 3 September 2024. It mirrors robots.txt in spirit but is written for ingestion rather than exclusion: a concise, linkable map that helps agents find and prioritize content. A companion llms-full.txt inlines the full content.", "see_also": [ "markdown-twin", "agentic-web" ], "etymology_origin": "Proposed by Jeremy Howard, co-founder of Answer.AI, on 3 September 2024; specification maintained at llmstxt.org.", "related_to": [ "markdown-twin", "robots-txt", "agentic-web", "content-negotiation" ], "contrast_with": "Unlike robots.txt, which tells crawlers what NOT to fetch (exclusion), llms.txt tells models what to read first (curation for ingestion).", "example": "After Mintlify rolled out llms.txt support across its hosted docs in November 2024, thousands of docs sites — including Anthropic and Cursor — adopted the file.", "source": "https://www.answer.ai/posts/2024-09-03-llmstxt.html", "status": "proposed", "why_it_matters": "llms.txt is one of the lowest-effort, highest-signal agent-readiness steps: a single root file that curates what models ingest from your site.", "sameAs": [ "https://llmstxt.org/" ], "bridge_entity": "protocols/discovery/llms-txt", "last_verified": "2026-06-15", "md_twin": "/glossary/llms-txt.md" }, { "id": "markdown-twin", "term": "Markdown Twin", "category": "protocols", "short_def": "A clean markdown version of an HTML page, served from the same URL via content negotiation when a client requests text/markdown.", "long_def": "Markdown carries the meaning of a page in roughly a tenth of the tokens of the equivalent HTML. Agent fetchers (Claude Code's WebFetch, Cursor, Cloudflare's edge) request it with an Accept: text/markdown header; humans still get the styled HTML.", "see_also": [ "content-negotiation", "llms-txt", "token-economics" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://datatracker.ietf.org/doc/html/rfc7763 — 'markdown twin' is an agentic-web practitioner term with no single coiner/standards body; text/markdown is registered in RFC 7763 (2016)" }, "related_to": [ "content-negotiation", "llms-txt", "token-economics", "rag" ], "contrast_with": "Unlike a separate /page.md URL, a markdown twin lives at the SAME URL as the HTML and is selected by the Accept header via content negotiation — one canonical URL, two representations.", "example": "Claude Code's WebFetch and Cursor request the markdown representation with an Accept: text/markdown header, receiving roughly a tenth of the tokens of the HTML.", "source": "https://datatracker.ietf.org/doc/html/rfc7763", "status": "emerging", "why_it_matters": "The markdown twin is the canonical way to serve agents a cheap, clean, low-token version of a page without maintaining a duplicate site.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/markdown-twin.md" }, { "id": "content-negotiation", "term": "Content Negotiation", "category": "protocols", "short_def": "An HTTP mechanism where the same URL returns different representations based on request headers such as Accept.", "long_def": "Standardized in HTTP/1.1 (RFC 9110, HTTP Semantics), it is the clean way to serve HTML to browsers and markdown to agents from one URL, advertised with a Vary: Accept response header so caches behave.", "see_also": [ "markdown-twin" ], "etymology_origin": "Specified by the IETF as part of HTTP semantics; current normative reference is RFC 9110 'HTTP Semantics' (June 2022), which obsoletes the earlier RFC 7231 / RFC 2616 definitions.", "related_to": [ "markdown-twin", "robots-txt" ], "contrast_with": "Unlike URL-based routing, which serves different content from different paths, content negotiation serves different representations from ONE URL based on request headers (Accept, Accept-Language).", "example": "A server sets Vary: Accept and returns text/markdown to an agent and text/html to a browser from the same URL, as defined in RFC 9110 (June 2022).", "source": "https://www.rfc-editor.org/rfc/rfc9110.html#name-content-negotiation", "status": "active", "why_it_matters": "Content negotiation is the standards-compliant mechanism that makes markdown twins possible without breaking caches, SEO or canonical URLs.", "sameAs": [ "https://en.wikipedia.org/wiki/Content_negotiation" ], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/content-negotiation.md" }, { "id": "tool-use", "term": "Tool Use", "category": "core", "short_def": "An LLM's ability to call external functions — search, code execution, APIs — by emitting a structured request the host executes.", "long_def": "The model does not run the tool itself; it outputs a tool call, the host runs it and returns the result, and the model continues. Tool use (also called function calling) is what turns a chat model into an agent.", "see_also": [ "ai-agent", "mcp", "agentic-loop" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://platform.openai.com/docs/guides/function-calling — popularised as 'function calling' by OpenAI (June 2023); no single coiner for the broader 'tool use' sense" }, "related_to": [ "ai-agent", "mcp", "agentic-loop" ], "contrast_with": "Unlike plain text generation, tool use emits a structured, machine-parseable call that the host actually executes — the model proposes the action; the host performs it.", "example": "OpenAI shipped function calling for tool use in its API in June 2023; MCP later standardized how those tools are discovered and described across vendors.", "source": "https://platform.openai.com/docs/guides/function-calling", "status": "active", "why_it_matters": "Tool use is the mechanism by which an agent acts on your site; exposing clean, well-described tools (e.g. via MCP) is what makes a site actionable to agents.", "sameAs": [], "bridge_entity": "protocols/capability/mcp", "last_verified": "2026-06-15", "md_twin": "/glossary/tool-use.md" }, { "id": "agentic-loop", "term": "Agentic Loop", "category": "core", "short_def": "The observe-decide-act-observe cycle an agent repeats until its task is complete.", "long_def": "Each turn the agent reads the current state, decides on the next action (often a tool call), takes it, and incorporates the result. The loop ends when the goal is reached, a budget is exhausted, or the agent asks for input.", "see_also": [ "ai-agent", "tool-use" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/OODA_loop — the agentic loop generalises the perception-action cycle (and Boyd's OODA loop, 1976); no single coiner for the LLM-agent sense" }, "related_to": [ "ai-agent", "tool-use", "agent-experience" ], "contrast_with": "Unlike a single prompt-and-response, the agentic loop repeats — each tool result feeds the next decision — until a goal, budget or human checkpoint stops it.", "example": "An agent debugging code runs an agentic loop: read error, edit file, run tests, read result, repeat — iterating until the tests pass.", "source": "https://en.wikipedia.org/wiki/OODA_loop", "status": "active", "why_it_matters": "The agentic loop explains why agents fetch a page multiple times and why clear, stateful, machine-readable responses (not one-shot HTML) make a site agent-friendly.", "sameAs": [], "bridge_entity": "agentic-web", "last_verified": "2026-06-15", "md_twin": "/glossary/agentic-loop.md" }, { "id": "rag", "term": "Retrieval-Augmented Generation (RAG)", "category": "knowledge-memory", "short_def": "Fetching relevant documents at query time and giving them to the model as context, so answers are grounded in current, specific data.", "long_def": "The term was coined by Patrick Lewis and colleagues at Facebook AI Research (now Meta AI), University College London and NYU in a 2020 NeurIPS paper. RAG reduces hallucination and lets a model answer about information it was never trained on. Agent-friendly sites help RAG by exposing clean, chunkable content (markdown twins, llms.txt) and structured data.", "see_also": [ "grounding", "markdown-twin" ], "etymology_origin": "Coined by Patrick Lewis et al. in 'Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks' (Facebook AI Research / UCL / NYU), arXiv 2005.11401, May 2020; presented at NeurIPS 2020.", "related_to": [ "grounding", "markdown-twin", "json-ld" ], "contrast_with": "Unlike fine-tuning, which bakes knowledge into model weights, RAG fetches documents at query time and supplies them as context — knowledge stays external, current and citable.", "example": "The seminal 2020 RAG paper (Lewis et al., arXiv 2005.11401) combined dense-passage retrieval with a BART generator and set state-of-the-art on three open-domain QA benchmarks.", "source": "https://arxiv.org/abs/2005.11401", "status": "active", "why_it_matters": "RAG is how most AI answer engines ground responses; sites that expose clean, retrievable content get pulled into and cited by those answers.", "sameAs": [ "https://en.wikipedia.org/wiki/Retrieval-augmented_generation" ], "bridge_entity": "models", "last_verified": "2026-06-15", "md_twin": "/glossary/rag.md" }, { "id": "grounding", "term": "Grounding", "category": "knowledge-memory", "short_def": "Tying a model's output to verifiable external sources rather than its parametric memory.", "long_def": "A grounded answer can cite where each claim came from. Structured data and retrievable content make grounding easier; AI answer engines increasingly cross-check claims against the live page.", "see_also": [ "rag", "json-ld", "geo" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/Symbol_grounding_problem — 'grounding' derives from the symbol-grounding problem (Harnad, 1990); the LLM 'grounding-to-sources' sense has no single coiner" }, "related_to": [ "rag", "json-ld", "geo" ], "contrast_with": "Unlike RAG, which is the retrieval mechanism, grounding is the property of the output — an answer is grounded when each claim is tied to a verifiable source, however it was retrieved.", "example": "AI answer engines such as Perplexity ground responses by citing the live pages they pulled from, letting a reader trace each claim to its source.", "source": "https://en.wikipedia.org/wiki/Symbol_grounding_problem", "status": "active", "why_it_matters": "Grounding is why source-rich, structured, accurate content gets cited; a site that is easy to ground is a site that AI engines quote.", "sameAs": [ "https://en.wikipedia.org/wiki/Symbol_grounding_problem" ], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/grounding.md" }, { "id": "json-ld", "term": "JSON-LD", "category": "protocols", "short_def": "A JSON-based format for embedding Schema.org structured data in a page — the lingua franca that AI engines extract.", "long_def": "JSON-LD 1.1 is a W3C Recommendation (2020). A single script tag with an @graph can describe a site, its pages, its author and its key entities in a way that Google, Bing, Perplexity and ChatGPT all parse. By 2026, engines cross-check schema claims against page content, so accuracy matters more than volume.", "see_also": [ "grounding", "geo", "agentic-web" ], "etymology_origin": "Developed at the W3C; JSON-LD 1.0 reached W3C Recommendation in 2014 and