# Agent Experience (AX)

> Agent experience (AX) is the practice of designing products, docs and APIs so AI agents can find, understand and operate them without human help.

Source: https://metavert.io/agent-experience  
Published: 2026-10-07  
Updated: 2026-10-07

**Agent experience (AX)** is the quality of the experience an AI agent has when it uses a product or platform, and the practice of designing for it. The term was coined by Netlify co-founder Mathias Biilmann in a January 2025 essay that defined AX as "the holistic experience AI agents will have as the user of a product or platform." It sits in a deliberate line of succession: user experience for people, developer experience for programmers, and now agent experience for the software that increasingly acts on behalf of both.

### Why It Became a Discipline

Biilmann's argument was competitive rather than aesthetic: "Platforms, tools or frameworks that are hard for large language models (LLMs) and agents to use will start feeling less powerful and require more manual intervention." Less than two years later the premise is easy to observe. JetBrains (August 2026) found 90% of professional developers using [AI coding agents](https://metavert.io/ai-coding-agents) at work at least weekly, and those agents routinely select and install tools without a person reading the vendor's website first. A product that an agent cannot sign up for, configure or debug is, for a growing share of sessions, a product that does not get used.

### The Four Questions

In a retrospective published in January 2026, Biilmann reduced AX to four areas, each phrased as a question a product team can answer honestly.

| Pillar | The question | What it looks like in practice |
| --- | --- | --- |
| Access | "Can the Agent access your product? Do they have the right permissions?" | Sign-up and authentication flows an agent can complete, scoped tokens, no step that requires a human in a dashboard |
| Context | "Does the LLM know about your product? Does it have the right context to use your product?" | Documentation as clean text, [llms.txt](https://metavert.io/llms-txt), accurate examples, repository instructions such as [AGENTS.md](https://metavert.io/agents-md) |
| Tools | "Are you building your product for agents? Do you offer the right tools for agents?" | A CLI, a stable API, an [MCP](https://metavert.io/model-context-protocol) server, machine-readable errors |
| Orchestration | "Can you trigger agent runs from your product? Can you pass the right context along?" | Hooks and events that start an agent with the relevant state attached |

The right-hand column is this page's interpretation, not Biilmann's wording. The pillars overlap with [context engineering](https://metavert.io/context-engineering), but AX is the supplier's side of the problem: what a product exposes, as opposed to what an agent's operator loads into the context window.

### Documentation Is Now Read Mostly by Machines

The clearest quantitative signal comes from documentation hosts. Mintlify reported in July 2026 that agents accounted for 66% of measured traffic across the developer docs it hosts (213 million agent requests against 105 million human page loads in the month), up from 15.2% at the start of the year. The figure is vendor-reported, covers only Mintlify customers, and counts AI crawlers and user-initiated fetchers alongside coding agents, so it describes machine readership broadly. In the same report, serving an llms.txt cut agent error rates by almost 90% in Mintlify's benchmark, because agents stopped guessing at URLs.

Ahrefs' June 2026 log study of 137,210 domains gives the complementary view. It found that 97% of llms.txt files received no requests at all, and that among the AI traffic that did arrive, coding agents and agent infrastructure were the largest category at 10.5% of requests, against 1.1% for AI retrieval bots. The reading consistent with both studies is that llms.txt has little demonstrated effect on [AI search](https://metavert.io/ai-search) visibility and real utility for agents doing a task against a product's documentation. AX and search optimization share techniques but answer different questions.

### In Practice

Most AX work is unglamorous: make the quickstart complete from a terminal, return errors that say what to do next, keep examples current, and publish the same content people read in a form machines can parse. Much of this is becoming a platform feature rather than hand-built plumbing. [LightCMS](https://metavert.io/lightcms), the CMS serving this site, publishes a Markdown copy of every page, generates llms.txt, and offers both an MCP server and a public read-only MCP endpoint for visitors' agents. Listing an MCP server in the [MCP Registry](https://metavert.io/mcp-registry) extends the same idea to discovery.

Evidence that AX investment changes which tools agents choose is still thin. Armature's September 2026 experiments show agents disagreeing with one another and often favouring familiar incumbents, which good documentation alone does not overcome. AX is best understood as removing reasons for an agent to fail or give up, not as a ranking lever.

## Related Topics

- [AI Coding Agents](https://metavert.io/ai-coding-agents) — The agents whose experience is being designed for
- [AGENTS.md](https://metavert.io/agents-md) — Context for agents working inside a repository
- [llms.txt](https://metavert.io/llms-txt) — Context for agents reading a website or docs
- [Model Context Protocol](https://metavert.io/model-context-protocol) — The standard way to offer agents tools
- [MCP Registry](https://metavert.io/mcp-registry) — Discovery for those tools
- [llms.txt vs MCP](https://metavert.io/compare/llms-txt-vs-mcp) — Static briefing versus live tools
- [Context Engineering](https://metavert.io/context-engineering) — The operator-side counterpart to AX
- [LightCMS](https://metavert.io/lightcms) — A CMS that ships Markdown twins, llms.txt and MCP
- [Developer Tools](https://metavert.io/developer-tools) — Where AX pressure is strongest

## Further Reading

- [Introducing AX: Why Agent Experience Matters](https://biilmann.blog/articles/introducing-ax/) — Mathias Biilmann, January 2025
- [One Year of AX](https://biilmann.blog/articles/one-year-of-ax/) — Mathias Biilmann, January 2026
- [The State of Docs Traffic: A 2026 Midyear Report](https://www.mintlify.com/blog/state-of-docs-traffic) — Mintlify, July 2026
- [llms.txt log-file study of 137,210 domains](https://ahrefs.com/blog/llmstxt-study/) — Ahrefs, June 2026
- [AI Coding Agents: Adoption Trends](https://blog.jetbrains.com/research/2026/08/ai-coding-agent-adoption-2026/) — JetBrains Research, August 2026
- [Which Tools Coding Agents Install](https://armature.tech/blog/which-tools-coding-agents-install) — Armature, September 2026