I am tortured in new and exciting ways by the latest developments in technology. Today, as a part of Google I/O, Google announced the "Universal Cart" and another push towards "agentic shopping", this time with backing from a large variety of retail partners. In this piece, I'll explore how shopping is another front of the culture war, AI continues to gild shit, and these developments push us towards new and awful futures.
I'm sure anyone reading this is already familiar, but for future readers confounded at what shopping in "the old days" must have been like, I'll attempt to summarize the landscape of shopping online as it existed in the present-past. The usual shopping experience looked like this:
Online-forward retailers often now have add-on details, products offered at a discount post-purchase to be added to your shipment before it's prepared. Medium to large stores may also offer a paid subscription, which offer discounted or free products, a la Amazon Prime (A suite of Amazon products and discounted shipping) or Savage X Fenty Membership (significant discounts on clothes and discounted shipping). Mystery bags/hauls/boxes are very big at this time as well, offering a random selection of a range of products at a discount, often to get rid of hard-to-move goods.
There are a lot of ways a consumer may see advertising on the way towards a purchase. As previously listed, there are things like sponsored search results and products, advertised similar products, and products recommended based on their consumer profile of the shopper. Additionally, there are things like brand integrations into content, sponsored media, product reviews, all of which surface the products via other media platforms to hopefully drive sales. Alongside these posts, a consumer might see an affiliate link or code, which tracks sales related to a particular campaign and may drive kickback to the affiliate. Finally, at the most abstract, we have things like SEO optimization to improve product ranking in search results (which can take the form of things like keyword stuffing, blog backlinking, or synthetic reviews and social media posts) and dark patterns which attempt to influence the consumer towards certain choices, implied bargains, and purchases they otherwise wouldn't have made. Dark patterns are a whole thing unto themself, for a cursory guide I recommend Deceptive Patterns.
All this has formed a consumer-hostile experience, yet somehow the current shopping experience is degrading as I type this. There's been a flood of cheap, lookalike products featuring shiny AI-generated promo shots with limited to no customer support onto all markets, including Amazon and Walmart. Fake retailers pop up overnight and disappear faster, scamming by either selling products that do not match their description or refusing to sell products at all. AI generation among other things makes it trivial to stand up a convincing enough looking marketplace to get a few sales. Finally, AI influencers and artificial accounts generate fake prestige and social proof for the utility of these products and are surfaced naturally like any other content onto social feeds. Social media sites like Reddit have been considered a stop gap for real reviews and personal experiences with products and recommendations, however they've been known to be blanketed in artificial accounts and paid-for social marketing. Financial struggles and aggressive marketing are pushing people to new retailers like Temu which are filled to the brim with fakes and dark patterns and exploitative design. Traditional retailers are implementing these dark patterns to keep up and appear trendy. People are less concerned with how a product works and more that it looks like it works. Welcome to the Slop Economy.
Shopping For The New Shopping Experience
"Agentic shopping" (meaning shopping driven through and by AI agents) is sold as an alternative to this shitshow of an experience online. In Google's vision of the future, which is by no means the only but likely to be the most popular given Google's traffic and its retail partners, the consumer interacts with their AI agent and a "universal cart" to do their online shopping. AI agents have been shopping in some cases for a minute now, but this attempts to solve two problems. First, retail partners were unhappy with these agents shopping as they undercut the advertising market that's in place currently and things like bundles, discounts, and add-ons that might encourage return shoppers. On the consumer side, these agents were either shopping from a select few retailers, greatly limiting what could be bought, or just being let loose on the web, which is a sure way to end up with something you didn't want and leads to significant uncertainty.
As outlined in their blog post, Google intends for this new Universal Cart to be the premier way to shop when you're on Google, the most visited site in the world. What they decide invariably moves how we see the web, and in this case, shopping. To get the obvious out of the way: This is a big play on Google's part, attempting to dominate the next wave of E-Commerce. They are trying to set a standard for how shopping is, and by extension, ensuring a large amount of online shopping not only goes through Google (as much already traffic does), but through their Universal Cart specifically. That amount of customer data, and it being shopping data especially, puts them in a position to print even more money than they already do (net FY25 profit of $130B). This by itself is nothing new; major companies try to set themselves up to be the next big thing all the time, this is just one of the biggest companies trying to ensure they crush this next generation of competitors.