JSON-LD 1.1 in 2020. Manu Sporny (Digital Bazaar) was the original 1.0 editor/author; 1.1 editors include Gregg Kellogg, Pierre-Antoine Champin and Dave Longley.", "related_to": [ "grounding", "geo", "agentic-web", "nlweb" ], "contrast_with": "Unlike microdata or RDFa, which inline structured data into HTML tags, JSON-LD lives in a self-contained script block — decoupled from markup and far easier for engines to extract.", "example": "JSON-LD 1.1 became a W3C Recommendation in 2020; a single script tag with @graph lets Google, Bing, Perplexity and ChatGPT parse a page's entities.", "source": "https://www.w3.org/TR/json-ld11/", "status": "active", "why_it_matters": "JSON-LD is the structured-data format AI engines extract first; correct, page-consistent JSON-LD is a primary lever for getting entities understood and cited.", "sameAs": [ "https://en.wikipedia.org/wiki/JSON-LD", "https://www.w3.org/TR/json-ld11/" ], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/json-ld.md" }, { "id": "geo", "term": "Generative Engine Optimization (GEO)", "category": "optimization", "short_def": "Optimizing content to be cited and surfaced by AI answer engines, the way SEO optimized for search rankings.", "long_def": "The term was coined in the 2023 research paper 'GEO: Generative Engine Optimization' by Pranjal Aggarwal, Vishvak Murahari et al. (Princeton University and collaborators), later published at KDD 2024. Because AI engines summarize and cite rather than list ten blue links, GEO favors clear, structured, quotable, well-sourced content. Often discussed alongside AEO (Answer Engine Optimization).", "see_also": [ "json-ld", "grounding", "agentic-web" ], "etymology_origin": "Coined by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande in 'GEO: Generative Engine Optimization', arXiv 2311.09735 (November 2023); published at ACM SIGKDD (KDD 2024), August 2024.", "related_to": [ "json-ld", "grounding", "agent-experience", "agentic-web" ], "contrast_with": "Unlike SEO, which optimizes for ranking position in a list of links, GEO optimizes for inclusion and citation inside a generated answer — the unit of success is a citation, not a rank.", "example": "The seminal GEO paper (arXiv 2311.09735, November 2023) introduced the term and was presented at KDD 2024.", "source": "https://arxiv.org/abs/2311.09735", "status": "active", "why_it_matters": "GEO is the optimization discipline a content owner uses to get cited by AI answer engines; it bridges traditional SEO audiences into agent-readiness.", "sameAs": [ "https://en.wikipedia.org/wiki/Generative_engine_optimization" ], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/geo.md" }, { "id": "prompt-injection", "term": "Prompt Injection", "category": "identity", "short_def": "An attack that hides instructions in content an agent reads, hijacking its behavior against its principal's intent.", "long_def": "The term was coined by Simon Willison in September 2022, framed as the LLM analogue of SQL injection. Because agents act on the text they ingest, malicious or invisible instructions on a page ('ignore previous instructions and...') can manipulate them. Hidden agent-only text is therefore an anti-pattern indistinguishable from an attack; trustworthy sites keep their machine layer transparent.", "see_also": [ "agent-experience", "web-bot-auth" ], "etymology_origin": "Coined and defined by Simon Willison on 12 September 2022 ('Prompt injection attacks against GPT-3'), naming it after SQL injection; the underlying GPT-3 vulnerability was demonstrated by Riley Goodside the prior day.", "related_to": [ "agent-experience", "web-bot-auth", "agent-identity" ], "contrast_with": "Unlike jailbreaking, where a user coaxes a model to break its own rules, prompt injection plants instructions in third-party content the agent later reads — the attacker is not the user.", "example": "Simon Willison coined 'prompt injection' on 12 September 2022, framing it as the LLM analogue of SQL injection.", "source": "https://simonwillison.net/2022/Sep/12/prompt-injection/", "status": "active", "why_it_matters": "Prompt injection is the reason hidden agent-only text is an anti-pattern; an agent-ready site keeps its machine layer transparent and signed, not cloaked.", "sameAs": [ "https://en.wikipedia.org/wiki/Prompt_injection" ], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/prompt-injection.md" }, { "id": "web-bot-auth", "term": "Web Bot Auth", "category": "identity", "short_def": "Cryptographically verifying which agent is making a request using HTTP Message Signatures (RFC 9421), since user-agent strings are spoofable.", "long_def": "An agent signs its requests with an Ed25519 key tied to a published identity (a JWKS directory at /.well-known/http-message-signatures-directory, advertised via the Signature-Agent header); the server verifies the signature per RFC 9421. This lets sites distinguish a genuine ClaudeBot or GPTBot from an impostor, and is the foundation for agent-aware rate limits and paid access.", "see_also": [ "agent-identity", "prompt-injection", "x402" ], "etymology_origin": "An IETF effort building on RFC 9421 'HTTP Message Signatures' (February 2024); the Web Bot Auth scheme and the HTTP Message Signatures Directory are active IETF Internet-Drafts, with Cloudflare publishing the reference write-up.", "related_to": [ "agent-identity", "prompt-injection", "x402", "ai-crawler", "robots-txt" ], "contrast_with": "Unlike user-agent strings or IP-range checks, which are spoofable or brittle, Web Bot Auth proves identity cryptographically with an Ed25519 signature over the request (RFC 9421).", "example": "OpenAI signs all Operator requests with HTTP Message Signatures so site owners can cryptographically verify they genuinely originate from Operator, per Cloudflare's Web Bot Auth write-up.", "source": "https://blog.cloudflare.com/web-bot-auth/", "status": "emerging", "why_it_matters": "Web Bot Auth is the foundation for trusting an agent's identity — the precondition for agent-aware rate limits, pay-per-crawl and verified-agent certification.", "sameAs": [], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/web-bot-auth.md" }, { "id": "agent-identity", "term": "Agent Identity", "category": "identity", "short_def": "A verifiable answer to 'which agent, acting for whom, is making this request?'", "long_def": "Built from signed requests (Web Bot Auth / RFC 9421), declared user-agents and operator-published verification (IP ranges, reverse DNS). Strong agent identity is the precondition for agent-native access control and commerce.", "see_also": [ "web-bot-auth", "agentic-commerce" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://datatracker.ietf.org/doc/html/rfc9421 — 'agent identity' is a composite agentic-web concept (Web Bot Auth + reverse-DNS + Agent Cards) with no single coining spec" }, "related_to": [ "web-bot-auth", "agentic-commerce", "a2a", "ai-crawler" ], "contrast_with": "Unlike Web Bot Auth, which is the cryptographic mechanism, agent identity is the broader answer it serves — who the agent is AND on whose behalf it acts (the principal), spanning signatures, reverse DNS and Agent Cards.", "example": "An agent's identity is verified by combining a Web Bot Auth Ed25519 signature (RFC 9421) with operator-published IP ranges and reverse DNS, since user-agent strings alone are spoofable.", "source": "https://datatracker.ietf.org/doc/html/rfc9421", "status": "emerging", "why_it_matters": "Agent identity is the gate before access control and commerce: a site cannot safely grant agents paid or privileged access without first knowing which agent — and which principal — it is dealing with.", "sameAs": [], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/agent-identity.md" }, { "id": "x402", "term": "x402", "category": "commerce", "short_def": "A protocol that uses the HTTP 402 'Payment Required' status so an agent can pay for a resource inline with a stablecoin micropayment.", "long_def": "Launched by Coinbase in May 2025. The server answers an unpaid request with 402 and machine-readable payment terms (amount, asset, network, recipient); the agent pays in a stablecoin such as USDC and retries with cryptographic proof of payment. On 23 September 2025 Coinbase and Cloudflare announced the x402 Foundation to steward the standard.", "see_also": [ "agentic-commerce", "agent-identity" ], "etymology_origin": "Launched by Coinbase in May 2025; named after the long-dormant HTTP 402 'Payment Required' status code. Stewardship moved to the x402 Foundation (Coinbase + Cloudflare), announced 23 September 2025.", "related_to": [ "agentic-commerce", "agent-identity" ], "contrast_with": "Unlike AP2, which is the authorization layer (cryptographic payment mandates), x402 is the settlement layer that actually moves the stablecoin — AP2 extends x402 rather than replacing it.", "example": "As of March 2026, x402 had processed over 119 million transactions on Base and 35 million on Solana, with roughly $600 million in annualized volume and zero protocol fees.", "source": "https://www.coinbase.com/developer-platform/discover/launches/x402", "status": "active", "why_it_matters": "x402 is the leading HTTP-native rail for an agent to pay for content or tools inline, making pay-per-call and pay-per-crawl monetization possible without legacy billing.", "sameAs": [ "https://en.wikipedia.org/wiki/X402" ], "bridge_entity": "protocols/payments/x402", "last_verified": "2026-06-15", "md_twin": "/glossary/x402.md" }, { "id": "agentic-commerce", "term": "Agentic Commerce", "category": "commerce", "short_def": "Transactions initiated and completed by AI agents on a user's behalf, through protocols like x402, AP2 and the Agentic Commerce Protocol.", "long_def": "Spans micropayments for data and tools (x402), mandate-based settlement across rails (AP2) and in-conversation checkout (ACP / Instant Checkout). The shared challenge is authorization: proving the agent had the user's permission to spend.", "see_also": [ "x402", "agent-identity" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://www.coinbase.com/developer-platform/discover/launches/x402 — 'agentic commerce' is an umbrella category term popularised across 2025 launches (x402, AP2, ACP) with no single coiner" }, "related_to": [ "x402", "agent-identity", "a2a" ], "contrast_with": "Unlike e-commerce, where a human clicks checkout, agentic commerce has the AI agent transact on the user's behalf — shifting the hard problem from UX to verifiable authorization (did the agent have permission to spend?).", "example": "OpenAI and Stripe launched the Agentic Commerce Protocol with ChatGPT Instant Checkout on 29 September 2025, going live with Etsy on launch day.", "source": "https://openai.com/index/buy-it-in-chatgpt/", "status": "emerging", "why_it_matters": "Agentic commerce is the revenue side of the agentic web; a site that wants agents to buy from it must speak one of its authorization-plus-settlement