One of the things that I would like to try this with is testing api/MCP data quality without having to go and signup for multiple services, add config, trial .. etc.
I’m referring to services like RapidAPI where it’s quite manual and painful to go to each service, subscribe (even for free) make calls, check the data.
I want to instruct ChatGPT for something like, from (x) service, find all providers for (y) data, spend max of ($z) for the following use cases and present me with a table based on some criteria
>Over time, we expect users to be able to delegate purchases entirely to their agent by setting clear parameters—such as spending limits, purchase categories, or time periods—and pre-approve those purchases with enhanced safeguards that build on today’s protections, including biometric checks and clear records of authorization.
While I understand many people are wealthy, I can't imagine being so insensitive to price that I would delegate the purchase entirely to an agent.
It depends on what you are buying. I can see many people using this to try to beat bots when buying gig tickets for very popular artists or for items in high scarcity.
Oh my yes. This feels like one of those classic, "you have to live in a Silicon Valley-like alternate reality bubble to think this is anywhere within the galaxy of normal" things.
I don’t think this is about wealth at all. When I’m looking to buy something, I usually have a few criteria in mind, like price, warranty, and quality. It’d be nice to have a bot check for options that meet those criteria so I don’t have to check manually every few hours. It's not a new idea either, just another N+1 attempt at it.
That's separate from automatic buying though, which seems to be the main point here.
This reminds me of the whole house automation proposition, where the problem with it is that even if you remove 100% of the friction in my life from turning on and off light switches, you really haven't removed that much from my life. Replacements don't get very much "budget" before they're more hassle than they are worth and it doesn't take very many technical snafus where I can't actually turn my lights on because of the automation for the whole thing to go net negative.
The actual approval to spend the money is not the hard part. I'm not doing it hundreds of times per day. The research of what to spend it on is the hard part. The agents already do that well enough. Even if you save me 100% of the effort of actually buying a product you're not really saving me that much, and most importantly, it doesn't take much for the whole proposition to go negative because there wasn't much positive in the first place. One product I didn't really want and have to physically return negates in one fell swoop quite a few "oh thank goodness I didn't have to approve this purchase". One misinterpretation that causes the agent to spend 10x what I really wanted to spend negates months of benefits.
This is one of those features where it's something OpenAI wants rather than something most people are going to want. Yes, of course OpenAI wants direct unfettered access to your credit card. Most businesses would love that. But just because they can wave the words "AI Agent!" in your face doesn't mean you should actually give it to them.
just did this last night with Grok Bot. I said buy it if it's less than $30, and it came back as $31 with tax so i had to give it another approval to actually purchase the item
Why can’t you tell the agent to be price sensitive and only buy products you have approved (or actually would have)? Perhaps there is an issue of trust, but it seems the idea of the wallet is to enable that trust where a certain amount of purchases are expected to be made in error and refunded without a fuss.
We are already at a place where the agent can surely budget better than most people if given the right instruction to do that. I don’t see this tech as being obviously as for the rich.
This is happening with regular users too. People give FULL control of browser to agents which are logged into Amazon and they get (sometimes pleasant) surprise receipts for objects they have told their AI that they are considering.
I mean, I don't, you don't, that doesn't mean this cannot be a problem.
Yeah, and little has changed. I run very limited workloads on my personal computer that has all my details and data, because the sandboxing tools are all still botched.
It works in two ways: first it’s a psychological thing, it’s easier to approve a payment when the money has already left your account and second it makes budgeting easier. Like, it’s safer at preventing multiple purchases beyond what your account has
There have been so many attempts at this. From "Alexa, buy a blue jacket" to embedded shopping for items in movie scenes to smart fridges ordering food...
It turns out that people like to do some research, look at a lot of images, read user reviews and more before making a purchasing decision. The subset of people that treat most everyday shopping as a completely automatable commodity is much lower than the tech world thinks.
