Skip to content
AX Framework
AX Evaluation

How agent-ready is your product?

Readiness is measured against real tasks, across the six pillars. Explore a sample AX Score report below to see how findings, severity, and recommendations fit together.

AX proposal · scoring methodMock data
Mock dataSample reports for fictional products. No real website or application was audited, and this page does not run audits.

Evaluated task: Buy a specific product within a budget and delivery date

Overall readiness

34/ 100

Maturity L1 · Legible

  • 2Critical
  • 2High
  • 3Medium
  • 1Low

Dimension scores

Each dimension is four checks scored 0–2, shown as a percentage. Select one to filter issues.

Identified problems 8/8

Suggested improvements

Highest severity first, then lowest effort.

  1. 1Accept an Idempotency-Key header on order creation and return the original order on replay.Recovery & Resilience · Critical · Effort M
  2. 2Introduce OAuth scopes such as orders:read and orders:create, with per-order spending limits.Trust & Control · Critical · Effort L
  3. 3Return a receipt with the order ID and a retrievable URL from the order action.Feedback & Observability · High · Effort S

Run it yourself

Evaluate today, by hand.

Everything you need for a manual evaluation is published: the checks for each pillar, the severity scale, and a task-based testing method. A focused evaluation of one task takes a few hours.

  1. 01Choose one task people would delegate
  2. 02Score the four checks for each pillar
  3. 03Run the task with at least one agent
  4. 04Classify issues by severity and fix the worst first

Future direction

Roadmap

Automated AX audits

A tool that runs parts of this evaluation automatically is a goal for the project. It does not exist yet, and nothing on this site performs an audit. This is how we expect it could work, and it will change with community input.

  1. PHASE 1

    Static inspection

    Fetch public pages and check for semantic structure, structured data, manifests, and machine-readable alternates.

  2. PHASE 2

    Contract review

    Read OpenAPI documents, MCP tool lists, or AX manifests and check descriptions, schemas, risk labels, and error vocabularies.

  3. PHASE 3

    Task runs

    Run scripted tasks with real agents against a staging environment and record success, guesses, and recovery.

  4. PHASE 4

    Report

    Produce a report in the shape shown above, with evidence attached to each issue.

Interested in shaping it? See how to contribute.