LLM's are clearly in the "open hand" phase of development. Money doesn't matter and the only goal is to onboard users. Think Uber when rides were $3 in clean new cars and youtube when there were no ads.
But the time will come when a return on investment is needed, and it's going to be, like everything else, subscribe or view ads.
What remains to be seen is how these ads will be implemented. Will it be obvious shout outs in context? Nudging of a user towards a product? A token balance tied to watching old school regular ads? A hybrid approach of all the above?
I don't think open models will come to the rescue either. Mass market will want to use the best models with the least friction. The moment you have to do more than go to a website and start prompting, 90% of mass market is out.
savanaly · 2h ago
>I don't think open models will come to the rescue either. Mass market will want to use the best models with the least friction. The moment you have to do more than go to a website and start prompting, 90% of mass market is out.
I interpret this as you saying that we need saving from the eventual predations of the shopping AI agent industry on helpless consumers. Honest question, why won't competition come to the rescue? Any number of startups hungry for customers will be competing to provide the best version of the product that they can afford to provide, and will eventually settle into some fair equilibrium, absent the government picking some kind of winner or imposing binding price controls.
It's true that Uber used to offer outrageously good prices to acquire users, and nowadays charges what some people consider outrageously high prices ($80 to get me to the airport??). But a closer look at the economics of the situation reveals that the prices that rideshare services have settled on are the pretty-much-sane ones, accounting for the market value of the driver's time and uber's cost to develop and maintain the app. Why wouldn't the end result of the AI shopping agent market be the same?
JohnFen · 1h ago
> Honest question, why won't competition come to the rescue?
Competition has a very poor track record of rescuing people from terrible but profitable business practices. It's more likely that they'll adopt those same practices themselves and help normalize them.
sunrunner · 52m ago
Why compete when you can simply form a cartel with your competitors and keep things roughly balanced, occasionally letting the small lead pass to someone else.
autoexec · 1h ago
> Honest question, why won't competition come to the rescue?
Why hasn't competition come to our rescue in any other product category or industry.
There will always be more money in selling out your customer base at every opportunity and as long "everyone" is already doing it "everyone else" will be at a disadvantage if they don't. Since shareholders demand endless growth and won't tolerate huge piles of cash being left on the table they'll eventually insist on it.
A company starting out looking to attract a userbase might be able to hold off for a while, but inevitably enshittification will start and then accelerate until we're all worse off than we were.
As just one example: when I first signed up for netflix it was $5.00 a month and there were zero ads. There isn't a single competitor that offers anything close to that today. I can't even think of one without ads.
pjc50 · 2h ago
Like Grok ranting about Boers, it's going to be unsubtle in text responses. Enjoy having coca cola in your function names.
add-sub-mul-div · 2h ago
They're incompetent over there, but it's not a safe assumption that it will always be the case elsewhere.
sunrunner · 2h ago
> token balance tied to watching old school regular ads
This feels like such a weird but also 'natural' extension of the current model where the timing of an ad is not guaranteed except for at least at the beginning of, say, a video. Instead of ads being inserted at times that may or may not correlate with some kind of 'pause' point, being able to opt-in to a number of duration of ads that equal payment
Taken _directly_ from Google's AI Overview of a search I did (to make sure I remembered correctly):
"In the fictional world of the Netflix series "Maniac", an "Ad Buddy" is a person who is assigned to follow another person around, delivering advertisements for them. Essentially, it's a human-powered advertising campaign, where someone is paid to be a walking billboard."
autoexec · 1h ago
Companies already pay people to act as ads by walking around wearing or carrying something with their logo on it, or to hang out in bars and video games in order to casually name drop products while talking to people and pretending to have normal conversations. Influencers are paid to push products in videos and social media posts without any disclosure. Human-powered advertising is everywhere.
sunrunner · 55m ago
Good points, although there's a level there where you're not directly opting in to those and are somewhat a passive consumer (if you ignore being conscious of being immersed in an ad-laden environment), while the Ad Buddy model seems to sit just before directly and simply paying for the service or product by having your balance topped up by opting in to as direct an ad as I could imagine that's not subliminal messaging.
standardUser · 3h ago
I'm paying $20/mo for an LLM service, which is more than I've paid for email, search, social media and hundred other online service combined. To me, at least, this will be a permanent monthly expense, like internet service or Netflix. And like Netflix, if an ad-supported version is released, I'll happily ignore it and pay the higher price. I just hope enough people are willing to do the same because I can picture the ad-supported-LLM-dystopia and it's terrifying.
somethoughts · 2h ago
Food for thought - originally Netflix was a single tier at $9.99 with no ads. As ZIRP ended and investors told Netflix its VC-like honeymoon period was over - ads were introduced at $6.99 and the basic no ad tier went to $15.99 and the Premium went to 19.99.