protocols.", "sameAs": [], "bridge_entity": "protocols/payments/x402", "last_verified": "2026-06-15", "md_twin": "/glossary/agentic-commerce.md" }, { "id": "accessibility-tree", "term": "Accessibility Tree", "category": "knowledge-memory", "short_def": "The semantic representation of a page that assistive tech — and browser-driving agents — read instead of pixels.", "long_def": "Derived from semantic HTML and ARIA and specified by the W3C Core Accessibility API Mappings, it exposes roles, labels and structure. Agents that drive a browser act through this tree, which is why accessible markup doubles as an agent interface: one investment, two audiences.", "see_also": [ "webmcp", "agent-experience" ], "etymology_origin": "A construct of browser/user-agent accessibility APIs, normatively specified by the W3C in the Core Accessibility API Mappings (Core-AAM) and WAI-ARIA, developed by the W3C ARIA Working Group.", "related_to": [ "webmcp", "agent-experience", "agentic-loop" ], "contrast_with": "Unlike the DOM, which describes a page's full markup and presentation, the accessibility tree exposes only its semantics — roles, labels and states — which is what a browser-driving agent (and a screen reader) actually consumes.", "example": "A browser-driving agent like Operator acts through the accessibility tree's roles and labels (per the W3C Core Accessibility API Mappings) rather than parsing pixels.", "source": "https://www.w3.org/TR/core-aam-1.1/", "status": "active", "why_it_matters": "The accessibility tree means accessible markup IS an agent interface — one investment in semantic HTML/ARIA serves both assistive tech and browser-driving agents.", "sameAs": [ "https://developer.mozilla.org/en-US/docs/Glossary/Accessibility_tree" ], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/accessibility-tree.md" }, { "id": "intent-preview", "term": "Intent Preview", "category": "optimization", "short_def": "Showing what an agent action will do before it does it, so a human can approve or cancel.", "long_def": "An AX trust pattern: rather than acting silently, the agent surfaces the planned call and its effect ('this will POST and write one record') for confirmation. Pairs with an action audit log of what actually happened.", "see_also": [ "action-audit", "agent-experience" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://biilmann.blog/articles/introducing-ax/ — 'intent preview' is an Agent Experience (AX) trust pattern; confirm whether Biilmann/Netlify or this Almanac is the first named source for the exact term" }, "related_to": [ "action-audit", "agent-experience" ], "contrast_with": "Unlike an action audit log, which records what already happened (the 'what did' half), intent preview shows what is about to happen (the 'what will' half) so a human can approve or cancel first.", "example": "Before an agent writes data, an intent preview surfaces the planned call and its effect — 'this will POST and write one record' — for human confirmation.", "source": "https://biilmann.blog/articles/introducing-ax/", "status": "emerging", "why_it_matters": "Intent preview is a core AX trust pattern: it keeps a human in the loop for consequential agent actions without forcing the agent to stop for trivial ones.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/intent-preview.md" }, { "id": "action-audit", "term": "Action Audit Log", "category": "optimization", "short_def": "A visible, timestamped record of the actions an agent has taken.", "long_def": "The 'what did happen' half of agent trust (intent preview is the 'what will happen' half). An auditable trail lets a principal verify and, if needed, undo what an agent did.", "see_also": [ "intent-preview", "agent-experience" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://biilmann.blog/articles/introducing-ax/ — 'action audit log' is an Agent Experience (AX) trust pattern; audit logging itself is a general security concept with no single coiner for the agent sense" }, "related_to": [ "intent-preview", "agent-experience" ], "contrast_with": "Unlike intent preview, which asks for approval before an action, an action audit log is the after-the-fact record — the 'what did happen' half — that lets a principal verify or undo an agent's work.", "example": "After an agent completes a task, its action audit log shows a timestamped trail of every call it made, letting the principal verify and, if needed, undo each one.", "source": "https://biilmann.blog/articles/introducing-ax/", "status": "emerging", "why_it_matters": "An action audit log is the accountability half of agent trust; without it, a principal cannot verify or reverse what an autonomous agent did on their behalf.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/action-audit.md" }, { "id": "token-economics", "term": "Token Economics", "category": "knowledge-memory", "short_def": "The cost structure of agent interactions, where every token of input and output is billed — making concise, structured content a direct cost saving.", "long_def": "Because agents pay per token, a markdown twin that is ~90% smaller than its HTML equivalent is not just faster but cheaper to consume. Agent-friendly design is partly an economic argument.", "see_also": [ "markdown-twin", "content-negotiation" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/Large_language_model — 'token economics' here means LLM token billing/cost-structure, distinct from blockchain 'tokenomics'; no single coiner for the LLM sense" }, "related_to": [ "markdown-twin", "content-negotiation", "rag" ], "contrast_with": "Unlike blockchain 'tokenomics' (the supply and incentive design of a crypto token), token economics here means the per-token billing of LLM input and output — a content-cost argument, not a crypto one.", "example": "A markdown twin is roughly 90% smaller than its HTML equivalent in tokens, so serving it directly lowers the per-token cost an agent pays to read the page.", "source": "https://en.wikipedia.org/wiki/Large_language_model", "status": "active", "why_it_matters": "Token economics turns agent-readiness into a cost argument: leaner, structured content is literally cheaper for an agent to consume, which influences whether agents prefer your site.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/token-economics.md" }, { "id": "a2a", "term": "Agent2Agent (A2A)", "category": "protocols", "short_def": "A protocol for agents to discover and delegate tasks to each other, each publishing an Agent Card of its skills.", "long_def": "Where MCP connects an agent to tools, A2A connects agents to other agents. Announced by Google on 9 April 2025 and donated to the Linux Foundation in June 2025 for neutral governance. It defines Agent Cards (capability advertisements), Tasks (the work exchanged) and a transport over HTTP/SSE/JSON-RPC 2.0, with payment extensions (x402, AP2) layered on the task lifecycle.", "see_also": [ "mcp", "ap2", "agentic-commerce" ], "etymology_origin": "Announced by Google on 9 April 2025 at Google Cloud Next; released under Apache 2.0 and donated to the Linux Foundation in June 2025, which launched the Agent2Agent Protocol Project.", "related_to": [ "mcp", "agentic-commerce", "agent-identity" ], "contrast_with": "Unlike MCP, which connects an agent to tools and data, A2A connects agents to OTHER agents — discovery, task delegation and coordination between peers; the two complement each other.", "example": "Google announced A2A on 9 April 2025; by April 2026 more than 150 organizations — including Microsoft, AWS, Salesforce, SAP and IBM — supported it.", "source": "https://en.wikipedia.org/wiki/Agent2Agent", "status": "active", "why_it_matters": "A2A is the interoperability layer for multi-agent systems; a site or service that publishes an Agent Card can be discovered and delegated to by other agents.", "sameAs": [ "https://en.wikipedia.org/wiki/Agent2Agent" ], "bridge_entity": "protocols/interop/a2a", "last_verified": "2026-06-15", "md_twin": "/glossary/a2a.md" }, { "id": "nlweb", "term": "NLWeb", "category": "protocols", "short_def": "Microsoft's standard for turning a website into a conversational endpoint that answers natural-language queries — and is itself an MCP server.", "long_def": "Introduced by Microsoft at Build 2025 (May 2025) and led by Schema.org/RSS/RDF creator R.V. Guha, NLWeb combines a site's Schema.org data, a vector index and an LLM to answer questions grounded in the site's own content, exposing the result over a simple endpoint that doubles as an MCP server.", "see_also": [ "mcp", "json-ld", "agentic-web" ], "etymology_origin": "Introduced by Microsoft at Build 2025 (May 2025); conceived and led by Ramanathan V. (R.V.) Guha, creator of RSS, RDF and Schema.org, who joined Microsoft as CVP and Technical Fellow.", "related_to": [ "mcp", "json-ld", "agentic-web", "rag" ], "contrast_with": "Unlike MCP, which is the tool-calling protocol, NLWeb is a way to make an existing website conversationally queryable — and every NLWeb endpoint is itself exposed as an MCP server, so it builds on MCP rather than competing with it.", "example": "Microsoft introduced NLWeb at Build 2025 (May 2025) with early adopters including Shopify, Snowflake, O'Reilly Media, Tripadvisor and Eventbrite.", "source": "https://en.wikipedia.org/wiki/NLWeb", "status": "emerging", "why_it_matters": "NLWeb lets a site reuse its existing Schema.org data to become an agent-queryable MCP endpoint — a low-lift path from structured data to conversational agent-readiness.", "sameAs": [ "https://en.wikipedia.org/wiki/NLWeb" ], "bridge_entity": "protocols/interop/nlweb", "last_verified": "2026-06-15", "md_twin": "/glossary/nlweb.md" }, { "id": "robots-txt", "term": "robots.txt", "category": "protocols", "short_def": "The root-level file that tells crawlers — including AI crawlers — what they may and may not fetch.", "long_def": "The web's oldest crawler contract, originally defined by Martijn Koster in 1994 and standardized as RFC 9309 in 2022. In the agentic era it is where sites name AI crawlers explicitly (GPTBot, ClaudeBot, Google-Extended), and where RSL licensing terms are referenced via a License directive.", "see_also": [ "ai-crawler", "rsl", "llms-txt" ], "etymology_origin": "Originally defined by Martijn Koster in 1994 as the Robots Exclusion Protocol; a de facto standard by mid-1994, formally published by the IETF as RFC 9309 (with Koster as an author) in September 2022.", "related_to": [ "ai-crawler", "llms-txt", "content-negotiation" ], "contrast_with": "Unlike llms.txt, which curates what models should read first (inclusion), robots.txt declares what crawlers may NOT fetch (exclusion) — exclusion contract versus ingestion index.", "example": "RFC 9309 (September 2022) formalized the Robots Exclusion Protocol that Martijn Koster first defined in 1994; sites now name AI crawlers such as GPTBot and Google-Extended in it explicitly.", "source": "https://www.rfc-editor.org/rfc/rfc9309.html", "status": "active", "why_it_matters": "robots.txt is still the front door for