You may have a point - there is a kitchen tool that helps you drain curd to make thicker yoghurts (greek yoghurt or something). And I saw it once in an online shop and though I would buy it later. But I just couldn't find the same thing again, despite trying different keywords. Searching by describing it would indeed have been useful.
ive been testing this for a while, because I am a serial over-researcher when I buy stuff.
It is a great research companion, but it is perhaps the most important use case I've run across to feed multiple models from different providers the same/similar prompts. You'll get wildly different suggestions sometimes.
I think the critique is focused on this part of the announcement:
> Over time, we expect users to be able to delegate purchases entirely to their agent by setting clear parameters—such as spending limits, purchase categories, or time periods—and pre-approve those purchases with enhanced safeguards
Researching with AI is one thing, but the announcement is suggesting that eventually they want to be able to take the human out of that loop too.
I agree, but I see a use case if this is integrated directly into standard chats. Many people use ChatGPT to research products and would probably check out within ChatGPT if they had the chance.
i dont think its being treated as an automatable commodity. one of the most popular uses of chatgpt is to aggregate and compare item information. in fact it improves the researching and reviewing experience
This just makes it so you can actually buy the thing from chatgpt vs. going to a specific site and filling out your info
What about "ChatGPT, buy regular groceries for this week" ? There are a whole class of items I don't do much research for. I can have a thread where I keep updating this list based on the season and week on week is just done and maybe it gives me useful suggestions based on behaviour too.
I think many people do "research" their groceries somewhat, and it can be highly personal.
Does ChatGPT know that my kid wants grapes but my wife wants him to eat only organic grapes, but doesn't like grapes herself, so we should only buy a small quantity of organic grapes?
Or that as a household we try to avoid Nestle products altogether?
Or that we want tortillas, but only made of flour and the burrito size from one specific manufacturer?
Maybe after enough time and enough coaching it could get it right, but I also have never ordered groceries online. I don't even trust a random person to do a good job picking produce, so I suppose I'm far from the imagined target market here.
Anecdotal, I've used ChatGPT to do most of my research on buying things lately. For instance, I am learning how to sew outdoor gear -- backpacks, duffel bags, pouches, etc. I basically take ChatGPT's recommendations at face value when I'm asking about what kind of thread, zippers, webbing, etc. I should by based on what I'm trying to create. While I'm doing the research, I have at times wished I could just say, "Okay, buy that for me from one of these handful of retailers."
"I take chatGPTs recommendations at face value" just means "I beleive whoever is paying the most to get their products into chatGPT, at face value". If you dont think this is what's happening, you're a fool.
This is going exactly the same route as Apple or Android's "forced" 15/30% cuts... they will start free, and then, start to take commissions, or else, they will not "recommend" your product.
This is yet another attempt at squeezing the money out of sellers.. and it will mostly be smaller ones that will be impacted.
One silver-lining is this might give some negotiating power to sellers on Amazon.. who have been ripped-off by Amazon more and more every passing year.. until of course, they both don't end up colluding.
We are going to enter an economy in which the quality of anything sold is irrelevant, and the only thing that matters is outbidding everyone at sponsored content. Even brand recognition will be meaningless when everybody just buys whatever their AI agents recommend them.
I think about this sometimes in relation to Adam Smith. He talks a lot about division of labour, making pins and so forth, but not at all about the cost of marketing products. Yet in the economy today the cost of finding and convincing your customers is often the dominating cost - especially in software. Are we really efficiently solving the economic problem any more, or just playing pretend like we are?
Adam Smith couldn't have predicted marketing to be a dominating cost back in the 18th century. Goods were cumbersome to transport, only commodities or luxury items would be worth sailing around, and those don't really need marketing.
At worst, at best maybe it's an honest user's review on reddit. Or a content marketing company's bot's review. Maybe it's a hobbyist's personal site that was scraped. Or an industry shill's site.
