Everything is gambling now

24:32And this is me making a defense of the prediction markets. I don't think the test should be, are they accurate about predicting things? I think the test really should be, does that price on a given contract accurately reflect conventional wisdom? Now, Now, if you – and in a sense, a prediction market is a replacement for your typical cable TV news pundit. And the pundit is talking about some issue and they're like, well, what do you think about the prospects that in the next 10 years China launches a blockade of Taiwan or something like that? And then they'll say, well, we think there's a 60 – we think there's a 40 percent chance that in the next 10 years.

25:17It's like, eh, these are totally useless. Everyone always is making these 60, 40 percent chances. It informs us nothing. There's no track record. We have no way of going back. But it's, you know, you sound smart. I think there's a 60 percent chance. I think the prediction market is like we can improve on that to some extent because we no longer have to rely necessarily on pundits to come up with that number. We can rely on the wisdom of crowds. I actually think there's something to that. And not only the wisdom of crowds, but the wisdom of crowds willing to put their money where their mouth is.

25:48Which I do actually find sort of compelling. Which I actually find to be very compelling. And I think there's a pretty legitimate reason to think now we – oh, you know what? It's 75 percent chance now. The people with money on the line are very concerned about X or Y happening, whatever it is. And there's a meaningful difference between 55 and 75 percent. And when you see it moving, something is going on. I think it's very interesting from the media perspective. To some extent, I would say that I think of prediction markets as media companies. You can go to polymarket.com and you can see what's moving.

26:25And if something has moved significantly in the last day or whatever, that is a reason to think that some development has happened in the news. So I would say that is very useful. A couple other things I want to say, too. You know, like a point that I've stressed for a long time when it comes to prediction markets is if you think about the U.S. government bond market in the United States or really anywhere, that is literally a prediction market. Let's say what is a six month T-bill? A six month T-bill is what the market expects that the overnight interest rate by the Fed is going to be on average over the next six months.

27:00Who sets the overnight interest rate by the Fed? The 12 voting members of the FOMC. So this is one of the most liquid popular instruments of the world. And it is literally, not metaphorically, it is literally a bet on what 12 people are going to decide over their meetings over the next six months. It is not like a prediction market. It is a prediction market in the most classical sense. And so to some extent, prediction markets not only are valid, vindicated, invalidated, they've been vindicated and validated for a decade because this is an instrument. People betting on what 12 people in a room are going to do is actually at the core of the financial system.

27:43Did you see this thing that happened recently where Spotify put out its year-end data and who was the most streamed artist? And there was this one person, I think it was on Polymarket, who was like, well, Taylor Swift has won the last two years. Taylor Swift is obviously going to do it. The market is way underpriced. They put, I think,$15 ,000 on it, and they were like, this is the best bet ever. I'm going to win. I'm a genius. And then 15 minutes later, they were like, oh, it's Bad Bunny, the number one streamed artist of the year. And they were just following up a tweet with like, well, never mind.

28:10You can't win them all. You know, there's some interesting. So let's go even going back to the most searched person of the year. Incidentally, from what I understand, Google's most searched person of the year is not, strictly speaking, the person whose name was searched. It's about trends, not volume. It's about trends. But this gets to something important, which is contract specification. And so what actually you have to like in many cases, there's a lot of disputes. And I think the companies probably have to mature on this and get better and be more clear about what we are actually betting on and so forth.

28:46And so there could be many of these situations where people think they're betting on one thing because it's like, who's the most searched person in the year is probably literally Donald Trump. And so it's based on some trends. But I think a lot of people get into these markets and don't read the exact contract specs and they get really frustrated because, oh, they didn't realize it. And so probably as these companies mature, they're going to have to get better about being more transparent so that people know what they're actually betting on. Yeah, agreed. So given all of that, you're a reporter.

