The big news in AI is that some people in AI think AI may destroy the world. And now the topic has broken through to regular people who couldn’t tell you what an LLM is.
But another big AI development is that “agentic” bots — bots that can act on their own, on your behalf — are starting to roll out to regular people, too. Instinct, a new service that’s generating a ton of interest from the tech set, promises it can do things like buy you groceries or book your flights. And Meta just debuted Muse, which makes the same pitch.
All of which means that the idea of “shopping” may radically change, and it could be at least as big a deal as the move from brick-and-mortar stores to the internet.
Then again: These are still niche ideas, and it’s entirely unclear where this goes. Will consumers trust a bot with their credit card? Do consumers want a bot to shop for them? And what about all the existing retail operations — especially Amazon — that expect human buyers to come to their sites and stores, where they can sell them other stuff? Can they slow this idea down or stop it altogether?
I asked Jason Del Rey, a veteran retail journalist who covers this stuff full-time at The Aisle, the newsletter he launched last year, to walk me through where things stand today, and where they might go soon. You can listen to our entire conversation on my Channels podcast; the following is an edited excerpt.
Peter Kafka: Bots that shop and do other transactions for me are something the AI guys have been talking about for years. Why aren’t they here yet?
Jason Del Rey: Last year, OpenAI and ChatGPT launched something called Instant Checkout — basically the ability for you to buy something without leaving ChatGPT. A couple of big companies signed on, Walmart being one of them, and then Walmart came out a few months later and said, “The conversion on this was awful.” People don’t want to buy one item at a time. They want to build a shopping cart. So that’s technical.
Another piece of this has been access to pricing and to product data that’s really detailed, in real time.
Real time is one of the things AI is particularly bad at, at the moment.
And in some cases, you have retailers who don’t want to make it easy to get that data, like Amazon.
The other piece of it is companies have been building stuff without really knowing what problem they’re solving.
But now there are new services that say they can do this. There’s a startup called Instinct, and Meta just rolled out Muse, which has a similar pitch. Do they actually work?
That is the big question. A lot of sites and retailers have bot-fighting technologies. But theoretically, if you give login information to a site that needs to be logged in, Instinct and Muse are saying, “We can go ahead and do this for you.”
I haven’t seen any horror stories about Muse yet, but there are some Instinct Screwed Up stories out there, like the guy who says Instinct didn’t do a flight refund correctly and cost him hundreds of dollars. Or a guy who lost his Resy account because Instinct was hammering the reservation site for a hot restaurant. When you see those anecdotes, does that make you think this stuff is far from working, or really close to working?
I have two thoughts: One is “wow, [these agents] are getting the attention of all these people at all these companies.” But do Muse and Instinct have enough leverage to get those people to the table to actually make this easy?
The other is that I think a lot of these companies are just going to try building this stuff internally and making their sites smarter and be able to do complex things more easily — but not to be agent-friendly, because they’re worried about disintermediation
Which brings us to Amazon. It seems super-obvious that if what you’re talking about comes to pass, it hurts Amazon. It turns Amazon into a dumb pipe, and they can’t do any of the things Amazon is good at, like showing you other stuff you might want to buy. They just become a warehouse, and they’re the same as somebody else’s warehouse. You may have no idea you bought through Amazon. As a consumer, maybe you’re 100% OK with that, but if you’re Amazon, you’re freaking out.
So Amazon CEO Andy Jassy has been asked about this, I think, on every earnings call the last six quarters. He’s been saying, “We will eventually do business with some of the third-party AI companies, but right now the experience they give shoppers for shopping is [bad].” I don’t think he’s been wrong.
Let’s say it gets really good. Isn’t the core problem that if it works, it’s bad for Amazon?
It would be bad, and that’s why they’re doing a couple things. They’re not letting bots easily crawl their sites. They’re not giving up-to-date pricing info. They’re suing Perplexity, which is also trying to do that.
And they’re doubling down on their own AI use cases within the Amazon ecosystem
Who does a shopping agent work for? In theory, it should work for you, the consumer, right?
That’s one of the big questions. I haven’t seen the language yet from Instinct, but I would be shocked if part of their messaging, once they start actually talking a lot …
Is “our job is to get you the best price regardless of where it is. We don’t have any preferred vendors, we don’t owe anyone anything.”
Right. But then it’s unclear how they will monetize their service. If it’s free for the user, the user’s the product, is typically what we’ve seen in the consumer internet. Once you involve advertisers, marketing, then I think that question is even a tougher one to answer.
But Mark Zuckerberg just said that in the future, he can imagine taking a small cut from whoever’s getting transactions funneled to them via Muse.
If things like Muse or Instinct take off, what happens to advertising? What’s the role of ad and brands if a bot is just finding stuff for you? Does it go away?
There is that possibility, and it would be an incredible story to tell. I am doubtful how much disruption we’re gonna see in the next 24 months in that regard.
And I just trust there will be ways for [advertisers and brands] to insert themselves into these conversations. But they are absolutely worrying about it.
I can imagine an argument where brands and advertisers say, “No, we’re even more important because we can stand out.”
I’ve heard that argument. I think it’s probably true for 1% of them.
It’s the same argument that all the publishers make — we’ll be fine because our audience values our brand and our work.
I don’t think many of those are looking at themselves with a straight face when they say that.