Some care is in order with this approach, especially when ChatGPT uses web searches. Some companies are known to market their products disguised as user feedback in forums etc. so when ChatGPT just goes ahead and reads these posts (or even reads marketing material directly) it will definitely steer the output. There are also cases in which websites cater to these agents directly.
Not saying a user couldn't also be tricked by this, but of course it is more nuanced than the models sometimes make it out to be.
Any information you could find is SEO-sponsored slop, all the accompanying images are AI generated, all the reviews LLM-written by bot farms. Go on with the times and just talk to your AI agent to get the latest sponsored answers.
People are already doing all their shopping benchmarks on ChatGPT and Claude. A wallet for agents like Stripe Link or ChatGPT will remove friction, and humans love being lazy.
Just saying something has not worked in the past doesn't make it true for the next iteration too. Many touch-enabled phones failed before the iphone.
I believe it is the same with travel. Some tech bros will plan their honeymoon with AI, yes, but choosing your favorite airbnb is sometimes part of the fun of a trip.
How often do you need something you really need to research?
I think you've got it backwards. The vast majority of people who don't have enough time to begin with will gladly just keep typing their shopping cart into their AI of choice.
I wouldn't be surprised if this ends up being the majority of usage for suburban housewives, along with webmd type things and relationship questions.
I've found looking for available products to be one of ChatGPT's most finicky use cases, it consistently recommends products that are unavailable, out of stock, or have wildly different reported prices than what it claims.
And miss out on the dopamine hit of tapping the purchase button when I'm not 100% sure I should or not??
Jokes aside, I do wonder if this is really something someone wants to delegate but I could see it decreasing the pain of adding to cart, inputing customer information, and doing payment if the site doesn't have Apple Pay or PayPal.
Slightly off topic, maybe, but the recent advertisements showing the elderly instructing a clanker to read their email and interact with their bank are absolutely unhinged.
And watching OpenAI and Anthropic lament the risks of the frankenstein monster that they are literally building as they lament it is something out of a comedy. Maybe don't build that LLM-connected wetlab or wire your monster into financial systems?
I wonder what the downstream implications for this will be.
For example, if say a report is only available on one website (eg Bloomberg), will it make ChatGPT more likely to just pay, rather than find a hosted copy on some other website?
People have oft suggested adding a tiny fee for each email to avoid spam. Would this move make it more likely to have every blogspam website now have a "Get full access only for <less than the threshold limit of purchase> amount!!!" paywall? What about traditional newspapers and media houses?
This runs a mild risk of making the web even more enshittified.
> To accept payment from ChatGPT Wallet, websites need to ensure agents are not inadvertently blocked, and payment and risk systems need to recognize legitimate purchases made by agents.
IIUC this is so people don't have to give full trust to agents. People neck deep in agentic AI future is having issues with their AIs --dangerously-skip-confirmation-before-completing-purcheses-on-their-behalves. AIs are clicking Buy Now without asking. A separate means of purchases AIs are required to use to buy things puts limits to that.
I'm about as Pro-AI as one can reasonably be, but I find many of OpenAI's examples lately to be a bit off-putting. The way the model recommends sending the gift to the user so they can add a recipe to make the gift seem more personal feels bad, to me. It's almost deceptive in the sense that we're recognizing having an AI model do the thinking, shopping, and purchasing of a gift has eliminated all of the human-touch that gift giving/receiving is about but we'll do it anyway and devise a way to make it seem like we cared enough to think about the person.
This made me realize that AI providers can build all the products they want, but can't be more than shovel sellers, because AI still sucks at many of the things that make a product... a product!
There is a place for agentic payments, but I think the "dots will buy X on your behalf" needs any kind of real proof. In my experience, ChatGPT is not really good at identifying products to buy, because it does not weight the various marketplaces the same way I would and constantly offers product that are more expensive and have insane shipping overhead instead of "just getting it from Amazon".
I am sure some people here will think this is a boon, but for the average user it's really not. If I want a DeWalt circular saw, I just want it from the cheapest place that brings it to my door the fastest, not just the first link that happened to show up in ChatGPT websearch tool.