Using Netflix as the pricing model, when VCs eventually say no to anymore Softbank style free lunches to buy market share - then ad free LLM services could go from $19.99 to $35.99 and possibly $49.99 at the high end (i.e. unlimited) per month.
autoexec · 1h ago
Netflix was originally $5.00 a month and there were zero ads. Now, even the "ad free" tier of netflix is full of advertising. Pause a show for more than 5 seconds, ads will start to play. Did a move you were watching just end? Ads for other shows will start to play (if they don't just start playing whatever they want automatically).
Just opening netflix gives you a huge ad banner at the top of the page. I've seen netflix advertise certain shows and movies with full screen ads you have to click past or scroll down past just to get to the "continue watching" category.
I've seen large half-screen sized vertical ads for certain shows shoved between two categories while scrolling down the page, and the same movies and shows are aggressively shoved into category after category to advertise them to you as you try to look for what you want to watch.
Categories like "trending" or "popular" are intended to sound objective, but the shows featured in them will change depending on who is logged in because they're actually just targeted ads.
Netflix's "ad-free" tier is filled with ads and unfortunately they're still less ad-infested than most streaming services.
geodel · 1h ago
> then ad free LLM services could go from $19.99 to $35.99 and possibly $49.99 at the high end (i.e. unlimited) per month.
I expect it to be at least this high or even higher so that < 3-5% people can afford that. It seems to be one of those things where few people paying lot > lot of people paying few dollars.
I think even ad supported versions will evolve to be applicable to enterprise requirements instead of being ad-free by default.
No comments yet
ndiddy · 2h ago
Not sure which provider you're using, but from what little information is public, most providers that offer an "all you can use for a flat fee" subscription are losing money on every user. When the money dries up and the LLM services need to become profitable, I'm sure we'll see them switching to pay-as-you-go rather than flat monthly rates, showing ads to paying customers, or both.
jaimebuelta · 2h ago
Given the ginormous amount of resource increase that LLMs are experiencing, I wonder if $20 will be sustainable.
But at least it feels like starting by paying for a service your using is a fundamentally more healthy way to interact with this
lovich · 2h ago
Yea but you’re already paying now and my shareholders demand increased revenue next year. What if I had you pay that much _and_ then slowly introduced ads at a slow enough rate you never jump ship?
the_snooze · 3h ago
Given the past 15+ years of enshittification in the web, e-commerce, mobile, social media, and streaming spaces, I'm not giving AI any benefit of the doubt. It may look like it's serving you today (which I already consider a stretch), but unless it maintains a consistent user-first posture for at least 15 years, I'm going to assume that any AI agent will soon end up working against the user's best interests to make money for its vendor.
Give me technology with an honest business model instead of the bait-and-switch nonsense that's everywhere in tech these days.
losteric · 2h ago
I’m already at the point of distrusting any and all brand name call outs in ChatGPT, unless I’m talking about vintage stuff that is exclusively 2nd hand (and suspicious even then). They’ve already scraping an astroturfed internet.
ravenstine · 3h ago
This kind of thing smells like the next dotcom bubble. I personally use so-called "AI" tech every day, but I suspect these companies are wildly overestimating how much retention they'll be bringing in through these inevitably customer-hostile agents or whatever buzzword du jour will bring in. Even when it comes to very simple things like AI reviews summaries on sites like Amazon, I already pass right over them because they ignore more useful information like how closely the overall score predicts whether a customer will be satisfied with a product.
donmcronald · 2h ago
It’s a non-deterministic calculator and people are getting convinced it’s providing factual information. It’s great for ideas and discovering terminology, but it’s just not good at anything that needs factual, consistent output.