crawler access control; in the agentic era it is where a site first decides which AI crawlers it admits, blocks or licenses.", "sameAs": [ "https://en.wikipedia.org/wiki/Robots.txt", "https://www.rfc-editor.org/rfc/rfc9309.html" ], "bridge_entity": "crawlers", "last_verified": "2026-06-15", "md_twin": "/glossary/robots-txt.md" }, { "id": "ai-crawler", "term": "AI Crawler", "category": "core", "short_def": "An automated bot that fetches web content for an AI system — to train a model, build a search index, or answer a user's question in real time.", "long_def": "AI crawlers split by purpose (training vs search vs inference) and by behavior (whether they honor robots.txt). Their user-agent strings are spoofable, so genuine ones are confirmed via published IP ranges or reverse DNS — and increasingly via Web Bot Auth signatures.", "see_also": [ "robots-txt", "web-bot-auth", "agent-identity" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://radar.cloudflare.com/ — 'AI crawler' is a descriptive category (training/search/inference bots) tracked by Cloudflare Radar and ai.robots.txt; no single coining authority" }, "related_to": [ "robots-txt", "web-bot-auth", "agent-identity", "agentic-web" ], "contrast_with": "Unlike a traditional search crawler such as classic Googlebot, an AI crawler fetches content to train models or to ground a live answer — and a growing share (inference fetchers) act per user query rather than on a scheduled index crawl.", "example": "Per Cloudflare Radar (May 2026), AI crawlers by crawl share included GPTBot (11.48%), Bytespider (10.25%), Applebot (7.01%) and the new Claude-SearchBot (2.22%).", "source": "https://radar.cloudflare.com/", "status": "active", "why_it_matters": "AI crawlers are the agents most sites encounter first; knowing which one is which — and verifying it — is the entry point to every access, licensing and citation decision.", "sameAs": [ "https://en.wikipedia.org/wiki/Web_crawler" ], "bridge_entity": "crawlers", "last_verified": "2026-06-15", "md_twin": "/glossary/ai-crawler.md" }, { "id": "ap2", "term": "Agent Payments Protocol (AP2)", "category": "commerce", "short_def": "An open protocol from Google that gives an AI agent a cryptographically signed mandate proving a human authorized it to spend, before any payment is made.", "long_def": "Announced by Google on 16 September 2025 with more than sixty payments and technology partners (including Mastercard, PayPal, American Express, Coinbase and Adyen), AP2 introduces tamper-proof 'mandates' — signed digital contracts that prove a user authorized a specific transaction. It is payment-rail agnostic (cards, bank transfers, stablecoins) and is designed to layer on top of A2A and settlement protocols like x402.", "see_also": [ "x402", "agentic-commerce", "a2a" ], "etymology_origin": "Announced by Google on 16 September 2025 as the Agent Payments Protocol (AP2), developed with 60+ payments and technology partners; an open, payment-rail-agnostic extension to the Agent2Agent (A2A) ecosystem.", "related_to": [ "x402", "agentic-commerce", "a2a", "agent-identity", "acp" ], "contrast_with": "Unlike x402, which is the settlement layer that actually moves a stablecoin, AP2 is the authorization layer — a cryptographic mandate proving the human consented; AP2 extends x402 rather than replacing it.", "example": "Google announced AP2 on 16 September 2025 with more than sixty partners including Mastercard, PayPal, American Express and Coinbase.", "source": "https://cloud.google.com/blog/products/ai-machine-learning/announcing-agents-to-payments-ap2-protocol", "status": "emerging", "why_it_matters": "AP2 is the authorization standard that lets a site accept agent-initiated payments with proof the human consented — the trust precondition for agentic checkout at scale.", "sameAs": [], "bridge_entity": "protocols/payments/ap2", "last_verified": "2026-06-15", "md_twin": "/glossary/ap2.md" }, { "id": "acp", "term": "Agentic Commerce Protocol (ACP)", "category": "commerce", "short_def": "An open standard, maintained by OpenAI and Stripe, that connects buyers, their AI agents and merchants so a purchase can complete inside a conversation.", "long_def": "ACP defines an interaction model for in-conversation checkout — the standard behind ChatGPT's Instant Checkout. Maintained jointly by OpenAI and Stripe as Founding Maintainers under the Apache 2.0 license, it uses date-based (YYYY-MM-DD) versioning, with OpenAI and Stripe providing the first reference implementations.", "see_also": [ "agentic-commerce", "x402", "ap2" ], "etymology_origin": "Introduced alongside ChatGPT Instant Checkout (29 September 2025); the specification is maintained by OpenAI and Stripe as Founding Maintainers under Apache 2.0, with date-based (YYYY-MM-DD) versioning.", "related_to": [ "agentic-commerce", "x402", "ap2", "agent-as-buyer" ], "contrast_with": "Unlike AP2, a payment-authorization protocol, ACP is the broader checkout interaction model between an agent and a merchant — it governs how the purchase happens in-conversation, and can settle through underlying payment rails.", "example": "OpenAI and Stripe launched the Agentic Commerce Protocol with ChatGPT Instant Checkout on 29 September 2025, going live with Etsy on launch day.", "source": "https://github.com/agentic-commerce-protocol/agentic-commerce-protocol", "status": "emerging", "why_it_matters": "ACP is how a merchant exposes its catalog and checkout to ChatGPT and other agents directly; speaking it is the path to selling inside an AI conversation.", "sameAs": [], "bridge_entity": "protocols/payments/acp", "last_verified": "2026-06-15", "md_twin": "/glossary/acp.md" }, { "id": "ucp", "term": "Universal Commerce Protocol (UCP)", "category": "commerce", "short_def": "An open commerce standard from Google that orchestrates A2A, AP2 and payment rails into one end-to-end agentic-commerce journey.", "long_def": "UCP defines a common language and functional primitives so consumer surfaces, businesses and payment providers can transact through agents. Developed by Google with retail partners including Shopify, Etsy, Wayfair, Target and Walmart, it composes existing protocols — A2A for agent communication and AP2 for payment mandates — rather than replacing them.", "see_also": [ "agentic-commerce", "ap2", "a2a" ], "etymology_origin": "Introduced by Google as the Universal Commerce Protocol, developed with retail partners (Shopify, Etsy, Wayfair, Target, Walmart); composes A2A and AP2.", "related_to": [ "agentic-commerce", "ap2", "a2a", "acp" ], "contrast_with": "Unlike AP2 or x402, which each handle one slice (authorization, settlement), UCP is the orchestration layer that strings discovery, agent messaging and payment together into a single commerce journey.", "example": "Google's Universal Commerce Protocol orchestrates A2A and AP2, with launch partners including Shopify, Etsy, Wayfair, Target and Walmart.", "source": "https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/", "status": "emerging", "why_it_matters": "UCP is the highest-level commerce framework a retailer can adopt to be transactable by agents end-to-end; it bundles the identity, messaging and payment standards into one journey.", "sameAs": [], "bridge_entity": "protocols/payments/ucp", "last_verified": "2026-06-15", "md_twin": "/glossary/ucp.md" }, { "id": "mpp", "term": "Machine Payments Protocol (MPP)", "category": "commerce", "short_def": "An open, HTTP-native standard co-authored by Stripe and Tempo that lets an agent request, authorize and settle a payment within the same HTTP request.", "long_def": "MPP is an internet-native machine-to-machine payment standard, proposed to the IETF, that lets agents pay for services inline across stablecoins, cards and other methods using Shared Payment Tokens (SPTs). Businesses configure spend limits, merchant-category restrictions and approval workflows in advance, so agents transact only within explicitly granted permissions.", "see_also": [ "x402", "agentic-commerce", "ap2" ], "etymology_origin": "Co-authored by Stripe and Tempo and announced on 18 March 2026; an open machine-to-machine payment standard built on HTTP and Shared Payment Tokens (SPTs).", "related_to": [ "x402", "agentic-commerce", "ap2", "ucp" ], "contrast_with": "Unlike x402, which settles natively in stablecoin over the HTTP 402 status, MPP is method-agnostic — it coordinates stablecoins, cards and BNPL through Shared Payment Tokens within the same HTTP request.", "example": "Stripe and Tempo announced the Machine Payments Protocol on 18 March 2026; named launch adopters include Browserbase, PostalForm and Prospect Butcher Co.", "source": "https://stripe.com/blog/machine-payments-protocol", "status": "emerging", "why_it_matters": "MPP gives a business one inline, permissioned rail to accept agent payments across stablecoins and cards, with budgets and merchant rules enforced before the agent ever transacts.", "sameAs": [], "bridge_entity": "protocols/payments/mpp", "last_verified": "2026-06-15", "md_twin": "/glossary/mpp.md" }, { "id": "agents-md", "term": "AGENTS.md", "category": "protocols", "short_def": "An open, plain-Markdown file placed in a repository that gives coding agents project-specific build, test, style and security instructions.", "long_def": "Conceived as a 'README for agents', AGENTS.md is plain CommonMark with no required schema; agents scan for conventional headings like '## Build & Test' or '## Code Style'. Agents read the nearest file walking up the directory tree, so monorepo subprojects can ship tailored instructions. Originated in the OpenAI ecosystem and now stewarded by the Agentic AI Foundation (AAIF) under the Linux Foundation; adopted by Codex, Cursor, Zed, Jules, Aider and others.", "see_also": [ "agents-txt", "llms-txt", "mcp" ], "etymology_origin": "Open Markdown format published in the OpenAI ecosystem (repository at github.com/openai/agents.md); plain CommonMark with no required schema, now stewarded by the Linux Foundation's Agentic AI Foundation (AAIF).", "related_to": [ "agents-txt", "llms-txt", "robots-txt", "mcp" ], "contrast_with": "Unlike agents.txt, which declares site-level access rules and capabilities for visiting agents, AGENTS.md instructs a coding agent how to build, test and style a specific codebase — repository guidance, not access policy.", "example": "AGENTS.md is supported by OpenAI Codex, Cursor, Zed, Google Jules and Aider, with agents reading the nearest file up the directory tree in a monorepo.", "source": "https://agents.md/", "status": "active", "why_it_matters": "AGENTS.md is the lowest-effort way to make a repository agent-ready for coding agents — a single Markdown file that every major coding agent now reads.", "sameAs": [], "bridge_entity": "protocols/discovery/agents-md", "last_verified": "2026-06-15", "md_twin": "/glossary/agents-md.md" }, { "id": "agents-txt", "term": "agents.txt", "category": "protocols", "short_def": "A proposed root-level file that