I agree with all of your points, but isn’t this basically an MVP (or close to it)?
I’m willing to bet that these types of services get optimized fairly quickly one way or another, either internally at OpenAI or externally via plugins of some sort.
Or said another way, if you gave your instructions above to your dot purchasing agent, how many weeks or months do you think it will take for OpenAI to make those instructions happen in a way that you would be satisfied with?
Note that I am not a fan of Sam Altman and I don’t trust OpenAI, but I struggle to see how your preferences will be a long-term hurdle.
Damn, they are going for Stripe and other PSPs now. I wonder if that's the right strategy.
Basically there's a whole load of companies working on that space, also for agentic payments.
Now, they basically shoved on everyone building that, that they want people to use their own solution (think Apple pay vs. thousands of payment providers).
Kinda silly, this will make first of all, those companies no longer see them as a partner.
Eventually every single business in earth will see OpenAI as a threat, then who will want to buy anything from it, or use it?
Gosh, I get the angle here, but I'm tired of the random OpenAI launches BEYOND models. It's Anthropic's IPO week and this is what you lead with?? Get your best and brightest models out the door, please...
> 1.2B people use ChatGPT each week, and many of them turn to it to research products, compare options, and decide what to buy...
> To accept payment from ChatGPT Wallet, websites need to ensure agents are not inadvertently blocked.
ChatGPT Wallet is a data gathering ploy. Sam Altman and co. want dense, up-to-date pricing signal, so they tell stores, "we will redirect traffic your way if you start giving us this data."
The stores already give customers this data, and should give customers' AI representatives this data. That does not extend to all agents. The stores should charge like $0.01/request, which is a rounding error for someone actually looking up a dozen products to buy, but prevents the DDoS of random agents scraping for training data or stupidity.
Won’t this enable people to trivially build things like shoe bots, restaurant reservation bots, etc? Is that something merchants are okay with?
That feels to me like the only time I would be okay pre-authorizing a bot to make a purchase on my behalf (which is where the announcement says we’re going). In other words, I’ve done this research and know I want this one thing that I can’t easily get myself. It also feels like exactly what merchants would block.
My wife has a small online shop, and she uses social media to push her products. She doesn't use marketplaces since her funnel will drop, as she'd need to compete against other brands for the customers she already has (5x more difficult to onboard a new customer than to retain an old one). There is no way small and medium-sized businesses will gain anything by using this "new" marketplace.
Oras | 8 hours ago
I’m referring to services like RapidAPI where it’s quite manual and painful to go to each service, subscribe (even for free) make calls, check the data.
I want to instruct ChatGPT for something like, from (x) service, find all providers for (y) data, spend max of ($z) for the following use cases and present me with a table based on some criteria
suddenlybananas | 8 hours ago
While I understand many people are wealthy, I can't imagine being so insensitive to price that I would delegate the purchase entirely to an agent.
netdevphoenix | 8 hours ago
goosejuice | 8 hours ago
simonw | 8 hours ago
suddenlybananas | 8 hours ago
folkrav | 8 hours ago
Waterluvian | 8 hours ago
Maken | 8 hours ago
azuanrb | 8 hours ago
jerf | 7 hours ago
This reminds me of the whole house automation proposition, where the problem with it is that even if you remove 100% of the friction in my life from turning on and off light switches, you really haven't removed that much from my life. Replacements don't get very much "budget" before they're more hassle than they are worth and it doesn't take very many technical snafus where I can't actually turn my lights on because of the automation for the whole thing to go net negative.
The actual approval to spend the money is not the hard part. I'm not doing it hundreds of times per day. The research of what to spend it on is the hard part. The agents already do that well enough. Even if you save me 100% of the effort of actually buying a product you're not really saving me that much, and most importantly, it doesn't take much for the whole proposition to go negative because there wasn't much positive in the first place. One product I didn't really want and have to physically return negates in one fell swoop quite a few "oh thank goodness I didn't have to approve this purchase". One misinterpretation that causes the agent to spend 10x what I really wanted to spend negates months of benefits.
