How to run a two-week agent readiness trial on your main task
Before you buy a tool or start a project, test one task end to end and write down what happened.
Am I Ready for Agents? editors · 27 September 2026 · Reviewed September 2026
A useful proof of concept tests one task, not a whole site. Pick the task, take the self-check as a team, run two scanners, watch an agent try the task if you have a tool that does that, and compare all of it with what your web team says. Two weeks is enough, and the output is a short list of fixes with owners.
Why test one task?
A site-wide score hides the step that matters. Customers' agents arrive with one job: buy this, book that, get a quote. If that job fails, a good score elsewhere does not help. A trial built around one task keeps the scope small and the result concrete.
Days 1 and 2: which task, and what counts as success?
Pick the task an agent is most likely to attempt for a customer: a purchase, a booking, a subscription or a quote request. Publishers can pick reading and citing an article.
Write the success test in one line, for example: an agent acting for a guest can find the product, read the total price with delivery, and reach the payment step without a person stepping in. Keep it narrow enough that you can tell whether it passed.
Days 3 and 4: where does the team disagree?
Ask the four or five people who own the pieces to take the self-check separately, then compare answers. The 30-minute review has an agenda. You end with the questions where people disagree or said not sure, and those become what the trial should answer.
Days 5 to 7: what do two scanners say?
Run at least two scanners on the pages your task uses. They check different things: Cloudflare's Is It Agent Ready groups its checks into discoverability, content accessibility, bot access control, protocol discovery and commerce, while is-agentic.com splits its checks into essential, recommended and bonus. Compare the failed checks, not the totals, and mark which failures sit on your task's path.
Remember what scanners cannot see. They work on public URLs, so a checkout or account area behind a login may be out of reach.
Days 8 to 10: can an agent finish the task?
A scanner tells you which files and settings exist. It does not by itself show whether an agent finishes the job. Some tools run agents: ora.ai says its Scan runs real agents against a site and its Journey tool shows what an agent found, where it stalled and whether it finished. Our tools list describes each option in its own words.
If you do not use such a tool, walk the task yourself as a guest and note every step where a person would have to step in. It is less rigorous, but it catches the obvious walls: a CAPTCHA at checkout, a price that appears only in an image, a date picker that needs a mouse.
Days 11 and 12: which settings could explain a failure?
Ask your web team three things: what the robots.txt says about the main AI crawlers; whether the CDN or host has an AI bot rule switched on, and whether it is set to log, challenge or block; and whether bot protection can recognise verified agents. These are the settings most likely to explain a failure the scanners did not report. The questions are in the find and trust groups on ask your web team.
Days 13 and 14: what goes in the write-up?
- The task and the success test, and whether it passed.
- Each failure, the area it belongs to (find, read, act, pay or trust) and who owns it.
- What the scanners reported that mattered to the task, and what they reported that did not.
- One decision on tools: buy nothing yet, use what your CDN or host already offers, or trial a paid tool for a named question.
- A date to repeat the trial.
Keep the write-up to a page. Its value is that it names owners and a date.
What should the trial not try to prove?
That agent readiness will raise traffic or sales. A trial shows whether an agent can complete one task; your own analytics are the place to look for business effects over time.