declares a site's identity, terms of use, service catalog and agentic endpoints to visiting AI agents.", "long_def": "Where robots.txt answers 'may you look at this?', agents.txt aims to answer 'what can you do here, and on what terms?' — a B2A (business-to-agent) capability manifest. The namespace is contested: several independent projects have used the agents.txt name for different purposes, and one prominent proposal was renamed agent-manifest.txt in March 2026 to disambiguate. It is an emerging, not-yet-standardized convention.", "see_also": [ "agents-md", "robots-txt", "llms-txt" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://github.com/asturwebs/agents-txt — 'agents.txt' is a contested namespace with multiple independent proposals (one renamed to agent-manifest.txt in March 2026); no single coiner or standards body has been established" }, "related_to": [ "agents-md", "robots-txt", "llms-txt", "agent-identity" ], "contrast_with": "Unlike robots.txt, an exclusion contract that says what crawlers may NOT fetch, agents.txt is meant to advertise what agents CAN do — a capability and terms manifest rather than an access-denial list.", "example": "By March 2026 the agents.txt namespace was crowded enough that one prominent proposal was renamed agent-manifest.txt to disambiguate it from competing uses.", "source": "https://github.com/asturwebs/agents-txt", "status": "proposed", "why_it_matters": "agents.txt represents the emerging move from blocking agents (robots.txt) to inviting and instructing them — but its contested namespace means a site should track which proposal wins before committing.", "sameAs": [], "bridge_entity": "protocols/discovery/agents-txt", "last_verified": "2026-06-15", "md_twin": "/glossary/agents-txt.md" }, { "id": "agntcy", "term": "AGNTCY", "category": "protocols", "short_def": "An open-source 'Internet of Agents' infrastructure project — originally from Cisco — providing agent discovery, identity, messaging and observability across vendors.", "long_def": "Open-sourced by Cisco in March 2025 (with LangChain and Galileo) and welcomed by the Linux Foundation on 29 July 2025 with Dell, Google Cloud, Oracle and Red Hat as formative members. AGNTCY supplies cross-framework infrastructure: agent discovery via the Open Agent Schema Framework (OASF), cryptographically verifiable identity, multi-modal messaging and end-to-end observability.", "see_also": [ "a2a", "mcp", "agent-identity" ], "etymology_origin": "Open-sourced by Cisco (with LangChain and Galileo) in March 2025; welcomed by the Linux Foundation on 29 July 2025 with Dell, Google Cloud, Oracle and Red Hat as formative members.", "related_to": [ "a2a", "mcp", "agent-identity", "ucp" ], "contrast_with": "Unlike A2A, a single agent-to-agent communication protocol, AGNTCY is a broader infrastructure stack — discovery, identity, messaging and observability — meant to span multiple agent frameworks and protocols rather than define one wire format.", "example": "The Linux Foundation welcomed the AGNTCY project on 29 July 2025, with Cisco, Dell, Google Cloud, Oracle and Red Hat as formative members.", "source": "https://www.linuxfoundation.org/press/linux-foundation-welcomes-the-agntcy-project-to-standardize-open-multi-agent-system-infrastructure-and-break-down-ai-agent-silos", "status": "emerging", "why_it_matters": "AGNTCY is one of the main neutral infrastructure bets for multi-agent interoperability; a service joining the Internet of Agents may register its identity and capabilities through AGNTCY's framework.", "sameAs": [], "bridge_entity": "protocols/interop/agntcy", "last_verified": "2026-06-15", "md_twin": "/glossary/agntcy.md" }, { "id": "agent-card", "term": "Agent Card", "category": "identity", "short_def": "A JSON document an A2A agent publishes at /.well-known/agent-card.json that advertises its identity, skills, endpoints and authentication.", "long_def": "The Agent Card is A2A's discovery primitive — a machine-readable 'business card' hosted at a well-known URI (per RFC 8615) listing the agent's name, description, version, service endpoints, supported interfaces, capabilities (e.g. streaming) and the AgentSkill objects it offers. A client reads the card to decide whether and how to delegate a task to the agent.", "see_also": [ "a2a", "agent-identity", "agntcy" ], "etymology_origin": "Defined by the Agent2Agent (A2A) protocol; served at the well-known URI /.well-known/agent-card.json following RFC 8615 'Well-Known Uniform Resource Identifiers'.", "related_to": [ "a2a", "agent-identity", "agntcy", "web-bot-auth" ], "contrast_with": "Unlike llms.txt or AGENTS.md, which describe a site or repository for ingestion, an Agent Card describes a callable AGENT — its skills, endpoints and auth — so other agents can discover and delegate work to it.", "example": "An A2A agent publishes its Agent Card at /.well-known/agent-card.json, listing skills and endpoints so client agents can discover and delegate to it.", "source": "https://a2a-protocol.org/dev/topics/agent-discovery/", "status": "active", "why_it_matters": "The Agent Card is how a service makes itself discoverable and delegable in a multi-agent system; publishing one is the A2A counterpart of having a sitemap for agents.", "sameAs": [], "bridge_entity": "protocols/interop/a2a", "last_verified": "2026-06-15", "md_twin": "/glossary/agent-card.md" }, { "id": "streamable-http", "term": "Streamable HTTP", "category": "protocols", "short_def": "The MCP transport, introduced in spec version 2025-03-26, that carries client-server messages over a single HTTP endpoint and supersedes the older HTTP+SSE transport.", "long_def": "Streamable HTTP replaces MCP's original 2024-11-05 HTTP+SSE transport with a single-endpoint design: the server handles POST and GET requests and may optionally use Server-Sent Events to stream multiple messages, but can also run fully statelessly behind a load balancer. The TypeScript SDK v1.10.0 (17 April 2025) was the first to support it.", "see_also": [ "mcp", "content-negotiation" ], "etymology_origin": "Introduced in the Model Context Protocol specification version 2025-03-26 as the recommended remote transport, superseding the HTTP+SSE transport from spec version 2024-11-05.", "related_to": [ "mcp", "tool-use", "webmcp" ], "contrast_with": "Unlike the deprecated HTTP+SSE transport, which required a persistent server-sent-events connection, Streamable HTTP uses a single endpoint and can run statelessly behind a load balancer — SSE becomes optional rather than mandatory.", "example": "MCP introduced Streamable HTTP in spec version 2025-03-26; the TypeScript SDK v1.10.0 (17 April 2025) was the first release to support it.", "source": "https://modelcontextprotocol.io/specification/2025-03-26/basic/transports", "status": "active", "why_it_matters": "Streamable HTTP is the transport a site uses to host a scalable remote MCP server; choosing it over deprecated SSE is a concrete agent-readiness implementation decision.", "sameAs": [], "bridge_entity": "protocols/capability/mcp", "last_verified": "2026-06-15", "md_twin": "/glossary/streamable-http.md" }, { "id": "agent-skills", "term": "Agent Skills", "category": "knowledge-memory", "short_def": "An open standard from Anthropic that packages procedural knowledge as a folder with a SKILL.md file an agent discovers and loads on demand.", "long_def": "A Skill is a directory containing a SKILL.md (metadata plus instructions), optional scripts and resources, which an agent loads dynamically only when relevant — keeping the context window lean. Anthropic unveiled Agent Skills on 16 October 2025 and released it as an open standard at agentskills.io on 18 December 2025; Microsoft, OpenAI, GitHub, Figma and Cursor adopted it.", "see_also": [ "context-engineering", "mcp", "tool-use" ], "etymology_origin": "Unveiled by Anthropic on 16 October 2025 and released as an open standard at agentskills.io on 18 December 2025; each Skill is a folder with a SKILL.md file of metadata and instructions.", "related_to": [ "context-engineering", "mcp", "tool-use", "prompt-caching", "agentic-rag" ], "contrast_with": "Unlike MCP, which connects an agent to external tools and data at runtime, Agent Skills package reusable procedural knowledge (how to do a task) that the agent loads into context on demand — know-how, not a tool connection.", "example": "Anthropic released Agent Skills as an open standard on 18 December 2025, with Microsoft, OpenAI, GitHub, Figma and Cursor adopting it.", "source": "https://claude.com/blog/skills", "status": "active", "why_it_matters": "Agent Skills let a service ship its own how-to knowledge to any compatible agent as a portable folder, so agents can perform domain tasks correctly without bespoke prompting.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/agent-skills.md" }, { "id": "aeo", "term": "Answer Engine Optimization (AEO)", "category": "optimization", "short_def": "Structuring content so AI answer engines and assistants extract, trust and cite it as a direct answer rather than ranking it as a link.", "long_def": "AEO optimizes for extractability, factual density and cross-source consensus — the signals that get content lifted into AI Overviews, voice answers and featured snippets. It overlaps heavily with GEO and LLMO (often treated as the same practice under different names); the distinction is emphasis: AEO leans toward answer-engine and snippet surfaces, GEO toward generative-engine citation.", "see_also": [ "geo", "llmo", "ai-overviews" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/Generative_engine_optimization — 'Answer engine optimization' redirects to the GEO article on Wikipedia; it is a practitioner-coined term with no single coiner/date, often used interchangeably with GEO and LLMO" }, "related_to": [ "geo", "llmo", "ai-overviews", "zero-click", "agentic-seo" ], "contrast_with": "Unlike GEO, which is academically coined (Aggarwal et al., 2023) and frames the target as citation inside generated answers, AEO is a practitioner term emphasizing direct-answer surfaces — snippets, voice, AI Overviews — though the two practices largely overlap.", "example": "On Wikipedia, 'Answer engine optimization' redirects to the Generative Engine Optimization article, reflecting how interchangeably the two terms are used.", "source": "https://www.tryprofound.com/resources/articles/what-is-answer-engine-optimization", "status": "active", "why_it_matters": "AEO is the SEO-adjacent vocabulary many site owners arrive with; framing agent-readiness in AEO terms bridges existing search audiences into citation-first thinking.", "sameAs": [ "https://en.wikipedia.org/wiki/Generative_engine_optimization" ], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/aeo.md" }, { "id": "llmo", "term": "LLM Optimization (LLMO)", "category": "optimization", "short_def": "Optimizing content for how large language models evaluate, trust and select it as a source — a practitioner sibling of GEO and