This is one of those features where it's something OpenAI wants rather than something most people are going to want. Yes, of course OpenAI wants direct unfettered access to your credit card. Most businesses would love that. But just because they can wave the words "AI Agent!" in your face doesn't mean you should actually give it to them.
johnsimer | 8 hours ago
swid | 8 hours ago
We are already at a place where the agent can surely budget better than most people if given the right instruction to do that. I don’t see this tech as being obviously as for the rich.
numpad0 | 7 hours ago
I mean, I don't, you don't, that doesn't mean this cannot be a problem.
sethops1 | 8 hours ago
CodeCompost | 8 hours ago
freakynit | 8 hours ago
steinvakt2 | 8 hours ago
xpct | 7 hours ago
steinvakt2 | 5 hours ago
algoth1 | 8 hours ago
sebzim4500 | 8 hours ago
algoth1 | 5 hours ago
dofm | 8 hours ago
ajs1998 | 8 hours ago
max_sendfeld | 8 hours ago
It turns out that people like to do some research, look at a lot of images, read user reviews and more before making a purchasing decision. The subset of people that treat most everyday shopping as a completely automatable commodity is much lower than the tech world thinks.
esafak | 8 hours ago
eckelhesten | 8 hours ago
esafak | 8 hours ago
Walmart has one too: https://corporate.walmart.com/news/2026/10/sparky-shopping-a...
whimsicalism | 8 hours ago
thisislife2 | 7 hours ago
jdc0589 | 8 hours ago
It is a great research companion, but it is perhaps the most important use case I've run across to feed multiple models from different providers the same/similar prompts. You'll get wildly different suggestions sometimes.
civicsquid | 7 hours ago
> Over time, we expect users to be able to delegate purchases entirely to their agent by setting clear parameters—such as spending limits, purchase categories, or time periods—and pre-approve those purchases with enhanced safeguards
Researching with AI is one thing, but the announcement is suggesting that eventually they want to be able to take the human out of that loop too.
danielfoster | 8 hours ago
skybrian | 8 hours ago
vanuatu | 8 hours ago
This just makes it so you can actually buy the thing from chatgpt vs. going to a specific site and filling out your info
LoganDark | 8 hours ago
vanuatu | 7 hours ago
automatable commodity is referring to buying stuff
tirumaraiselvan | 8 hours ago
Carrok | 7 hours ago
Does ChatGPT know that my kid wants grapes but my wife wants him to eat only organic grapes, but doesn't like grapes herself, so we should only buy a small quantity of organic grapes?
Or that as a household we try to avoid Nestle products altogether?
Or that we want tortillas, but only made of flour and the burrito size from one specific manufacturer?
Maybe after enough time and enough coaching it could get it right, but I also have never ordered groceries online. I don't even trust a random person to do a good job picking produce, so I suppose I'm far from the imagined target market here.
Maken | 7 hours ago
prescriptivist | 8 hours ago
bad_haircut72 | 7 hours ago
freakynit | 7 hours ago
This is yet another attempt at squeezing the money out of sellers.. and it will mostly be smaller ones that will be impacted.
One silver-lining is this might give some negotiating power to sellers on Amazon.. who have been ripped-off by Amazon more and more every passing year.. until of course, they both don't end up colluding.
Maken | 7 hours ago
morkalork | 7 hours ago
bad_haircut72 | 7 hours ago
piva00 | 4 hours ago
morkalork | 7 hours ago
NikxDa | 7 hours ago
Not saying a user couldn't also be tricked by this, but of course it is more nuanced than the models sometimes make it out to be.
whimsicalism | 7 hours ago
Maken | 8 hours ago
Eldodi | 8 hours ago
Just saying something has not worked in the past doesn't make it true for the next iteration too. Many touch-enabled phones failed before the iphone.
rosstex | 8 hours ago
syshyk | 7 hours ago
dingaling911 | 6 hours ago
I think you've got it backwards. The vast majority of people who don't have enough time to begin with will gladly just keep typing their shopping cart into their AI of choice.
