Last Mile · The maturity model

Four dimensions of agent readiness

Two axes to find & read you (Discoverability, Understanding) and two to act. Acting is split by the money line: Interaction — the agent provides inputs and performs unpaid actions — and, below the line, Transaction — money-moving actions with payment and trust. Best practices are the behaviors; standards are the specs that implement them.

The four dimensions of agent readiness

Standard status Established Emerging Experimental — caution, low-cost bet only
The four dimensions of agent readiness — Discoverability, Understanding, Interaction and Transaction — with what each means, example agent tasks, best practices, supporting standards, and the deciding signal Last Mile tests for.
Dimension What it means Example agent tasks Best practices Standards Deciding signal · how we test it
Discoverability
Do agents find you?
What it meansAgents find your site and your capabilities — content as well as the tools and actions you expose. Example agent tasks
  • Surfacing directly for "find me a flight to Lisbon"
  • Locating your booking tool or MCP server
  • Appearing in an assistant's results, not only via an OTA
Best practices
  • Allow AI crawlers
  • Publish sitemap & headers
  • Expose a capability index
  • Authenticate trusted agents
Standards
robots.txt Sitemap Link headers DNS-AID MCP Server Card API Catalog OAuth discovery Web Bot Auth llms.txt Agent Skills identity.json
Deciding signal · how we test it
Discoverability
A real agent is prompted for the task — does the airline (or its function) surface and get reached at all?
Understanding
Can agents read you?
What it meansAn agent can read, parse and comprehend you — offers, rules, returned results — accessibility plus structure. Example agent tasks
  • Extracting a fare's baggage allowance & change fee without a PDF
  • Comprehending the flight options returned by a search
  • Following a clean path to the right page
Best practices
  • Serve clean text to agents
  • Expose structured data
  • Keep a clean structured path
  • Readable inputs & errors
  • Accessibility & ARIA labels
  • Semantic HTML5 structure
  • Server-side rendering (SSR)
  • Visual stability (low CLS)
Standards
Markdown negotiation schema.org / JSON-LD NDC WAI-ARIA WCAG HTML5 Core Web Vitals
Deciding signal · how we test it
Structured Path · Input legibility · Error legibility
Can the agent parse offers and read the inputs and errors, or is content locked in JS / PDFs?
Interaction
Can agents operate you? (unpaid)
What it meansThe agent actively operates you — supplies inputs, runs searches, fills forms and performs account actions that don't move money. Example agent tasks
  • Searching a flight — entering route and dates
  • Remote check-in; prefilling passenger data
  • Selecting a seat or meal (no payment)
Best practices
  • Operable forms & inputs
  • Expose actions as tools
  • Write access with identity
  • Passable consent wall
  • Recoverable errors
Standards
WebMCP MCP (actions) OAuth Protected Resource
Deciding signal · how we test it
Unpaid task completion · Failure recovery
Does the agent complete a search, check-in or selection end to end — recovering from errors — with no human handoff?
Money line — everything below moves real money
Transaction
Can agents pay? (the highest bar)
What it meansThe agent completes money-moving actions on the customer's behalf — with payment, agent auth and trust. Depends on Interaction. Example agent tasks
  • Booking and paying for a flight
  • Rebooking or rebuying during a disruption
  • Buying a seat, bag, or paid reservation
Best practices
  • Programmatic checkout
  • Accept verified-agent payment
  • Don't bot-block at checkout
Standards
ACP UCP x402 MPP AP2 NDC OrderCreate
Deciding signal · how we test it
Paid task completion
Does a verified agent complete a real paid booking end to end — payment cleared — with no human in the loop?

v2 · Actionability split by the money line into Interaction (unpaid) & Transaction (paid)