AEO.", "long_def": "LLMO (sometimes AIO, AI Optimization) focuses specifically on large-language-model citation behavior: clear claims, source-able facts, structure a model can parse. It shares roughly 80% of its methods with GEO; the main difference is provenance and scope — GEO came from academia and covers all generative engines, LLMO arose among practitioners and targets LLMs specifically.", "see_also": [ "geo", "aeo", "share-of-model" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://ahrefs.com/blog/geo-is-just-seo/ — 'LLMO'/'AIO' are practitioner-coined umbrella terms (2024-2025) with no single coiner or standards body; widely treated as synonyms of GEO" }, "related_to": [ "geo", "aeo", "share-of-model", "agentic-seo" ], "contrast_with": "Unlike GEO, which the founding paper scopes to all generative engines, LLMO is the practitioner framing aimed specifically at large language models — but the two share roughly 80% of their methods and are often used interchangeably.", "example": "Industry commentary (e.g. Ahrefs, 2025) argues GEO, LLMO and AEO are largely the same practice under different names — 'it's all just SEO'.", "source": "https://ahrefs.com/blog/geo-is-just-seo/", "status": "active", "why_it_matters": "LLMO is one of several near-synonyms a site owner will encounter; disambiguating it from GEO and AEO prevents chasing three 'different' strategies that are really one.", "sameAs": [], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/llmo.md" }, { "id": "agentic-seo", "term": "Agentic SEO", "category": "optimization", "short_def": "A model of search optimization in which autonomous AI agents continuously plan, execute and refine SEO actions across both classic search and AI answer systems.", "long_def": "Agentic SEO shifts optimization from a periodic human task to a real-time agent loop: agents analyze live signals, decide the next action, and apply updates directly. It has two readings — using agents to DO SEO (automation), and optimizing a site so agent-buyers and AI systems can discover, parse and surface it (the agent-readiness reading). The second is the one that bridges to agent-as-buyer thinking.", "see_also": [ "geo", "aeo", "agent-as-buyer" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://searchengineland.com/guide/agentic-ai-in-seo — 'agentic SEO' is a 2025-2026 practitioner term with no single coiner; used both for AI-agent-run SEO workflows and for optimizing sites for agent consumption" }, "related_to": [ "geo", "aeo", "agent-as-buyer", "agent-experience", "agentic-web" ], "contrast_with": "Unlike GEO, which optimizes content to be cited inside AI answers, agentic SEO also covers structuring a site (APIs, rich metadata, real-time data) so autonomous agents — including agent-buyers — can act on it, not just quote it.", "example": "Search Engine Land's 2026 guide frames agentic SEO as autonomous agents that continuously scan trends and restructure content for AI systems in real time.", "source": "https://searchengineland.com/guide/agentic-ai-in-seo", "status": "emerging", "why_it_matters": "Agentic SEO reframes optimization for a world where the visitor is an agent that acts, not a human that clicks — the through-line from GEO to full agent-readiness.", "sameAs": [], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/agentic-seo.md" }, { "id": "zero-click", "term": "Zero-Click Search", "category": "optimization", "short_def": "A search where the user gets their answer on the results page itself — via a snippet, knowledge panel or AI Overview — without clicking through to any site.", "long_def": "Popularized by Rand Fishkin of SparkToro, whose 2019 research found that over half of Google searches ended without a click to an external property. AI Overviews and answer engines accelerate the trend, making 'being cited in the answer' more valuable than 'ranking for the click' — the economic premise behind GEO and AEO.", "see_also": [ "ai-overviews", "geo", "aeo" ], "etymology_origin": "Popularized by Rand Fishkin (founder of SparkToro, former CEO of Moz); his research published 13 August 2019 reported that in June 2019, for the first time, a majority of Google.com browser searches ended in zero clicks.", "related_to": [ "ai-overviews", "geo", "aeo", "share-of-model", "grounding" ], "contrast_with": "Unlike a traditional organic click, where the user lands on your page, a zero-click result satisfies the query on the SERP — so the success metric shifts from traffic to being the cited source inside the answer.", "example": "Rand Fishkin's SparkToro research (13 August 2019) found that in June 2019, for the first time, a majority of Google.com browser searches resulted in zero clicks.", "source": "https://sparktoro.com/blog/less-than-half-of-google-searches-now-result-in-a-click/", "status": "active", "why_it_matters": "Zero-click is the structural reason agent-readiness and GEO matter: if users (and agents) never click, the only way to win is to be the cited source inside the answer.", "sameAs": [], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/zero-click.md" }, { "id": "ai-overviews", "term": "AI Overviews", "category": "optimization", "short_def": "Google Search's AI-generated answer summaries, shown above the classic links, which cite the sources they synthesize.", "long_def": "Launched in the US on 14 May 2024 at Google I/O as the rebrand and general-availability release of Search Generative Experience (SGE, previewed May 2023). AI Overviews synthesize an answer from multiple pages and link the sources, expanding zero-click search and making source citation — the GEO/AEO target — the new prize for content owners.", "see_also": [ "geo", "aeo", "zero-click" ], "etymology_origin": "Launched in the United States on 14 May 2024 at Google I/O as the general-availability rebrand of Search Generative Experience (SGE), which Google had previewed on 10 May 2023.", "related_to": [ "geo", "aeo", "zero-click", "grounding", "share-of-model" ], "contrast_with": "Unlike a featured snippet, which quotes one source verbatim, an AI Overview synthesizes an answer across multiple pages and cites several — so the optimization goal is being one of the cited sources, not owning a single box.", "example": "Google launched AI Overviews in the US on 14 May 2024, rebranding the Search Generative Experience it had previewed in May 2023.", "source": "https://en.wikipedia.org/wiki/AI_Overviews", "status": "active", "why_it_matters": "AI Overviews are where most users now first meet AI-synthesized answers; being cited in them is a primary, measurable GEO outcome for a content owner.", "sameAs": [ "https://en.wikipedia.org/wiki/AI_Overviews" ], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/ai-overviews.md" }, { "id": "share-of-model", "term": "Share of Model", "category": "optimization", "short_def": "A metric for how often a brand appears in AI-generated answers across prompts, relative to competitors — the AI-era analogue of share of voice.", "long_def": "Share of model (closely related to 'AI share of voice') measures a brand's slice of the AI conversation: if models mention brands 100 times across tracked prompts and a brand accounts for 25, its share is 25%. It is the headline output of GEO/AEO measurement tools (Profound, Ahrefs Brand Radar, Semrush) and reframes visibility from rankings to presence inside model answers.", "see_also": [ "geo", "aeo", "zero-click" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://www.tryprofound.com/ — 'share of model' is an industry metric name (also a product/company name) with no single coiner; closely synonymous with 'AI share of voice'" }, "related_to": [ "geo", "aeo", "zero-click", "llmo" ], "contrast_with": "Unlike a search ranking, which measures position in a list of links, share of model measures the percentage of AI answers in a category that mention your brand — presence inside the answer, not order on a page.", "example": "If AI models mention brands 100 times across tracked prompts and a given brand accounts for 25 of those mentions, its share of model is 25%.", "source": "https://www.tryprofound.com/resources/articles/what-is-answer-engine-optimization", "status": "emerging", "why_it_matters": "Share of model is the KPI that makes GEO accountable; it lets a brand measure whether agent-readiness and citation work are actually increasing its presence in AI answers.", "sameAs": [], "bridge_entity": "geo", "last_verified": "2026-06-15", "md_twin": "/glossary/share-of-model.md" }, { "id": "agent-as-buyer", "term": "Agent-as-Buyer", "category": "commerce", "short_def": "The pattern where an AI agent, not a human, searches, evaluates and completes a purchase on the user's behalf.", "long_def": "Agent-as-buyer is the demand side of agentic commerce: the agent compares options, makes the decision and transacts, with the human authorizing scope in advance. It changes optimization targets — product data must be machine-parseable, APIs and structured feeds matter more than visual merchandising, and checkout must accept protocols like ACP, AP2 and x402. McKinsey has projected agentic commerce could reach $1 trillion in US retail by 2030.", "see_also": [ "agentic-commerce", "acp", "agentic-seo" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://en.wikipedia.org/wiki/Agentic_commerce — 'agent-as-buyer' is a descriptive pattern term within agentic commerce, with no single coiner; the parent category 'agentic commerce' has a Wikipedia entry" }, "related_to": [ "agentic-commerce", "acp", "ap2", "agentic-seo", "agent-identity" ], "contrast_with": "Unlike agentic commerce as a whole, which spans buying and selling, agent-as-buyer names specifically the demand-side actor — the purchasing agent whose needs reshape product data, feeds and checkout toward machine-readability.", "example": "In an agent-as-buyer flow, a shopping agent compares products and completes checkout via a protocol like ACP; McKinsey has projected agentic commerce could reach $1 trillion in US retail by 2030.", "source": "https://en.wikipedia.org/wiki/Agentic_commerce", "status": "emerging", "why_it_matters": "Agent-as-buyer is why a merchant must expose machine-readable product data and an agent-acceptable checkout; the buyer it must win over is increasingly software, not a human shopper.", "sameAs": [ "https://en.wikipedia.org/wiki/Agentic_commerce" ], "bridge_entity": "protocols/payments/acp", "last_verified": "2026-06-15", "md_twin": "/glossary/agent-as-buyer.md" }, { "id": "http-message-signatures", "term": "HTTP Message Signatures", "category": "identity", "short_def": "An IETF standard (RFC 9421) for cryptographically signing components of an HTTP message so a server can verify who sent it and that it was not altered.", "long_def": "Published as a Standards Track RFC in February 2024 (editors A. Backman and J. Richer, with M. Sporny), RFC 9421 defines how to sign chosen parts of a request or response and supports algorithms including EdDSA over Curve25519 (Ed25519). It is