Claude is finally entering the agentic commerce race, but as a buyer, why would I want to limit myself to stores Claude or even ChatGPT whitelist? LLMs are slowly eating away the moat that marketplaces have since an agent in the terminal can scour the web and find the product you need without the product search dark patterns and sponsored products everywhere.
Automating the checkout is tricky, but with WebMCP, I feel like agents will be able to go and checkout directly on websites soon. I myself am building an open-source agentic commerce chat that uses MCP-UI under the hood and brings the full checkout experience to the chat.
Also, the entire premise of selling specific agents seems a bit fishy to me, my first thought is apps on the iPhone that got banned when Apple introduced an official app covering the same functionality in a new iOS.
I'm not sure I would risk building an agentic app that competes with this on Anthropic models out of fear of that scenario.
I would suggest that bringing the "full checkout experience" to chat is backward thinking--that approach adds more friction than a traditional checkout. The ideal flow is when checkout happens behind-the-scenes and the chat is simply a confirmation of purchase.
Yes it is a bit awkward going thru the checkout process in the chat. MCP-UI returns the cart UI so you can interact with the cart directly like change quantity or change shipping method. You can also paste in your address and the chat LLM will figure out how to handle the address submission for that specific site.
Also, we bring in the PayPal and Stripe Elements directly into the chat so you can enter your credit card info. Just like entering it on that merchants website.
I would rather not let agent use checkout. I think they are not "smart" enough to tell scam from real website apart. Even recently Claude/Opus referred me to scammers website telling me to buy a product there.
Agents checkout + risk of prompt injection => big nop!
What is even the legal basis here? Can the agent sign contracts on your behalf? What if it is tricked by the vendor into signing for a monthly subscription you never wanted?
Anyone else worried about the premise of Anthropic selling specific agentic tools that also use their models underneath? My first thought is apps on the iPhone that got banned after Apple introduced an official option with the same functionality in a new iOS.
I'm not sure I would risk building an agentic app that competes with this on Anthropic models out of fear of that scenario.
>Retailers running shopping agents on Claude have seen carts up to 35% larger and shoppers 60% more likely to complete a purchase.
I feel like this is a messed up thing to brag about. From dark patterns now to whatever this is. Like a dark nudge from something that can appear intelligent, acting as a person, with the goal of selling more and more.
The flip side is that it helps people find items they want, leading to less friction and buying more. Not that I trust it to be to the benefit of the consumer by default, but we've definitely seen better conversion and AOV increases after moving from straight lexical search to hybrid vector.
I said almost the same thing. From a business perspective, I get it. 35% increase in cart size??!! Sign me up!
From a consumer perspective, I'm going to avoid coming anywhere near something like this. They've got data harvested from all of humankind and are weaponizing it as this discrete marketing tactic, convincing me to spend 35% more money on things I didn't actually need.
Such a bleak outlook. "Robot tells me to buy so I buy"...
Who is still using Anthropic seriously, their price and limits are ridiculous these days, you can't get anything done unless you're paying top dollar on a metered pricing at ridiculous levels. Subscriptions only get you frustrated, and any agentic pipeline that needs to make a profit, that has actual margins, I seriously doubt is using any of their models.
The way they keep spinning off these niche offerings tells you one of 2 things, it's either working pretty well, or it is failing very, very badly.
They know they need to differentiate from other offerings that are catching up, but I don't know anyone who has just left their work up to any of these specialised products. Most try it say "it's interesting" and then move on.
What do you mean spinning off these nieche offerings?
This is akin to a demo / SDK ready to plug into a store, it’s just making it easy for devs to get started. it’s still using core models and APIs, no special magic models.
>> The way they keep spinning off these niche offerings tells you one of 2 things, it's either working pretty well, or it is failing very, very badly.
You posited two extremes, with no evidence for either, and with no other possibilities in between, or even outside the spectrum. That's beyond mere conjecture: it's just nonsense. I hope you realize that.
Maybe at this point it's all nonsense. I can only speculate with the framing of what what would lead me to barely push into so many verticals with no follow up.