29:18You're a markets guy. You tweet obsessively about jobs data every 15 minutes, which I used to think was like a bit. And now I think it's just actually central to your personality in a way that I really love about you. That's right. No bits here. No bits. How do you think about Polymarket and Kashi? Do you look at them as useful sources of data in your work? Yeah, I regularly check the prices of things. I want to know, like, who is the favorite in some primary that's going to come? And so absolutely. I, you know, we have here at Bloomberg, we have put the we have we have Polymarket and Kelsey data on the Bloomberg terminal, which I think is like a pretty, you know, we don't just put on data willy nilly.

30:03And so I think there's like people use this stuff. People find it to be helpful. You can do other things, too. You know, it's fun. Like occasionally you can find a chart of, you know, I'm going to track Donald Trump's odds of winning the 2024 election along with the S &P or along with the series of energy stocks within the S &P or various sectors that are perceived to perhaps do well under Trump. And you can see sometimes they move together. So there are a lot of reasons for even people who don't bet, but who are just in sort of news media or want to be informed consumers of the news to check these markets from time to time.

30:43OK, that's interesting. All right. Two more things I want to talk to you about. thing number one is if all of this seems to be growing at the pace that it is and seems to be sort of culturally hitting in these really interesting ways if i'm fanduel or draft kings or robin hood or bank of america like why why wouldn't absolutely everyone anywhere who does anything with your money get into prediction markets like is this just the future of finance there's two things there so one is it certainly feels that way i mean what the bigger trend is not just prediction markets, but the complete obliteration of any lines between sort of legacy financial institutions, speculative trading and, you know, gambling.

31:24So like one of the more surprising news developments, the Chicago Mercantile Exchange has a partnership, I believe, with FanDuel, where they like allow FanDuel users to make bets in small denominations on the price of gold and the price of oil and stuff like that. That was a very surprising move to me. So we really have just completely lost all definitional difference between all of this stuff. Kind of feels like it. And then, you know, the other thing is that the brokerages, so IBKR, Interactive Broker, is one of the, I would say it's the most popular brokerage for the sort of prosumer trading audience.

32:05So it's like someone who wants a little bit more than like a Schwab or Robinhood or something like that. They have, you can trade prediction markets right there. If you go to Robinhood, if you go to the app store right now and you download Robinhood, one of the main things that you'll see there is that you could trade football games through Robinhood. So I think it is absolutely the case that all of the lines between trading, speculating, gambling are just being, yeah, they're just being completely torn apart. Where do we even go from here? If we're at this moment where like, you know, we go all the way back to the pandemic and it's like every this is when crypto went nuts and everybody started betting on everything.

32:50And then it's like we're at the death of the American dream and the male loneliness epidemic. And it's like you can kind of thread everything into what if I could just bet on everything that happened in the world? And it feels like for better or for worse, And I think in many ways it feels scary. It feels it feels tenuous, whatever it's going to be. It feels tenuous. But it is hard for me to imagine that if that collapse you're describing continues, that we're not headed towards some just like insane rewiring of what it means to like be a person in the world. I feel the same way. And I always think, oh, this is also crazy.

33:28Like we're at some sort of like turning point in the world. It's all going to like collapse or something. And then I'm like, no, you know what? I'm 45 years old. This is just like middle-aged man anxiety. Like every 45-year-old man through history has had the same fear. But then I like talk to everyone else and like, oh, shoot, it's not just me. Everyone sort of feels this way. I want people to say, what are you talking about? Like, you're just like, you're just a boomer at this point. No wonder you're like anxious and don't like all this change. But everyone I talk to more or less feels the same way, unfortunately.

33:57I don't really like that. I want people to tell me I'm wrong. But it does – I don't know where this is all going. I don't know how we could possibly know. But this feeling that you have or this intuition, I would say it's, like, shared by, like, by almost everyone I can think of. Yeah. I did read a study yesterday that said that young people think sports betting is bad. Yeah. And that made me happy. Yeah. Maybe we can have some of this, but let's find a little bit of equilibrium here somewhere. I think as a society, we have lost the concept of the difference between differences of degree and or the differences of degree can manifest as differences of kind.

