Agent Readiness CompareA square mark containing five stacked bars, the fourth in ultramarine.Agent Readiness Compare

UNIVERSITY · LESSON 7 · RUNNING IT · 5 MIN READ

How to run your first agent task test

One task, four page checks, and a written record of where the agent stopped.

Agent Readiness Compare editors · Spec status checked September 2026

Answer

Pick one task that matters commercially, such as explaining your pricing, and write down what a correct answer looks like. Check the pages the task depends on with a few file checks, then run the task with a real agent, record each step and where it stalled, fix the first failure and run it again.

On this page
  1. 1. Which task should you start with?
  2. 2. What should you write down first?
  3. 3. Which pages should you check before the run?
  4. 4. How do you run the task?
  5. 5. How do you read the result?
  6. 6. What do you do after the first task?

1.Which task should you start with?

Start with the task closest to revenue that an agent can attempt without an account. For many B2B sites that is: explain our pricing and recommend a plan for a team of a given size. It touches the pricing page, plan limits and often a contact or trial step, and a wrong answer is easy to spot.

2.What should you write down first?

Before running anything, write the expected answer: the plan names, the prices as published, the limits that decide the recommendation, and the page the agent should end on. Without it, you will be judging the agent's answer from memory, and two people will judge it differently.

3.Which pages should you check before the run?

  1. Fetch the pricing page with JavaScript disabled and confirm plan names, prices and limits are in the HTML (checklist L2.1 and L2.5).
  2. Confirm robots.txt and CDN bot rules do not block the agent you will use (L1.2 and L1.3).
  3. Open /llms.txt and confirm it links to the pricing page (L2.2).
  4. Check that the next step, a form or a trial link, has labelled fields and no challenge an agent cannot pass (L3.2 and L4.2).

A scanner covers much of this in one pass; ora.ai Scan and Cloudflare Is It Agent Ready both scan public URLs.

4.How do you run the task?

Give an agent the task in plain words, starting from your home page or from a search result, and record every step: which pages it opened, what it read, what it answered and where it stopped. ora.ai Journey records this path for an intent you choose. You can also run the task by hand with an assistant that browses, noting each step in a simple table with the time, the URL and what the agent did.

5.How do you read the result?

Compare the agent's answer with the expected answer line by line. A wrong price can mean the agent read an old or third-party page; a missing plan can mean a client-side toggle; a stop before the form can mean a challenge or an unlabelled field. Fix the first failure only, then run again, because a later step often behaves differently once an earlier one is fixed.

6.What do you do after the first task?

Add the next task from the list in lesson 6, keep the log dated, and repeat the runs after releases. Lesson 8 covers how to make that a routine.

Source: journey.ora.ai · ora.ai llms.txt · isitagentready.com · RFC 9309 · llmstxt.org · Reviewed Sep 2026

Back to University