The focus for the product is around its AI integration, being built on top of their Gemini model, but it's important to note that this has tendrils that creep into every Google product, including Gmail and YouTube. Their idea is this:
- The user sees a product in a Google Service (mentioned in a YouTube video, ad in their Gmail inbox) or searches for one (the products being surfaced in the Search or in Gemini)
- The product(s) get added to their Universal Cart to which the user can add products from other participating retailers (which will be any that surface in Google Shopping, as far as I understand)
- Their AI model will analyze what is in the cart to "anticipate... needs and help solve problems", which includes things like (supposedly) flagging incompatible PC parts or noting if a product is available cheaper elsewhere
- At checkout, the carts are processed through Google Pay, so any linked cards and addresses are available (I will note that there is a "Checkout on Target button" below an Or in one of the screenshots, I'm unsure why someone would choose that at this point)
- Assumedly you get the normal confirmation to your email from the retailer, and it's linked and tracked through Google Shopping/Gemini
Keen-eyed readers might pick up that that process is not significantly different from current shopping. In fact, outside of being able to see multiple retailer carts at once, I don't think there's really any difference at all from a standard online shopping experience. Most computer part retailers will flag incompatible parts. Most major brands will flag if you don't have a loyalty card or credit card attached for bonus points (or whatever). The real difference here is that this experience is built for AI first, or as they like to say "for the agentic web".
The Agentic Web Is Different Just For The Hell Of It
Here's a video demo of the AP2 in action, and by extension, what the purchasing process looks like for an AI-first purchase.
Ignoring the bits that are only visible because this is a demo of it for developers of payment systems, this process seems rather slow. From first prompt to process complete, I am counting no less than 9 steps, and that's assuming that nothing needs to be customized, updated, or negotiated further. It's also assuming the user take the program's suggestions with no questions about reviews, fit, and anything else they might be curious about. Other demos I've seen don't seem demonstrably different from this, just in those cases they use more pictures and there's more back and forth.
Then it hit me: this process isn't meant to be faster, more consistent, or even easier, it's just meant to be different. The "Agentic Web" doesn't enable a web that is better for people, it just makes a web that's better for those that control the agents. It's for larping having a personal assistant, and for that privilege users are expected to give up a significant amount of control and understanding of what's going on behind the scenes.
An Example To Drive It Home
Let's say I'm shopping for a new shoulder bag; I'm simply not vibing with it and I want something new. Traditionally, I'm searching for shoulder bags, checking out reliable retailers, checking Reddit for reviews, and scrolling through images to get inspiration. I track down a manufacturer I like (shout out Herschel, not sponsored I just like their stuff) and search through their product lineup. When I find a model I like, I can plug that name into a search engine and see where it's available if I feel like searching for deals. If a customer already has a favorite brand, maybe they skip straight to the retailer's site and browse. Throughout this process I am attempting to make an informed decision on what I buy.
Through an Agent however, there's a lot of assumptions made and I either have to a.) trust the agent or b.) ask it to "explain" itself each step of the way.¹ When I tell the program I want a new shoulder bag, maybe it pops up a bunch of options from random retailers across the web. I select one that looks nice, but not quite right, and ask for more like it. It filters down to what it determines to be similar looking bags, although it can't see anything. When I pick one, shows the name of the product and the manufacturer, but notably no reviews (at least from what I've seen in demos²). If I'm happy enough with it, I just assume that the product is of reasonable quality and that it in fact does match my specifications, as AI are known to fabricate information arbitrarily. I'm not nearly as engaged as I would be when compared to the traditional shopping experience and that's by design.
Keeping a customer distanced from any points of friction, including any information that might give them pause, alternatives, or even just the simple process of browsing, creates an environment where purchases are made before we can even think about it. If I ask someone to purchase something for me, I give it significantly less thought than I would doing it myself. I'm simply not involved with most of the decisions anymore. At each step of the process, the customer is asked to trust that the Agent is only offering them the best options it could find from the whole internet. That is not how it works by a long shot.
Pitfalls of Agentic Shopping
Agents are prone to all the same problems that are already present when searching for a product and more. There's no reason to believe that keyword stuffing and general SEO shenanigans won't still work. They can say they're mitigating it, but they already say that about current SEO practices. It's a cat and mouse game where actors try to find exploits to move them to the top and the searches try to calibrate around that, and the cycle repeats forever.
There's an added vulnerability of prompt injection as gestured at earlier, but I need to drive one thing home. Any claim of being resistant to prompt injection is anecdotal at best, as these programs are inherently non-deterministic. That means that performing the same action does not mean you will get the same result, which is the opposite of what we expect in traditional computing. Some methods of exploiting AI even just include asking the same question repeatedly, or saying it more forcefully, adding urgency to the request.
On top of that, since the Agent is surfacing specific options, it is being guided by an unseen hand ("the algorithm"), which may be influenced by partner retailers or other (nefarious) actors. Let's also recall that AI systems have routinely been bypassed and exploited in increasingly silly ways.³ It will tend towards things that are eye-catching and flattering, as that's what gets clicks and eyes, even if the product is noted to not be as described. This is the same sort of problem that current content algorithms have as well. The program can only prioritize some semblance of aesthetics because it can't determine quality or consistency. These agents are ripe for manipulation and pushing shit you don't actually want. There is no shame or discomfort from stopping these programs from trying to upsell you again and again and again like there would be with a human sales rep. They will make that sale if possible.
