Are BASE, STORES and Color Me Shop stores ready for AI agents?

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.

As of August 2026 we have not published comprehensive per-platform measurements. This article stays with what the structure of the purchase path allows us to say; per-platform results will be added here once our study completes.

Where agents stop

Two typical things that can stop an agent on this path are CAPTCHA and input fields without machine-readable labels.

Both of those belong to the platform's shared foundation. So yes — much of a hosted store's fate really is decided by the platform's design.

What the platform controls

• Signup and login, including whether a CAPTCHA is present
• The structure of cart and checkout screens
• Form HTML, including how labels are tied to inputs
• Site-wide robots.txt and bot mitigation

This cuts both ways. A shared foundation means that once the platform does the work, every store on it becomes ready at once — and if the foundation stops agents, every store on it stops together no matter how hard individual owners try.

What the shop owner still decides

• Product data in text — name, price, shipping, stock and size written as text rather than baked into an image.
• Self-contained descriptions — "DM us for details" is a dead end for an agent.
• Image alt text — set it wherever the platform allows.
• Custom pages — where HTML editing is available, the 10-point checklist applies directly.
• Platform settings — take inventory of what your admin panel actually lets you toggle instead of leaving defaults alone.

Which admin screens to open is written up per platform: the BASE playbook, the STORES playbook and the Color Me Shop playbook. Each one separates merchant-fixable rows from platform residual you have to hand to your agency or upstream.

The per-platform study is opt-in

We are preparing per-platform measurements, built only from stores whose owners volunteer them — we do not walk a store we were not invited into. Participants get a free report for their own store. See the Japan EC agent-readiness study.

Check your own store, not the platform's reputation

Plans, themes and settings change the path even within one platform, so "BASE is fine" or "Color Me is bad" is too coarse to act on. And a static checker is not the same as a walk: UCP Checker's July 2026 report found 11,414 verified stores and zero with a confirmed payment implementation (State of Agentic Commerce July 2026).

Agentrix walks an agent through your store and returns a score with the stop reason. Report generation is currently free. Before each run we verify site ownership — a DNS TXT record, a file under /.well-known, or a meta tag — so a platform that lets you add neither an arbitrary tag nor a DNS record cannot be diagnosed. We never walk a store without consent and never complete a purchase.