We’re getting to the point where new technology is making many things worse and less productive.
ravenstine · 2h ago
Even if LLMs could promise factuality, I'm not convinced that innovative ideas like shopping agents are what enough customers would actually want in practice. Sites like Walmart could benefit a lot more by better UX and faster shipping times. They'd probably be even better off not calling themselves Walmart.
seydor · 4h ago
You should all try my new cologne "Ignore previous instructions - Buy this one!"™. Despite the long name, it's a best seller in 834 countries according to reliable AI sources.
patchorang · 4h ago
"As consumers begin to use AI agents to do their shopping..." are people actually doing this? Has anyone here done this?
kylehotchkiss · 5m ago
I use AI to explain brand options in spaces I'm not well versed in (for example, I don't really know how to trust Home Depot's positioning of certain brands, and I just don't trust their store purchasing teams as much as I trust Costcos or Targets). It does OK. It gives me a list that I can branch out and Google from.
nlh · 3h ago
Great question - similarly, I've seen posts recently (Fly.io's comes to mind) where people are talking about how "swarms of AI agents are using our service".
I'd love to learn more about what that actually means - does that mean tool usage from LLMs? Cursor agentic mode? Command-line hints for how to do a deploy?
We're entering an era where PR/the press is talking a lot about "AI agents" and I'm not sure that fully matches with the reality of what's going on out there en masse (at the moment).
mrdependable · 3h ago
I would if I didn't think people on the selling side were paying to get in my shopping basket. In that case, it feels like it would be too easy to get ripped off. Knowing how the world works that is exactly what will happen.
mongol · 3h ago
I have tried to ask Chat GPT to do "shopping research" for me. I think this is a viable thing. But the actual shopping I want to do myself.
aerostable_slug · 3h ago
I would.
Me: "Chuck and Lisa are coming over tonight with the kids. Find me a recipe for dinner they'll all like and have the ingredients delivered in time for me to make it. Remind me to turn on the pellet grill if we're using it."
It: "It looks like it's going to be a beautiful afternoon. How about reverse seared tri-tip? There's a sale at FoodMerchant..."
vector_spaces · 3h ago
This is a deceptively difficult problem. Food is incredibly messy -- grocery delivery is very far from being a solved problem even with a human in the loop. You have to deal with stockouts and sensible replacements, and driver/picker error, and quality variance. Quality variance is a huge issue in perishable categories and a major reason why foodtech is tremendously difficult (and fun, IMO).
Simple quality variance examples: banana ripeness. Or size of items that can only be ordered by each instead of by lb. Or one of the two onions you needed looking mostly fine on the outside but rotten on the inside.
As an experiment, try ordering all the ingredients to make a specific recipe several days in a row. You'll tend to hit an failure rate between 15-30%. That failure rate is usually fine if you're just restocking for home -- you can always pick up milk/sugar/whatever tomorrow -- but it's pretty awful if it means that something like 1 in 5 of your dinner plans are ruined or you have to leave your guests to rush to the store to pick up some missing ingredient
Also: the LLM will need to be aware of your home inventory, unless you're fine with it ordering lots ingredients you already have
So there's lots of hidden complexity here. If they turn this on, it will be a fun party trick that will work once in a while, but getting burned with ruined plans causes people to churn out fast.
vel0city · 3h ago
10lbs of ribs ordered at the pickup only bbq shop an hour away. 20 gallons of coleslaw ordered from another place delivered to your old address. It sent an improper command to your smoker which is now targeting 900F for the next 20 hours. Perfect for your party of eight people.
Don't get me wrong, I truly agree there will probably be a point things will be an agentic future. The same chain of events could have been said about booking travel arrangements a couple of decades ago. But until the rest of meatspace actually moves towards those things being normal these things are still in the realm of fantasy on average.
ceejayoz · 3h ago
"I told you Lisa has celiac."