Large businesses who choose enterprise solutions over cheaper indie/open-source solutions for similar reasons (dedicated support, sticky products designed for their industry, less regulatory concerns, etc.)
It's pretty clear that there is massive unmet demand for AI, and compute cannot keep up. So you can pretty reliably dial in the "free compute for inference" situation for any given lab based on their pricing and friction tactics.
Anthropic/Google getting stuffed (Google is an odd case here)
OAI doing alrightish
Grok/Meta basically doing giveaways
Chinese models trying to fill vacancies on 3rd party compute
FWIW, Claude is an available model of Enterprise Copilot, and I'm sure Microsoft is paying a pretty penny to keep that as an option. Between Chat and Claude, most of my workplace prefers to use Claude. To your point though, the user isn't paying, and doesn't need to concern themselves with cost. I'd imagine if any of them have an active subscription to either, it's Chat
For most tasks I tried the choice is either expensive Claude that will actually deliver vs Codex or others that are cheaper but won’t.
Having said that Terra is awesome for clearly defined (by Opus/Fable) research task, and Sol is amazing for adversarial review, but I won’t allow them to touch my codebase any more.
If someone has a workflow where they keep tight control over agents, reviews the code by hand and so on, I can imagine them being useful.
I seriously doubt that. There's some myth going on here (lol, myth->Mythos).
For example, this exact narrative you're framing was said back during Opus 4.6/4.7 and fast forward to today we have open weight models that perform at Opus 4.6 level.
> Who is still using Anthropic seriously...Subscriptions only get you frustrated
If you're using Claude in a pair-programming/"exoskeleton" sense, you'll probably never get close to maxing your subscription allotment unless you really lean on Fable.
The "you can't get anything done" comment confounds me, frankly.
This is going to be an extremely interesting space to watch. Will online shops open their perfectly sculpted and sales-optimized webpages to accept requests from chat bots? such that they can just do highly targeted requests for shoppers? Or will they implement custom chat bots that guide and nudge you along your purchasing journey on the page itself? I'm very curious how consumer behavior will evolve from here. is every online search still going to start with a google query? or maybe soon a query on claude and chatgpt? or maybe three separate queries to the chat bots of the respective stores?
I don't want this. I want the diametric opposite of this: I want stores that expose their entire real-time catalog as RAG endpoints and let my agents go bonanza on them.
I'd happily a large premium for this, if any hungry startup people are looking for ideas.
I'm not interested in using agents controlled by vendors. Our interests are adverse! Even simple search on modern storefronts is an anti-consumer feature that promotes (upranks) defective products and things I wasn't interested in, and hides what I actually need, to maximize my eyeball time in their store. I don't want that. I would like to view as few commercial offers as possible. I would like to waste as little time as possible in the meaningless-to-my-life task of navigating stores.
Ok Shopify, you called the demons, now Anthropic makes you obsolete!
You know, maybe merchants now have to "justify to the bank why they are using Shopify instead of Anthropic" just like your software engineers have to "justify not using AI".
A new AI purity spiral is emerging that devours one by one.
The LAST thing you want to be non-deterministic is a payment system, right?
What is the possible use of AI in the payment process? I fill in fields and that gets whisked away to an API somewhere to verify. Why do I possibly need Claude involved?
What I find peculiar about this repository is that Anthropic expects that the buying agent will sit inside the merchant's own platform or application. This seems antithetical to what most people building in agentic commerce are expecting, which is that product discovery will happen in another LLM Chat interface, outside the merchant's platform.
Either they are deliberately leaving out a space to build or, which is more likely, they just want to get as many merchants accustomed to relying on Claude as the intelligence and will then export their data to their own chat interface.
57 comments
[ 15.5 ms ] story [ 1225 ms ] threadThats more of a 2 year old thing...
This has nothing to do with Claude - just showing neat things out into the world for people to find.
Automating the checkout is tricky, but with WebMCP, I feel like agents will be able to go and checkout directly on websites soon. I myself am building an open-source agentic commerce chat that uses MCP-UI under the hood and brings the full checkout experience to the chat.
