Yann Borie

Yann Borie

Founder & CEO

linkedin.com/in/yannborie

I'm building Serge. When a customer asks Claude, ChatGPT, or Operator to buy something on an e-commerce site, the agent either succeeds or gives up. Serge measures which — and tells the team exactly what to fix when it's the second one.

Scan now

Why Serge

Why I'm building this

An engineering lead I work with recently asked Claude Desktop to buy a pack of AA batteries on their own store. The agent landed on the homepage, navigated to the right product page, hit the variant selector — and stopped. The dropdown was a custom React component without an accessible name or proper role. Claude couldn't tell it was a dropdown. The session ended with no error, no log entry, no GA4 row. Just a quiet loss. That gap — the place where a customer's intent meets a structurally incomplete storefront — is what Serge measures. The engineering leads I've shown this to have had the same reaction: the silence is the worst part. You don't see the failure in any existing dashboard.

Background

What I've worked on

Before Serge, I spent time building products and infrastructure in the developer-tools and e-commerce space. The pattern I keep coming back to: the tools that succeed in this category are the ones that show you a number you can't see anywhere else, and ship a fix you can paste into your repo. Serge is shaped exactly that way.

The bet

Why agent commerce, why now

Three things stacked up in 2026. Claude Desktop and ChatGPT Operator made browser-driving agents a daily tool for normal users — not a research curiosity. A material share of e-commerce conversion funnels started coming via agent traffic that GA4 categorises as "direct." And most e-commerce stacks I look at carry structural gaps that prevent agents from completing buying tasks — variant selectors without ARIA, schema.org Product blocks missing offers, cart endpoints that 403 on agent user-agents. The market for "tell me whether this works" + "tell me exactly what to fix" is real and the category is largely empty. So I'm building Serge.

Right now

What I'm thinking about

Right now the scanner ships deterministic product-findability scores to anyone who pastes a domain. The paid product runs real Claude, ChatGPT, Operator, Perplexity, and Gemini agents against customer storefronts with a defined buying task, and surfaces the failure point as structured data with a fix per gap. The current focus is scaling beyond the first cohort of early customers and opening the scoring methodology as a published specification. If you're curious whether your store passes, paste your domain into the free scanner. If you want a deeper conversation — methodology, roadmap, what we're seeing in the wild — LinkedIn DM is open.

Related

Read next