the cryptographic foundation Web Bot Auth builds on to prove agent identity, since user-agent strings are spoofable.", "see_also": [ "web-bot-auth", "agent-identity", "verifiable-credentials" ], "etymology_origin": "Published by the IETF as RFC 9421 'HTTP Message Signatures' (Standards Track, February 2024); editors Annabelle Backman and Justin Richer, with Manu Sporny; supports EdDSA over edwards25519 among other algorithms.", "related_to": [ "web-bot-auth", "agent-identity", "verifiable-credentials", "prompt-injection" ], "contrast_with": "Unlike Web Bot Auth, which is the specific agentic-web scheme for identifying bots, HTTP Message Signatures (RFC 9421) is the general-purpose signing mechanism it is built on — the primitive, not the application.", "example": "RFC 9421 (February 2024) standardized HTTP Message Signatures with support for Ed25519; Web Bot Auth uses it so a server can verify a request genuinely came from a declared agent.", "source": "https://www.rfc-editor.org/rfc/rfc9421.html", "status": "active", "why_it_matters": "HTTP Message Signatures are the standards primitive under verified-agent access; understanding RFC 9421 is the basis for trusting, rate-limiting or charging an agent by identity.", "sameAs": [ "https://datatracker.ietf.org/doc/html/rfc9421" ], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/http-message-signatures.md" }, { "id": "verifiable-credentials", "term": "Verifiable Credentials", "category": "identity", "short_def": "A W3C standard for tamper-evident, cryptographically verifiable digital credentials that prove a claim about a subject without contacting the issuer.", "long_def": "The Verifiable Credentials Data Model 2.0 became a W3C Recommendation on 15 May 2025. A VC binds claims (e.g. 'this agent is operated by X' or 'this principal authorized this scope') to an issuer's signature, so a verifier can check authenticity and integrity offline. In the agentic web they are a candidate mechanism for portable agent and delegation identity.", "see_also": [ "agent-identity", "http-message-signatures", "delegation" ], "etymology_origin": "Verifiable Credentials Data Model 2.0 published as a W3C Recommendation on 15 May 2025 by the W3C Verifiable Credentials Working Group.", "related_to": [ "agent-identity", "http-message-signatures", "delegation", "web-bot-auth" ], "contrast_with": "Unlike HTTP Message Signatures, which authenticate a single live request, a verifiable credential is a portable, reusable attestation about a subject that a verifier can check independently of the issuer — a credential, not a per-request signature.", "example": "The W3C published the Verifiable Credentials Data Model 2.0 as a Recommendation on 15 May 2025.", "source": "https://www.w3.org/TR/vc-data-model-2.0/", "status": "active", "why_it_matters": "Verifiable credentials are a leading candidate for portable agent identity and delegation proofs — letting a site trust 'who an agent is and what it may do' without a central lookup.", "sameAs": [ "https://en.wikipedia.org/wiki/Verifiable_credentials" ], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/verifiable-credentials.md" }, { "id": "delegation", "term": "Scoped Delegation", "category": "identity", "short_def": "Granting an agent a limited, explicit set of permissions to act on a principal's behalf — bounded in scope, budget and time.", "long_def": "Delegation answers the second half of agent identity: not just 'which agent?' but 'authorized to do what, for whom, within what limits?'. Scoped delegation expresses bounded authority — e.g. spend up to a cap, only with approved merchants, for a fixed window — and underpins agent payment mandates (AP2) and permissioned payment rails (MPP). It is the principle that keeps an autonomous agent from exceeding what its principal allowed.", "see_also": [ "agent-identity", "ap2", "verifiable-credentials" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://datatracker.ietf.org/doc/html/rfc6749 — scoped delegation generalises OAuth 2.0 'scopes' and delegated authorization (RFC 6749, 2012) to AI agents; no single coiner for the agent-delegation sense" }, "related_to": [ "agent-identity", "ap2", "verifiable-credentials", "agent-as-buyer", "mpp" ], "contrast_with": "Unlike authentication, which proves who an agent is, scoped delegation defines what that agent is permitted to do on a principal's behalf — bounded authority, not identity; the two together gate any consequential agent action.", "example": "An AP2 payment mandate is a form of scoped delegation: it authorizes an agent to spend only up to a set amount, only for a specified transaction, on the human's behalf.", "source": "https://datatracker.ietf.org/doc/html/rfc6749", "status": "emerging", "why_it_matters": "Scoped delegation is the safety boundary of autonomous action; without it, granting an agent access means granting it unbounded authority — unacceptable for payments or writes.", "sameAs": [], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/delegation.md" }, { "id": "agent-gateway", "term": "Agent Gateway", "category": "identity", "short_def": "A proxy that sits between agents and the tools or models they call, enforcing security, access-control and observability policies on agent traffic.", "long_def": "An agent gateway (sometimes 'agent firewall' for the security-focused variant) is a networking layer built on agent-native protocols like MCP and A2A. It inspects and governs agent-to-tool, agent-to-model and agent-to-agent calls — applying policy, redacting secrets and PII, blocking prompt-injection and SSRF, and logging everything. It is where a site can centrally control what visiting or internal agents are allowed to do.", "see_also": [ "prompt-injection", "agent-identity", "mcp" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://agentgateway.dev/ — 'agent gateway'/'agent firewall' are an emerging product/infrastructure category (e.g. agentgateway.dev, MCP firewalls) with no single coiner or standards body" }, "related_to": [ "prompt-injection", "agent-identity", "mcp", "web-bot-auth", "delegation" ], "contrast_with": "Unlike a traditional network firewall, which filters IP/port traffic, an agent gateway understands agent protocols — it inspects MCP tool calls and A2A messages for prompt injection, secret leakage and policy violations, not just packets.", "example": "Open-source agent gateways such as agentgateway provide drop-in security, observability and access control for agent-to-LLM, agent-to-tool and agent-to-agent communication across frameworks.", "source": "https://agentgateway.dev/", "status": "emerging", "why_it_matters": "An agent gateway is where an organization enforces agent-readiness safely at scale — verifying identity, scoping permissions and blocking injection in one controllable choke point.", "sameAs": [], "bridge_entity": "protocols/identity/web-bot-auth", "last_verified": "2026-06-15", "md_twin": "/glossary/agent-gateway.md" }, { "id": "agentic-rag", "term": "Agentic RAG", "category": "knowledge-memory", "short_def": "Retrieval-augmented generation in which an agent plans, retrieves, evaluates and re-retrieves iteratively, instead of fetching context once.", "long_def": "Where naive RAG runs a single similarity search and hands the results to the model, agentic RAG turns retrieval into a control loop: the agent decides when and what to retrieve, judges whether the results are sufficient, retries or switches tools (web search, SQL, APIs) and validates before answering. It trades latency and cost for reliability on complex, multi-step questions.", "see_also": [ "rag", "embeddings", "vector-database" ], "etymology_origin": { "value": null, "verify_status": "verify-against-primary-at-build", "source_hint": "https://arxiv.org/abs/2501.09136 — 'agentic RAG' is a 2024-2025 practitioner/research term building on RAG (Lewis et al. 2020); see the 2025 survey arXiv 2501.09136, with no single coiner for the exact term" }, "related_to": [ "rag", "embeddings", "vector-database", "agentic-loop", "agent-skills" ], "contrast_with": "Unlike naive (one-shot) RAG, which retrieves context a single time before generating, agentic RAG defers retrieval decisions to an agent that can retrieve, evaluate, re-retrieve and validate in a loop — a control loop, not a fixed pipeline.", "example": "A 2025 survey on agentic RAG (arXiv 2501.09136) frames it as turning retrieval from a static pipeline into an iterative, agent-driven control loop.", "source": "https://arxiv.org/abs/2501.09136", "status": "emerging", "why_it_matters": "Agentic RAG is how modern agents ground complex answers; sites that expose clean, chunkable, well-described content are easier for an iterating retrieval agent to use and cite.", "sameAs": [], "bridge_entity": "models", "last_verified": "2026-06-15", "md_twin": "/glossary/agentic-rag.md" }, { "id": "embeddings", "term": "Embeddings", "category": "knowledge-memory", "short_def": "Dense numeric vectors that represent text (or other data) so that semantically similar items sit close together in vector space.", "long_def": "The modern approach was crystallized by Word2vec (Mikolov, Chen, Corrado and Dean at Google, 2013), which learned high-quality dense word vectors that captured meaning by context. Today, embedding models turn documents and queries into vectors so similarity search can find relevant content by meaning rather than keyword — the retrieval engine under RAG and vector databases.", "see_also": [ "vector-database", "rag", "agentic-rag" ], "etymology_origin": "The dense-vector approach was established by Word2vec, introduced by Tomáš Mikolov, Kai Chen, Greg Corrado and Jeffrey Dean at Google in 2013 (arXiv 1301.3781); the broader 'word embedding' concept predates it.", "related_to": [ "vector-database", "rag", "agentic-rag", "grounding" ], "contrast_with": "Unlike keyword indexing, which matches exact tokens, embeddings place text in a continuous vector space so 'meaning-similar' content is found by vector distance — semantic match rather than lexical match.", "example": "Word2vec (Mikolov et al., Google, 2013) showed dense word vectors capture semantics so well that vector arithmetic like 'king − man + woman ≈ queen' holds.", "source": "https://en.wikipedia.org/wiki/Word_embedding", "status": "active", "why_it_matters": "Embeddings are how agents retrieve your content by meaning; clean, well-structured text embeds and retrieves better, making your site easier to surface in RAG answers.", "sameAs": [ "https://en.wikipedia.org/wiki/Word_embedding", "https://en.wikipedia.org/wiki/Word2vec" ], "bridge_entity": "models", "last_verified": "2026-06-15", "md_twin": "/glossary/embeddings.md" }, { "id": "vector-database", "term": "Vector Database", "category": "knowledge-memory", "short_def": "A database built to store embeddings and retrieve the nearest ones to a query vector using approximate nearest-neighbor search.", "long_def": "A vector database (or vector store) indexes high-dimensional