What agent experience is
UX describes how comfortable a product is for a person to use. Agent experience (AX) asks the same question about an AI agent: can it understand the site, complete a task on the user’s behalf and not get stuck halfway. The term was coined by Netlify co-founder and CEO Mathias Biilmann in the essay “Introducing AX: Why Agent Experience Matters” (28 January 2025). His definition: the holistic experience AI agents have as the users of a product or platform.
For e-commerce the question is practical: if an agent cannot find a size, runs into a CAPTCHA or cannot parse the delivery terms, it moves on to another store — and the person may never learn you were among the options.
What AX is made of
Biilmann singled out access to the platform on the user’s behalf, clean APIs, documentation written for LLMs, and flows in which humans and agents work together. In an online store that breaks down as follows:
| Layer | What the agent needs | How to provide it |
|---|---|---|
| Understanding | Explicit data on products, prices, stock | Schema.org schema markup, a complete feed |
| Orientation | A site map and a description of capabilities for LLMs | llms.txt, API documentation |
| Action | A stable way to search and place orders | An API, Web MCP, agentic protocols (ACP, UCP, YCP) |
| Access | A path without dead ends for verified agents | Cryptographic identification instead of a CAPTCHA — Web Bot Auth |
| Terms | Delivery, returns and warranty as data | Dedicated fields in the feed and markup |
| Hand-off to the human | A clear moment for confirmation | A payment confirmation step for the user |
AX versus agent-readiness
Agent-readiness answers “does the store have the necessary components”; AX answers “do they work together when a real agent shows up”. You can have a feed, markup and an API and still lose agents — because the price in the feed differs from the page, the API is slow, or a pop-up covers the order button.
Important: AX can only be verified by running real tasks. A component checklist shows what exists, not where the agent gets stuck.
How to test and improve AX
- Write 3–5 typical tasks: “find a size M jacket under $150 delivered tomorrow”, “compare these two models”, “repeat my last order”.
- Run them through several agents — an assistant with web access and an agentic browser.
- For each run, record the outcome: did the agent finish, how many steps it took, where it got stuck.
- Fix in order: data first (attributes, prices, terms), then access (CAPTCHAs, pop-ups), then APIs and protocols.
- Repeat the runs after every major front-end release and track the share of completed tasks.