Also, the entire premise of selling specific agents seems a bit fishy to me, my first thought is apps on the iPhone that got banned when Apple introduced an official app covering the same functionality in a new iOS.
I'm not sure I would risk building an agentic app that competes with this on Anthropic models out of fear of that scenario.
Also, we bring in the PayPal and Stripe Elements directly into the chat so you can enter your credit card info. Just like entering it on that merchants website.
What is even the legal basis here? Can the agent sign contracts on your behalf? What if it is tricked by the vendor into signing for a monthly subscription you never wanted?
I'm not sure I would risk building an agentic app that competes with this on Anthropic models out of fear of that scenario.
I feel like this is a messed up thing to brag about. From dark patterns now to whatever this is. Like a dark nudge from something that can appear intelligent, acting as a person, with the goal of selling more and more.
From a consumer perspective, I'm going to avoid coming anywhere near something like this. They've got data harvested from all of humankind and are weaponizing it as this discrete marketing tactic, convincing me to spend 35% more money on things I didn't actually need.
Such a bleak outlook. "Robot tells me to buy so I buy"...
They know they need to differentiate from other offerings that are catching up, but I don't know anyone who has just left their work up to any of these specialised products. Most try it say "it's interesting" and then move on.
This is akin to a demo / SDK ready to plug into a store, it’s just making it easy for devs to get started. it’s still using core models and APIs, no special magic models.
You posited two extremes, with no evidence for either, and with no other possibilities in between, or even outside the spectrum. That's beyond mere conjecture: it's just nonsense. I hope you realize that.
Too many people, hence all the friction and cost.
It's pretty clear that there is massive unmet demand for AI, and compute cannot keep up. So you can pretty reliably dial in the "free compute for inference" situation for any given lab based on their pricing and friction tactics.
Anthropic/Google getting stuffed (Google is an odd case here)
OAI doing alrightish
Grok/Meta basically doing giveaways
Chinese models trying to fill vacancies on 3rd party compute
Having said that Terra is awesome for clearly defined (by Opus/Fable) research task, and Sol is amazing for adversarial review, but I won’t allow them to touch my codebase any more.
If someone has a workflow where they keep tight control over agents, reviews the code by hand and so on, I can imagine them being useful.
For example, this exact narrative you're framing was said back during Opus 4.6/4.7 and fast forward to today we have open weight models that perform at Opus 4.6 level.
If you're using Claude in a pair-programming/"exoskeleton" sense, you'll probably never get close to maxing your subscription allotment unless you really lean on Fable.
The "you can't get anything done" comment confounds me, frankly.
To share more on my comment, that was when contrasting it with the competition and their credit allowances.
If we go by your scenario, then you could get even more with just a $20 plan from some of the others.
likely they hooked many fat enterprise customers
I'd happily a large premium for this, if any hungry startup people are looking for ideas.
I'm not interested in using agents controlled by vendors. Our interests are adverse! Even simple search on modern storefronts is an anti-consumer feature that promotes (upranks) defective products and things I wasn't interested in, and hides what I actually need, to maximize my eyeball time in their store. I don't want that. I would like to view as few commercial offers as possible. I would like to waste as little time as possible in the meaningless-to-my-life task of navigating stores.
Dropshipping as a "make $10k/week and be your own boss" idea is back?
You know, maybe merchants now have to "justify to the bank why they are using Shopify instead of Anthropic" just like your software engineers have to "justify not using AI".
A new AI purity spiral is emerging that devours one by one.
this is just marketing for them, they will do these kinds of posts for every possible industry.
The LAST thing you want to be non-deterministic is a payment system, right? What is the possible use of AI in the payment process? I fill in fields and that gets whisked away to an API somewhere to verify. Why do I possibly need Claude involved?
Either they are deliberately leaving out a space to build or, which is more likely, they just want to get as many merchants accustomed to relying on Claude as the intelligence and will then export their data to their own chat interface.