How Agentrix works

Correction (2026-09-15): This article previously relied on one of our runs against Shopify's public demo store (hydrogen.shop), recorded as "stopped at a signup CAPTCHA, 20/100". We no longer use that record as evidence. The run saved no screenshots or HTML, and the same detection does not reproduce on hydrogen.shop today. On review, our detector at the time flagged a CAPTCHA whenever the page HTML merely contained strings such as "recaptcha", and this run's record consists only of such string matches. We cannot say a CAPTCHA was actually shown, and we now consider it most likely a false positive. Our 2026-09-02 note kept the number "for reference"; that was not enough. The affected sentences have been rewritten.

Three steps

1. Pick a path — view the public demo, or diagnose your store. Zero-knowledge users usually start with the demo.
2. Diagnose the site — a real agent walks a real browser through signup → browse → cart → pre-checkout. It never completes a purchase, only checks whether the flow is passable. CAPTCHA stops the run.
3. Share the report — Agentrix Score, failure reasons, and fix hints in one document (plus screenshots when captured).

The example run we published, our Shopify demo-store article, was corrected on 2026-09-15. Its record said "13 actions, $0.53, CAPTCHA detected at signup", but there was no screen evidence and the stop was most likely a false positive of our detector.

Turning a report into a conversation

Share the report on screen and use it as the basis for a few questions:

• Looking at this score, which step matters most to fix first?
• Is bot verification, missing form labels, or forced signup already a known issue internally?
• Where does an agent stopping cost you — search, chat, somewhere else?
• If it reaches pre-checkout, who owns calling that a success?
• Can we set a target score and deadline for the next re-run right now?

Continuous monitoring (paid plans)

Sites keep changing, so a single score goes stale fast. Starter/Business/Agency plans automatically re-scan on a cadence based on plan and site count, and email you if the score drops. See Pricing for details.