The real points of frustration with online shopping nowadays are around trust and sincerity. Can I trust this random Shopify store to ship my item as described, or at all? Does this TikTok shop have the quality and safety mechanisms in place that I should be buying supplements from them? (No, please don't do this.) Is this review genuine that says it worked great for them, or was it a paid review where the product was received for free? Do they even exist? AI (and Agents) don't fix any of the problems already present while shopping, nor does the universal cart- it instead obfuscates it. These systems make it less clear who you're going to and what their reputation is, so long as it's good enough to fool the AI it must be good enough for the consumer.
Agentic shopping attempts to squeeze even more purchases out of the average internet shopper. If you as a user feel catered to, you're more likely to engage in shopping habits, including shopping addictions which are already very prevalent. An agent allows the user to be guided along a very clear path, but made to feel like a friend is just helping them pick out a fun new outfit or treat. Computers are not our friends and should not act like it. If you feel like you need that in your life, you need real-life friends.
The Temu Of It All, Or: The Leftist Screed
Temu has been an unavoidable force for a hot minute now. Its tagline is "Shop like a billionaire", and I think that strikes at the heart of what's at play here. Americans are, unfortunately, a large driver of AI and global culture at large. In a quote typically attributed to John Steinbeck that is instead actually a paraphrasing of Steinbeck by Robert Wright, he says "socialism never took root in America because the poor see themselves not as an exploited proletariat but as temporarily embarrassed millionaires."⁴
We Americans, largely, like to play ourselves as the downtrodden hero. We are simply down on our luck Richie Riches who haven't grinded our way to success yet. We "shop like a billionaire" because we're going to save our way to success! Give me deals galore, every mechanic you can think of to ring those dopamine receptors, I want it all! Don't ask why I need all this cheap tat to be rich, that's just how it works! The rich have lots of things you see, so to be rich, I too should have lots of things.
AI presents the same sort of lifestyle paradigm. Do I need to have answers custom tailored to me and my sensibilities to be able to surf the internet? Certainly not, but AI offers a solution which can tailor the internet to my liking. It can also hold my hand and gently guide me to products and experiences I like. Hell, I can have e-sex with it if I want. A mommy/assistant/personal shopper/prostitute all in one.
The rich have assistants and people waiting on them hand and foot, why should I, a temporarily embarrassed millionaire, not have one too? Is this technology not democratizing assistants? What I'm doing is good actually, if you ignore all the parts about how we're reshaping the economy around it for no reason, aggressively increasing our electricity usage, and our dramatically upping our need for rare earth minerals. AI is a force for good; just don't look at the scams, vulnerabilities, and exploitation taking place at an unprecedented rate!
AI does little but allow for more, and that's not a good thing. More is not needed with where we're at. We have plenty of food but it gets wasted due to how we organize our society. Americans have more debt than ever, despite being more productive than ever. Curious; I'm sure that's not correlated with more wealth than ever being concentrated in the hands of the ultra-wealthy.
More is not a virtue, and should not be treated like one. We are finite beings with limits and that's okay. We should improve our lives, those around us instead of just searching for more. This is the existential threat AI poses.
In Summation
So, does the Universal Cart solve any of the problems that customers have faced previously with shopping on the web? No, AI provides an opaque wrapper for old problems and introduces many new ones. This new world envisioned by Google & Co prioritizes just one thing: reducing friction in online shopping. Now, anyone who has shopped online in the past decade might be saying "Things are already pretty seamless though, it takes me maybe three clicks to order something through most stores." They'd be right! I'd argue things are actually too seamless already; it's much too easy to spend hundreds or thousands of dollars in just a few clicks, address and payment information pre-filled by Link or Shop, with shipping sometimes measured in hours instead of days. Our spending is as easy as ever, while making that money is becoming increasingly fraught and dependent on at will contracts or contractor work.
Footnotes
¹ I will keep yelling it each time this comes up, but these programs don't "understand" anything, nor can they "explain" themselves. What it is is a complex relational set of probabilistic weights which resemble reasoning when you assign characters to the tokens it processes. Anyone who tells you it is reasoning or anything like that is lying and trying to whitewash what is going on to make it seem more trustworthy by associating it with something we as humans already understand. [Return to Article]
² It's also interesting to note that Google has you "pick out a size" for your avatar, which I will note that women's sizing is extremely inconsistent for those that aren't aware. No clue on if/how it handles that, if you need to be aware ahead of time that a brand runs small or large, etc. More guesswork that you just have to assume is being done by the agent. [Return to Article]
³ I was going to have separate links to different bypasses that have happened but it feels impossible to find news on that stuff because of the deluge of SEO spam from AI companies trying to sell you a solution, so here's a paper on prompt injection instead that features many of the major way Agents are undermined and subverted. Major ways I can remember seeing these models be exploited include "TyPInG LiKE ThiS", typing in poetic prose, and just saying you need it done now or someone will die. [Return to Article]
⁴ Wright, Ronald. A Short History of Progress. Toronto, Anansi, 2004, p. 124, Archive.org. Accessed 20 May 2026. [Return to Article]