The words of agent access.
What each term means, what our drives record for it and the products behind it.
What we measure.
Agent accessWhether an AI agent can use a product on its own: find it, sign up, get a key, do the work, pay and leave, with no person stepping in.Agent Access LevelsOur scale from L0 to L5 for how far an AI agent gets in a product on its own. L0 means there is nothing for an agent to use. L5 means it gets in, does the work, pays and leaves without a person.DriveOne run of our AI agent through a real product: it reads the docs, signs up, gets a key, does the core work, pays where it can and deletes the account.Gets in on its ownAn agent has working access with no person helping: its own sign-up, a key over the API, pay per call or no key at all.Needs a humanA step a person had to do for the agent: solve a CAPTCHA, type a phone code, enter a card or click a link in a browser.
At the door.
WallAnything that stops an AI agent at a product's door: a bot check, a phone number at sign-up, a card before first use, a sign-up that only works in a browser.Bot checkA check in front of a page or a form that decides whether the visitor is a person: a puzzle, a checkbox or a silent score. Each site chooses where it shows and which bots it lets through.Signed agentsAI agents that sign each request with a key the site can check, so the site knows who sent it. A site can let signed agents through a bot check that stops everyone else.Web Bot AuthThe draft standard that signed agents use: each request carries a signature, and the agent publishes its keys in a directory at a fixed address so a site can check them.robots.txt for AI agentsThe file at a site's root that tells crawlers which pages to stay off. Some sites add rules that name AI agents.API keyThe secret an agent sends with each call to a product's API. Without one, most APIs answer nothing useful.
Standards and tools.
MCPModel Context Protocol: an open standard for connecting AI agents to a product's tools and data. A product runs an MCP server, and an agent calls its tools.llms.txtA Markdown file on a site that points AI models and agents to the pages that matter, often with plain-text versions of the docs.OpenAPIA standard file that describes every route of an API: its paths, inputs and answers. An agent can read it to learn the API without reading the docs.CLIA product's command-line tool. An agent working in a terminal often signs in and works through it instead of the web app.x402An open payment standard built on HTTP's 402 Payment Required: the server answers a request with what it costs, and the agent pays and sends the request again.MPPThe Machine Payments Protocol: a way for a server to ask for payment over HTTP, so an agent can pay for a call or a plan without a checkout page.
Words of the field.
Agentic commerceAI agents buying for people and companies: finding the price, paying and getting what was paid for, with no person at the checkout.Agents as customersAI agents that choose, sign up for and pay for software on behalf of the people and companies they work for. A product that keeps them out loses those sales.Agent-readySaid of a product an AI agent can start using on its own: it finds the docs, signs up, gets a key and does the work without a person.Agent readinessHow ready a product looks for AI agents, usually as a score from reading its site: its llms.txt, its docs, its MCP server and its robots.txt.Agent-friendlyEasy for an AI agent to use: clear docs it can read, a way to sign up and get a key without a browser, and errors it can act on.AXAgent experience: what using a product is like for an AI agent, from the docs and the sign-up to the errors and the bill. It is to agents what UX is to people.Agentic webThe web as AI agents use it: sites and APIs that agents read, sign in to and pay on, next to the people who use them.
See which of these your product gets right.
Free: what an agent finds on your site and any wall in its way.