I built an open standard for AI shopping agents after noticing my products were invisible to them
Five weeks ago I noticed something. People were asking ChatGPT and Claude to help them shop, and the AI was recommending specific products by name. Not categories. Not "check Google." Actual products, with reasons.
I run an ecommerce business and my products were not showing up. I looked for a solution and found nothing. Google Merchant Center feeds are built for price comparison engines, not for AI reasoning models. Schema.org Product markup helps but it does not have the intent layer that makes an AI confident enough to recommend something specific.
So I built Agenticfeed.
The core idea: AI agents do not need better keywords. They need intent data. The questions a product answers. The pain points it addresses. The scenarios it fits. That is what gets a product recommended when someone describes their situation to an AI assistant.
Here is what I shipped in five weeks:
A Cloudflare Workers backend that imports product catalogues from Shopify, WooCommerce, and Google Merchant Center feeds
An AI pipeline (Claude Haiku) that generates questions, problems, and use cases for every product
A daily cron worker that keeps stock and prices in sync
Shopify OAuth integration that injects a discovery tag into the merchant's theme automatically
Stripe billing with three tiers
An open standard on GitHub so anyone can implement it independently
The first live customer is a UK board game retailer. Their products are now discoverable by AI agents.
I published the feed format as an open spec at github.com/bluestratus/agenticfeed because the value of a discovery mechanism increases when more sites implement it. Same reason RSS autodiscovery worked.
Still very early. Would love feedback from anyone who sells physical products or has thought about AI commerce.
Interesting build.
The thing I'd be careful with now is assuming the problem is merchant awareness.
The harder question may be what merchants need to believe before they care about becoming discoverable by AI agents in the first place.
Right now the product, the standard, and the technology are all clear. The first merchant decision feels less clear.
I wouldn't try to solve that loosely in the thread because it changes the positioning, proof, and first acquisition path.
If you're open to it, share your email and I'll put the tighter first-customer angle together properly.
I am happy for you to share my email, I have been going to EC ( Entrepreneur Circle in UK ) and so far I 100% customer acquisition. Bots are crawling the content more and more every day.
Appreciate it.
What's the best email to send it to?
I'll keep it focused on the first-customer decision and merchant adoption angle rather than turning it into a long teardown.
This is a really interesting direction.
Most ecommerce optimization has historically been built around search engines and human browsing behavior, but AI-assisted shopping changes the problem entirely. Instead of matching keywords, the model is trying to understand intent, context, constraints, and use cases before making a recommendation.
What stands out to me is the idea that product discovery data should answer questions like "Who is this for?" and "What problem does it solve?" rather than just exposing product attributes. That's much closer to how people actually ask AI assistants for recommendations.
I'm curious how you're thinking about adoption. Do you see this becoming a standalone standard that AI platforms eventually consume directly, or more of a merchant-side optimization layer that increases the likelihood of products being understood and surfaced by models?
Also, shipping Shopify integration, billing, synchronization, and an open spec in five weeks is impressive. It'll be interesting to see whether AI commerce evolves toward a few dominant standards the same way RSS and schema markup eventually did.
Thanks. You've put your finger on exactly the tension I'm navigating.
The honest answer is: both, but in sequence. Right now it's a merchant-side optimisation layer. The path of least resistance is helping merchants get their products understood and surfaced by the models that already exist. That's the immediate value and the thing I can ship and prove.
But the longer game is the standalone standard. The RSS analogy you used is the one I keep coming back to. RSS didn't win because a single platform mandated it. It won because the format was useful enough that both publishers and readers adopted it independently, and the network effect did the rest. I think agentic feeds could follow the same arc: enough merchants publish structured intent data, enough AI platforms learn to consume it, and it quietly becomes infrastructure.
The risk is fragmentation. Google's Universal Commerce Protocol, IAB's AAMP, and a few others are all staking out similar territory. I'm watching A2A closely too, since an agenticfeed node that's A2A-compliant would let shopping agents discover and delegate to it directly rather than passively hoping to be cited. That's a meaningful shift from optimisation layer to active participant.
On the five weeks: the scope was deliberately narrow. Shopify webhooks for sync, Stripe for billing, a spec that's readable in an afternoon. The temptation with something like this is to boil the ocean before anyone's using it. I'd rather have ten merchants getting real value from a simple format than a comprehensive standard nobody's implemented yet.
Keen to hear how you're thinking about the standards question. Do you see AI platforms actively consuming open specs, or building proprietary layers and making merchants come to them?
I think you're taking the right approach. It's much easier to get adoption around a simple solution that delivers value today than a comprehensive standard that's years ahead of the market.
The RSS comparison is interesting too. Sometimes the winners aren't the most sophisticated standards—they're the ones that are simple enough for people to actually implement. Looking forward to seeing how AgenticFeed evolves.