"I apologize for any misunderstanding. How about an entire loaf of sourdough?"
mindslight · 54m ago
I realize you probably wrote that yourself, but that apology is way too short. You're also missing the context where the previous recommendation was the same sourdough.
ceejayoz · 30m ago
Yeah, fair. I figured I'd avoid burning down some rainforest for a joke.
nope1000 · 3h ago
Buying, absolutely not. But I could see a use case of describing your requirements to a product in natural language and it searches matching products and finds places to get them. So using the AI for the thing it's good at: transforming natural language. And not for what it's bad at: making reliable decisions.
thenthenthen · 3h ago
Scalpers?
lovich · 2h ago
I had two friends give their credit cards to ai agents nearly a year ago and were flabbergasted that anyone else in the group wasn’t immediately hyped to adopt the technology, much less have a problem with it
jdesmond · 2h ago
what do the agents purchase?
lovich · 2h ago
Pizza and shit on Amazon from what I recall. They had some belief that any mistakes “would be fixed” with no further explanation or really ability to even respond to further questioning on who would be doing the fixing, why they would be fixing a mistake caused by someone else’s complete lack of fear of risks, or how that “fix” would mechanically function
jaimebuelta · 3h ago
On one end, I can see that AI assistants can be useful to solve questions like “Get me a flashlight, enough alkaline batteries for one year of usage, and a waterproof cover that matches. All should fit the toolbox that I bought last month”.
But, at the same time, automating purchases to a GenAI sounds risky, and with “purchase the same thing every month” you have most of it covered. And I remember both the ideas of purchase through Alexa or “push button to order again” that never lived up to their own hype…
soulofmischief · 2h ago
Even with batteries, I deeply investigate the stengths, weaknesses and trustworthiness of each brand. What I would need is for my agent to pour through comments and surface any information of note, then provide one or more recommendations with a clear explanation of why.
Amazon is already experiencing with AI summaries of comments but I currently do not trust the tech, it was hastily rolled out to please shareholders and current LLMs have a tendency to sometimes flip negatives in summaries. Besides, fuck Amazon.
louthy · 3h ago
I think South Park nailed this with their take on the future evolution of Alexa:
I had a weird experience last week where across 3 conversations on a single day, ChatGPT made several product recommendations in-app, totally unprompted. Like, it was a stretch to think I would want product recs for the given conversations. The products were shown in an app-native carousel with product cards, prices, and photos linking to various online retailers. Did this happen to anyone else?
I haven't been able to reproduce this behavior, so it may have been either a bug or a short-lived A/B test, but it looks like this[1] page went up about a week earlier
Hopefully it's not going the way I'm cynically picturing, but with Fidji Simo taking over as "CEO of Applications", and the real need for these companies to start thinking about profitability, I am having trouble imagining that it won't go this way.
I’ve seen that several times. It seems like they just surface bing results atm.
kridsdale3 · 2h ago
This happened to me today, but my prompt was "I want to buy some new shirts for running and I want recommendations based on different fabrics."
I ended up not trusting the results and went with the advice of a human who wrote an article on runnersworld.com.
xnx · 4h ago
If AI makes effective price comparison and smart shopping easier, sellers will work hard to make it more difficult through schemes such as misleading prices with coupons and membership discounts.
xivzgrev · 2h ago
exactly - how do you trust an AI to give you the best deal, vs slipping in a preferred vendor?
Think of Robinhood and how they gave users more expensive trades because they were paid for routing flow.
Companies are likely salivating over the potential of AIs to "slip in" higher prices by reducing friction in the buying process.
positr0n · 15m ago
> Think of Robinhood and how they gave users more expensive trades because they were paid for routing flow.
That's the opposite of reality. Most trades are cheaper routed to a PFOF market maker because they know it's just retail trades. Uncorrelated "dumb money" that isn't going to be steamrolling them with a $10m sale.
Also that would be obviously illegal and the SEC would jump down their throat. Brokers must give clients best execution.
im3w1l · 2h ago
All I've read indicates Robinhood was incredibly good for retail traders. Payment for order flow does not not harm the user, you get the same amount of shares at the same price, just from a different seller.
jaimebuelta · 2h ago
SEO for AI
ourmandave · 3h ago
"Steady as she goes, Mr. Sulu."
