AI Agent purchasing interface?

[LabelKing]

There is a lot of fuss recently around Agentic AI, with agents running around with access to their users funds, and authorisation to do stuff - like booking things, or buying things, or whatever.

So I'm wondering if this is something we want BP to be looking at proactively?

Some info from AI about this emerging technology:

The two routes agents can take

The first is what you might call the "brute force" route: an agent that can operate a web browser like a human does, clicking through pages, filling in forms, and completing checkout. Tools like Anthropic's Claude with computer use, OpenAI's Operator, and similar browser-automation agents can in principle do this on any website. Today's tools in this mode look more like fancy search than truly autonomous shopping, and some polished demos don't actually work in the real world. The friction points are real: CAPTCHAs, login walls, multi-factor authentication, and checkout flows that weren't designed to be machine-navigated. It is technically possible on many sites, but it is fragile and unreliable.

The second route is purpose-built agent interfaces, which is where the serious momentum is. OpenAI's Agentic Commerce Protocol (ACP), developed with Stripe, has been live since September 2025 in ChatGPT. Google announced its own Universal Commerce Protocol (UCP) in January 2026, backed by Walmart, Target, Shopify, and 20+ other partners. These protocols give agents a structured, machine-readable way to query product catalogues, check availability, and complete checkout programmatically without ever "navigating" a website in the human sense.

In agentic commerce, AI agents make purchasing decisions via structured APIs rather than by navigating storefront interfaces. A shopping agent translates user intent into API calls, queries product catalogues, pricing services, availability feeds, and shipping estimators. If product data is unstructured, delayed, or inconsistent, the agent cannot reliably include it.

So if we're not supporting this kind of interface we could be increasingly losing sales to those that do. And of course we could be missing out on the chance to grab new customers while other e-commerce websites are still sleeping (though I expect all the big ones will be on this already - Etsy and shopify are already integrating with chatgpt, plus all those companies above are already backing the new UCP).

We could find that the very nature of customer acquisition changes (very quickly) so that not just agents will be ordering, but people will be finding us via AI rather than PPC or google search, or even order directly from chatgpt itself. (the chatgpt and etsy/shopify link up uses "Buy in ChatGPT" rather than just pushing people to the e-commerce website like PPC...

[Neil]

Nope! Not within BP, absolutely not, it’d slow servers down and BP are better concentrating on the core system. If they expanded the api then a ai agent operating outside on a different server with work just as well, without the complexity of it being coded into BP.

we already have ai functionality working from a webform on our site, I’d much rather see the time spent building the blog into the api, and extra functionality.

There are better uses for ai such as AdWords optimisation, we’ve halved our conversations costs in 10 weeks thanks to Claude

[LabelKing]

Nope! Not within BP, absolutely not, it’d slow servers down and BP are better concentrating on the core system. If they expanded the api then a ai agent operating outside on a different server with work just as well, without the complexity of it being coded into BP.

It doesn't have to be built in, there could be piece of "middleware" sitting on another server that handles the interface with the agent protocol, and then communicates with Bp via the API.

There are 3 main parts to selling to an agent via chatgpt (and presumably similar for other systems) - 1) providing a list of products, 2) some form of checkout (correctly calculating the total cost, and putting the actual order into BP, 3) the payment. No 1 is trivial, No3 is trivial if using stripe (apparently a single line of code), so No 2 is the one that needs some work - I think it could mostly be done, but may need a few tweeks to the API.

I'd rather BP did the middleware and got it working with their API than have to do it ourselves and possibly have to fudge some of the checkout stuff.

we already have ai functionality working from a webform on our site, I’d much rather see the time spent building the blog into the api, and extra functionality.

When you say AI functionality, do you mean talking to external AI agents, or something different.
And yes, blog functionality in the API!

[Loxta]

I have to agree with Neil, I would love to see the API functionality increase to allow for purchase order functionality.

Blog functionality would also be great, but I suspect that is open for some serious abuse. Even an upload and download function on the blogs would be very helpful to update existing blogs.

[Neil]

It doesn't have to be built in, there could be piece of "middleware" sitting on another server that handles the interface with the agent protocol, and then communicates with Bp via the API.

There are 3 main parts to selling to an agent via chatgpt (and presumably similar for other systems) - 1) providing a list of products, 2) some form of checkout (correctly calculating the total cost, and putting the actual order into BP, 3) the payment. No 1 is trivial, No3 is trivial if using stripe (apparently a single line of code), so No 2 is the one that needs some work - I think it could mostly be done, but may need a few tweeks to the API.

I'd rather BP did the middleware and got it working with their API than have to do it ourselves and possibly have to fudge some of the checkout stuff.

When you say AI functionality, do you mean talking to external AI agents, or something different.
And yes, blog functionality in the API!

If you are doing your site optimisation properly, you should be appearing in ai citations already, it’s all about reputation, and good content will get noticed. I’m actually not interested in having people checkout on ai, with orders passed over, I want them on my site where we can cross sell, up sell and get full details for re marketing. Not a one off sale.

At the end of the day the winner in the ai battle is going to be ……. Google. People trust it and google isn’t going anywhere. Chat gpt figures aren’t adding up.

Yes we currently have various ai functions working via web forms so we can offer our customers personalised advice based on their location. This also includes airtable, and make in the mix.

[LabelKing]

If you are doing your site optimisation properly, you should be appearing in ai citations already, it’s all about reputation, and good content will get noticed. I’m actually not interested in having people checkout on ai, with orders passed over, I want them on my site where we can cross sell, up sell and get full details for re marketing. Not a one off sale.

I guess it depends on your customers and your products. With business customers just buying specific off the shelf products from a narrow product range there isn't much value to having a human on the website (if they want 100 widgets then they want a 100 widgets, our only goal would be to make that purchase as quick and simple as possible), and if they start adopting AI agents to purchase their supplies to save them time on basic repetitive tasks then it would be good to at least be capable of taking those orders rather than losing them. I guess we're just in radically different markets!