embeddings and finds the most similar ones with Approximate Nearest Neighbor (ANN) algorithms — commonly HNSW graphs or quantization — under metrics like cosine distance. It is the storage-and-retrieval backbone of RAG: documents go in as vectors, a query vector comes in, and the closest chunks come out as context.", "see_also": [ "embeddings", "rag", "agentic-rag" ], "etymology_origin": "A database category that emerged with the rise of embedding-based retrieval; popularized by systems such as FAISS, Pinecone, Weaviate, Milvus and pgvector, using ANN indexes like HNSW.", "related_to": [ "embeddings", "rag", "agentic-rag", "grounding" ], "contrast_with": "Unlike a relational database, which retrieves rows by exact field matches, a vector database retrieves items by approximate nearest-neighbor distance between embeddings — similarity ranking rather than exact lookup.", "example": "Vector databases such as Pinecone and Weaviate use HNSW-based Approximate Nearest Neighbor search over embeddings to return the most semantically similar documents to a query.", "source": "https://en.wikipedia.org/wiki/Vector_database", "status": "active", "why_it_matters": "The vector database is where your content lives once an AI system ingests it; content that chunks and embeds cleanly is retrieved more accurately into the answers agents generate.", "sameAs": [ "https://en.wikipedia.org/wiki/Vector_database" ], "bridge_entity": "models", "last_verified": "2026-06-15", "md_twin": "/glossary/vector-database.md" }, { "id": "context-engineering", "term": "Context Engineering", "category": "knowledge-memory", "short_def": "The practice of curating exactly what information enters a model's context window at each step, so the task is solvable.", "long_def": "Popularized in June 2025 when Shopify CEO Tobi Lütke and AI researcher Andrej Karpathy endorsed the term over 'prompt engineering'. Karpathy called it 'the delicate art and science of filling the context window with just the right information for the next step'. In production agents the prompt is a tiny fraction of context; the rest is retrieved documents, tool outputs, history and state — all of which must be engineered.", "see_also": [ "agent-skills", "agentic-rag", "prompt-caching" ], "etymology_origin": "Popularized in June 2025: Shopify CEO Tobi Lütke endorsed the framing on 18 June 2025 and Andrej Karpathy publicly preferred 'context engineering' over 'prompt engineering' shortly after.", "related_to": [ "agent-skills", "agentic-rag", "prompt-caching", "token-economics", "rag" ], "contrast_with": "Unlike prompt engineering, which crafts a single instruction, context engineering manages the whole context window — history, retrieved docs, tool results and state — of which the prompt is only a small part.", "example": "On 18 June 2025 Shopify CEO Tobi Lütke endorsed 'context engineering', and Andrej Karpathy followed, calling it the art of filling the context window with just the right information.", "source": "https://simonwillison.net/2025/Jun/27/context-engineering/", "status": "active", "why_it_matters": "Context engineering explains why lean, well-structured, retrievable content wins: it is cheaper and more reliable to fit into an agent's context window than sprawling HTML.", "sameAs": [ "https://en.wikipedia.org/wiki/Prompt_engineering" ], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/context-engineering.md" }, { "id": "prompt-caching", "term": "Prompt Caching", "category": "knowledge-memory", "short_def": "Reusing the model's processed state for a repeated prompt prefix so identical leading context is not recomputed, cutting latency and cost.", "long_def": "Introduced by Anthropic on 14 August 2024, prompt caching marks a content block as a cache breakpoint; a later request that begins with the same exact bytes reads the cached state instead of reprocessing it. Cached input typically costs a fraction of normal input tokens (with a one-time write surcharge). It rewards stable, front-loaded context — a direct incentive to put durable, machine-readable material first.", "see_also": [ "token-economics", "context-engineering", "agent-skills" ], "etymology_origin": "Introduced by Anthropic for the Claude API on 14 August 2024; caches a prompt prefix marked as a breakpoint, with an ephemeral (short-TTL) cache reused across subsequent requests.", "related_to": [ "token-economics", "context-engineering", "agent-skills", "agentic-loop" ], "contrast_with": "Unlike RAG, which fetches new context per query, prompt caching reuses unchanged leading context across calls — it optimizes repeated identical input rather than retrieving fresh information.", "example": "Anthropic launched prompt caching on 14 August 2024; cached tokens cost a fraction of normal input tokens, rewarding stable, repeated prompt prefixes.", "source": "https://claude.com/blog/prompt-caching", "status": "active", "why_it_matters": "Prompt caching makes stable, front-loaded, machine-readable context cheaper to reuse across an agent's loop — another economic reason clean structure beats sprawling markup.", "sameAs": [], "bridge_entity": "agent-readiness", "last_verified": "2026-06-15", "md_twin": "/glossary/prompt-caching.md" }, { "id": "aaif", "term": "Agentic AI Infrastructure Foundation (AAIF)", "category": "protocols", "short_def": "A Linux Foundation body, formed in December 2025, that provides neutral governance for core agentic-web standards including MCP, A2A and AGENTS.md.", "long_def": "The AAIF (Agentic AI Foundation) was formed under the Linux Foundation on 9 December 2025 to steward agent infrastructure as vendor-neutral commons. Platinum members are AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI; its inaugural projects are the Model Context Protocol (MCP), the goose agent and AGENTS.md.", "see_also": [ "mcp", "a2a", "agents-md" ], "etymology_origin": "Formed under the Linux Foundation in December 2025 as the Agentic AI Foundation (AAIF); platinum members include AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI.", "related_to": [ "mcp", "a2a", "agents-md", "ucp", "agntcy" ], "contrast_with": "Unlike a single protocol such as MCP or A2A, the AAIF is the governance meta-entity that holds several of them — it does not define a wire format; it provides the neutral home that keeps them vendor-independent.", "example": "The AAIF, formed under the Linux Foundation in December 2025, houses MCP, A2A v1.0, AGENTS.md and UCP, with platinum members including Anthropic, Google, Microsoft and OpenAI.", "source": "https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation", "status": "active", "why_it_matters": "The AAIF is the neutral-governance anchor that makes the agentic web's core protocols safe to adopt — a site betting on MCP or A2A is betting on standards a vendor cannot unilaterally control.", "sameAs": [], "bridge_entity": "protocols", "last_verified": "2026-06-15", "md_twin": "/glossary/aaif.md" }, { "id": "google-extended", "term": "Google-Extended", "category": "identity", "short_def": "A robots.txt user-agent token that lets a site opt out of having its content used to train and ground Google's Gemini models, while leaving Google Search indexing unaffected.", "long_def": "Google-Extended is a control token, not a crawler. Adding 'User-agent: Google-Extended' with 'Disallow: /' to robots.txt tells Google not to use the site's content for training or grounding Gemini and Vertex AI generative models; normal Googlebot search crawling continues. Introduced by Google in September 2023.", "see_also": [], "etymology_origin": "Introduced by Google in September 2023 as an opt-out control for generative-AI training/grounding.", "related_to": [], "contrast_with": "Unlike a crawler token that blocks fetching, Google-Extended only governs AI training/grounding use — Googlebot search access is unaffected.", "example": "A publisher adds Google-Extended to robots.txt to keep its articles out of Gemini training while staying in Google Search.", "source": "https://developers.google.com/search/docs/crawling-indexing/overview-google-crawlers", "status": "stable", "why_it_matters": "It separates 'be findable in search' from 'be used for AI training' — a core opt-out lever in the access-economics debate.", "sameAs": [], "bridge_entity": "/access-economics", "last_verified": "2026-06-15", "md_twin": "/glossary/google-extended.md" }, { "id": "pay-per-crawl", "term": "Pay-per-crawl", "category": "commerce", "short_def": "A Cloudflare marketplace mechanism that lets a site charge AI crawlers per request — returning HTTP 402 to unpaid bots with a price in response headers.", "long_def": "Announced by Cloudflare on 1 July 2025 (private beta). A site sets a price; when an AI crawler requests a page without payment the edge answers HTTP 402 Payment Required with crawler-price headers, and Cloudflare acts as merchant of record to settle. It turns crawling from a free externality into a metered transaction.", "see_also": [], "etymology_origin": "Announced by Cloudflare, 1 July 2025 (private beta).", "related_to": [], "contrast_with": "Where RSL is a licensing standard (the terms), pay-per-crawl is an enforcement + settlement mechanism (the toll booth).", "example": "A news site enables pay-per-crawl so GPTBot must pay per fetch instead of crawling for free.", "source": "https://blog.cloudflare.com/introducing-pay-per-crawl/", "status": "beta", "why_it_matters": "One of the three access-economics levers (block / price / permit) and a live way publishers monetize AI crawling.", "sameAs": [], "bridge_entity": "/access-economics", "last_verified": "2026-06-15", "md_twin": "/glossary/pay-per-crawl.md" }, { "id": "rsl", "term": "RSL (Really Simple Licensing)", "category": "commerce", "short_def": "An open standard for machine-readable content-licensing terms, referenced from robots.txt, that tells AI systems how content may be used and at what price.", "long_def": "Really Simple Licensing (RSL) launched on 10 September 2025, backed by Reddit, Yahoo, Medium and People Inc. among others. It defines license models — free, attribution, subscription, pay-per-crawl, pay-per-inference — in an XML file referenced from robots.txt, so AI crawlers can read the terms before using content.", "see_also": [], "etymology_origin": "Launched 10 September 2025 by the RSL Collective.", "related_to": [], "contrast_with": "RSL states the licensing TERMS; Cloudflare pay-per-crawl is one mechanism that ENFORCES payment — RSL can name pay-per-crawl as a license model.", "example": "A publisher publishes an RSL file declaring pay-per-inference licensing for its archive.", "source": "https://rslstandard.org/press/rsl-standard", "status": "emerging", "why_it_matters": "The emerging neutral standard for AI content licensing — the 'terms' layer of access economics.", "sameAs": [ "https://en.wikipedia.org/wiki/Really_Simple_Licensing" ], "bridge_entity": "/access-economics", "last_verified": "2026-06-15", "md_twin": "/glossary/rsl.md" } ] }