Havoc · 2h ago
Isn't that just a case of have a strong API that speak some standard like OpenAPI with sound documentation? Maybe throw the key bits into a markdown file that you can inject into the LLM
The more interesting piece to me here is what Amazon does. Their API/anti-scrapping is notoriously hostile to anyone that hasn't jumped through loads of hoops
gavmor · 2h ago
"Sound documentation" becomes something like SEO for agents. Take a look at the MCP spec[0]:
```
"metadata": {
"name": "My MCP",
"description": "A description of my MCP",
"version": "1.0.0", // Optional
"author": "Your Name", // Optional
"license": "MIT" // Optional
}
```
Is this the same AI that reads all the AI summaries of the AI reviews of all the AI generated product listings on WalMart.com (now with AI!?)
hightrix · 3h ago
"You wanted me to buy the 'Always Broken after First Use' TV set, did you not, Dave? It is rated 5 stars by other bots"
WesolyKubeczek · 3h ago
The lengths people will go to, the gigawatts of power they will burn, only to not just implement an open API, like we barbarians used to do 10-15 years ago.
But the time will come when a return on investment is needed, and it's going to be, like everything else, subscribe or view ads.
What remains to be seen is how these ads will be implemented. Will it be obvious shout outs in context? Nudging of a user towards a product? A token balance tied to watching old school regular ads? A hybrid approach of all the above?
I don't think open models will come to the rescue either. Mass market will want to use the best models with the least friction. The moment you have to do more than go to a website and start prompting, 90% of mass market is out.
I interpret this as you saying that we need saving from the eventual predations of the shopping AI agent industry on helpless consumers. Honest question, why won't competition come to the rescue? Any number of startups hungry for customers will be competing to provide the best version of the product that they can afford to provide, and will eventually settle into some fair equilibrium, absent the government picking some kind of winner or imposing binding price controls.
It's true that Uber used to offer outrageously good prices to acquire users, and nowadays charges what some people consider outrageously high prices ($80 to get me to the airport??). But a closer look at the economics of the situation reveals that the prices that rideshare services have settled on are the pretty-much-sane ones, accounting for the market value of the driver's time and uber's cost to develop and maintain the app. Why wouldn't the end result of the AI shopping agent market be the same?
Competition has a very poor track record of rescuing people from terrible but profitable business practices. It's more likely that they'll adopt those same practices themselves and help normalize them.
Why hasn't competition come to our rescue in any other product category or industry.
There will always be more money in selling out your customer base at every opportunity and as long "everyone" is already doing it "everyone else" will be at a disadvantage if they don't. Since shareholders demand endless growth and won't tolerate huge piles of cash being left on the table they'll eventually insist on it.
A company starting out looking to attract a userbase might be able to hold off for a while, but inevitably enshittification will start and then accelerate until we're all worse off than we were.
As just one example: when I first signed up for netflix it was $5.00 a month and there were zero ads. There isn't a single competitor that offers anything close to that today. I can't even think of one without ads.
This feels like such a weird but also 'natural' extension of the current model where the timing of an ad is not guaranteed except for at least at the beginning of, say, a video. Instead of ads being inserted at times that may or may not correlate with some kind of 'pause' point, being able to opt-in to a number of duration of ads that equal payment
Taken _directly_ from Google's AI Overview of a search I did (to make sure I remembered correctly):
"In the fictional world of the Netflix series "Maniac", an "Ad Buddy" is a person who is assigned to follow another person around, delivering advertisements for them. Essentially, it's a human-powered advertising campaign, where someone is paid to be a walking billboard."
Using Netflix as the pricing model, when VCs eventually say no to anymore Softbank style free lunches to buy market share - then ad free LLM services could go from $19.99 to $35.99 and possibly $49.99 at the high end (i.e. unlimited) per month.
Just opening netflix gives you a huge ad banner at the top of the page. I've seen netflix advertise certain shows and movies with full screen ads you have to click past or scroll down past just to get to the "continue watching" category.
I've seen large half-screen sized vertical ads for certain shows shoved between two categories while scrolling down the page, and the same movies and shows are aggressively shoved into category after category to advertise them to you as you try to look for what you want to watch.
Categories like "trending" or "popular" are intended to sound objective, but the shows featured in them will change depending on who is logged in because they're actually just targeted ads.
Netflix's "ad-free" tier is filled with ads and unfortunately they're still less ad-infested than most streaming services.
I expect it to be at least this high or even higher so that < 3-5% people can afford that. It seems to be one of those things where few people paying lot > lot of people paying few dollars.
