# Agent experience

> Agent experience (AX) is the design discipline of making a product usable by AI agents, not just people. As users delegate tasks to agents, a product's capabilities need to be discoverable and callable as typed tools with stated effects (through MCP, llms.txt and a server card) so an agent can act correctly on a user's behalf.

Canonical: https://peeve.ai/glossary/agent-experience/

## Why does agent experience matter now?

Because your next users increasingly won't have eyes. A person asks their agent to change a setting, and the agent tries to use your product. If your only interface is a chat widget built for humans, the agent scrapes and guesses. AX is the practice of giving agents a first-class surface instead.

## What makes a product agent-ready?

Machine-readable surfaces (llms.txt, AGENTS.md, an MCP server card) plus capabilities exposed as typed tools scoped to the acting user, each declaring its effect before it runs. Agent-readiness is testable: an agent should be able to discover what your product can do and do it, without a human translating the UI for it.

**Related:** MCP (Model Context Protocol), Capability registry, Deflection rate.