I wouldn't be surprised if this ends up being the majority of usage for suburban housewives, along with webmd type things and relationship questions.
jefb | 8 hours ago
pvab3 | 8 hours ago
hectdev | 8 hours ago
Jokes aside, I do wonder if this is really something someone wants to delegate but I could see it decreasing the pain of adding to cart, inputing customer information, and doing payment if the site doesn't have Apple Pay or PayPal.
breakwaterlabs | 8 hours ago
And watching OpenAI and Anthropic lament the risks of the frankenstein monster that they are literally building as they lament it is something out of a comedy. Maybe don't build that LLM-connected wetlab or wire your monster into financial systems?
user43928 | 8 hours ago
Stripe's Link integration for Grok Bot doesn't achieve the same?
user_7832 | 8 hours ago
For example, if say a report is only available on one website (eg Bloomberg), will it make ChatGPT more likely to just pay, rather than find a hosted copy on some other website?
People have oft suggested adding a tiny fee for each email to avoid spam. Would this move make it more likely to have every blogspam website now have a "Get full access only for <less than the threshold limit of purchase> amount!!!" paywall? What about traditional newspapers and media houses?
This runs a mild risk of making the web even more enshittified.
simonw | 8 hours ago
> To accept payment from ChatGPT Wallet, websites need to ensure agents are not inadvertently blocked, and payment and risk systems need to recognize legitimate purchases made by agents.
numpad0 | 8 hours ago
x187463 | 8 hours ago
hisfraudulency | 7 hours ago
xpct | 7 hours ago
egorfine | 7 hours ago
egorfine | 7 hours ago
netdur | 7 hours ago
belval | 7 hours ago
I am sure some people here will think this is a boon, but for the average user it's really not. If I want a DeWalt circular saw, I just want it from the cheapest place that brings it to my door the fastest, not just the first link that happened to show up in ChatGPT websearch tool.
csa | 3 hours ago
I’m willing to bet that these types of services get optimized fairly quickly one way or another, either internally at OpenAI or externally via plugins of some sort.
Or said another way, if you gave your instructions above to your dot purchasing agent, how many weeks or months do you think it will take for OpenAI to make those instructions happen in a way that you would be satisfied with?
Note that I am not a fan of Sam Altman and I don’t trust OpenAI, but I struggle to see how your preferences will be a long-term hurdle.
thiago_fm | 7 hours ago
Basically there's a whole load of companies working on that space, also for agentic payments.
Now, they basically shoved on everyone building that, that they want people to use their own solution (think Apple pay vs. thousands of payment providers).
Kinda silly, this will make first of all, those companies no longer see them as a partner.
Eventually every single business in earth will see OpenAI as a threat, then who will want to buy anything from it, or use it?
bluepeter | 7 hours ago
programjames | 7 hours ago
> To accept payment from ChatGPT Wallet, websites need to ensure agents are not inadvertently blocked.
ChatGPT Wallet is a data gathering ploy. Sam Altman and co. want dense, up-to-date pricing signal, so they tell stores, "we will redirect traffic your way if you start giving us this data."
The stores already give customers this data, and should give customers' AI representatives this data. That does not extend to all agents. The stores should charge like $0.01/request, which is a rounding error for someone actually looking up a dozen products to buy, but prevents the DDoS of random agents scraping for training data or stupidity.
civicsquid | 7 hours ago
That feels to me like the only time I would be okay pre-authorizing a bot to make a purchase on my behalf (which is where the announcement says we’re going). In other words, I’ve done this research and know I want this one thing that I can’t easily get myself. It also feels like exactly what merchants would block.
alexk307 | 7 hours ago
Machado117 | 5 hours ago
johndae | 3 hours ago