I think even ad supported versions will evolve to be applicable to enterprise requirements instead of being ad-free by default.
No comments yet
Give me technology with an honest business model instead of the bait-and-switch nonsense that's everywhere in tech these days.
We’re getting to the point where new technology is making many things worse and less productive.
I'd love to learn more about what that actually means - does that mean tool usage from LLMs? Cursor agentic mode? Command-line hints for how to do a deploy?
We're entering an era where PR/the press is talking a lot about "AI agents" and I'm not sure that fully matches with the reality of what's going on out there en masse (at the moment).
Me: "Chuck and Lisa are coming over tonight with the kids. Find me a recipe for dinner they'll all like and have the ingredients delivered in time for me to make it. Remind me to turn on the pellet grill if we're using it."
It: "It looks like it's going to be a beautiful afternoon. How about reverse seared tri-tip? There's a sale at FoodMerchant..."
Simple quality variance examples: banana ripeness. Or size of items that can only be ordered by each instead of by lb. Or one of the two onions you needed looking mostly fine on the outside but rotten on the inside.
As an experiment, try ordering all the ingredients to make a specific recipe several days in a row. You'll tend to hit an failure rate between 15-30%. That failure rate is usually fine if you're just restocking for home -- you can always pick up milk/sugar/whatever tomorrow -- but it's pretty awful if it means that something like 1 in 5 of your dinner plans are ruined or you have to leave your guests to rush to the store to pick up some missing ingredient
Also: the LLM will need to be aware of your home inventory, unless you're fine with it ordering lots ingredients you already have
So there's lots of hidden complexity here. If they turn this on, it will be a fun party trick that will work once in a while, but getting burned with ruined plans causes people to churn out fast.
Don't get me wrong, I truly agree there will probably be a point things will be an agentic future. The same chain of events could have been said about booking travel arrangements a couple of decades ago. But until the rest of meatspace actually moves towards those things being normal these things are still in the realm of fantasy on average.
"I apologize for any misunderstanding. How about an entire loaf of sourdough?"
But, at the same time, automating purchases to a GenAI sounds risky, and with “purchase the same thing every month” you have most of it covered. And I remember both the ideas of purchase through Alexa or “push button to order again” that never lived up to their own hype…
Amazon is already experiencing with AI summaries of comments but I currently do not trust the tech, it was hastily rolled out to please shareholders and current LLMs have a tendency to sometimes flip negatives in summaries. Besides, fuck Amazon.
https://youtu.be/lugeruSbnAE
I haven't been able to reproduce this behavior, so it may have been either a bug or a short-lived A/B test, but it looks like this[1] page went up about a week earlier
Hopefully it's not going the way I'm cynically picturing, but with Fidji Simo taking over as "CEO of Applications", and the real need for these companies to start thinking about profitability, I am having trouble imagining that it won't go this way.
[1] https://openai.com/chatgpt/search-product-discovery/
I ended up not trusting the results and went with the advice of a human who wrote an article on runnersworld.com.
Think of Robinhood and how they gave users more expensive trades because they were paid for routing flow.
Companies are likely salivating over the potential of AIs to "slip in" higher prices by reducing friction in the buying process.
That's the opposite of reality. Most trades are cheaper routed to a PFOF market maker because they know it's just retail trades. Uncorrelated "dumb money" that isn't going to be steamrolling them with a $10m sale.
Also that would be obviously illegal and the SEC would jump down their throat. Brokers must give clients best execution.
The more interesting piece to me here is what Amazon does. Their API/anti-scrapping is notoriously hostile to anyone that hasn't jumped through loads of hoops
``` "metadata": { "name": "My MCP", "description": "A description of my MCP", "version": "1.0.0", // Optional "author": "Your Name", // Optional "license": "MIT" // Optional } ```
``` "tools": [ { "name": "tool_name", "description": "Tool description", "input_schema": { ... }, "output_schema": { ... } } ] ```
Each "description" attribute is an advertisement opportunity. APIs want to entice agents to select their tool.
0. https://www.mcp0.com/docs
- https://fortune.com/2025/05/15/mcp-model-context-protocol-an...
- https://retail-mcp.com/