34:40I think we look at things very black and white these days. So you say, like, I want to regulate, like, the idea of just going back to, you know what, you can still bet on sports, but the only condition is you have to drive an hour to a casino and place the bet in person. Like, that strikes me as, like, a very healthy balance. I don't want to, like, ban sports betting such that the only people in the business are mafioso who will, like, break your leg if you don't pay back. I don't think it's great that like all of sports coverage seems to revolve around gambling or increasing or you can't watch sports without ads for gambling.

35:16The idea that like, you know what, let's just like find a moderate position where we allow it. That just seems so antithetical to our contemporary values. People just can't accept middle grounds anymore. But it does seem like there is a bit of a backlash forming. I think a lot of people know people who are addicted to their phones, addicted to betting on sports. They've lost a lot of money. betting on sports. The question is whether we as a society or really as a political system have the capacity to sort of do productive regulation anymore, draw lines. I don't know what the answer is to that.

35:51This is what I mean. I think you can tell the whole story of the world through prediction markets. And it just makes it so sort of head wrangling to me. It's crazy. You're a guy who knows things. Do you ever find yourself staring at Polymarket being like, Should I bet on the jobs numbers? No, I think I will. I'll save my money for losing at poker. Sounds good. All right, Joe, thank you for doing this. This is super fun. I appreciate it. Thanks for having me. It was a blast. All right. Thanks again to Joe for coming. Go listen to his podcast, Odd Lots. Subscribe to his newsletter if you're a Bloomberg subscriber.

36:19It's good stuff. We're going to take a break and then we're going to come back and we're going to talk AI protocols because that's what we do here. We'll be right back. Support for this show comes from Shopify. If you've been thinking about starting a business, but have just been waiting for a sign, here it is. 2026 is the year, and Shopify can help. Shopify gives you everything you need to sell online and in person. Millions of entrepreneurs have already made this leap, from household names to first-time business owners just getting started. Shopify gives you all the tools to easily build your dream store.

36:54Choose from hundreds of beautiful templates that you can customize to match your brand. Setup is fast with Shopify's built-in AI tools. They can write product descriptions and headlines and help you edit product photos. Marketing is built-in too. You can create email and social campaigns that reach customers wherever they scroll. And as you grow, Shopify grows with you. That means you can handle more orders, expand to new markets, and do it all from the same dashboard. In 2026, stop waiting and start selling with Shopify. Sign up for your$1 per month trial and start selling today at shopify.com slash vergecast.

37:33Go to shopify.com slash vergecast. That's shopify.com slash vergecast. Here you're first this new year with Shopify by your side. Support for this show comes from LinkedIn. If you've ever hired for your small business, you know how important it is to find the right person. That's why LinkedIn Jobs is stepping things up with their new AI assistant. So you can feel confident you're finding top talent that you can't find anywhere else. And the best part is that those great candidates you're looking for are already on LinkedIn. In fact, according to LinkedIn data, employees hired through LinkedIn are 30 % more likely to stick around for at least a year compared to those hired through the leading competitor.

38:18That's a big deal when every hire counts. With LinkedIn Jobs AI Assistant, you can skip the confusing steps and recruiting jargon. It filters through applicants based on criteria you've set for your role and surfaces only the best matches so you're not stuck sorting through a mountain of resumes. LinkedIn Jobs AI Assistant can even suggest 25 great fit candidates daily so you can invite them to apply and keep things moving. Hire right the first time. Post your job for free at linkedin.com slash track, then promote it to use LinkedIn Jobs' new AI assistant, making it easier and faster to find top candidates.

38:56That's linkedin.com slash track to post your job for free. Terms and conditions apply. Support for the show comes from Rippling. Your IT team shouldn't be endlessly chasing tickets and manually solving access and identity problems. But when you're running multiple disconnected tools, this is exactly what you have to deal with. And that's not just SaaS. That's SAD. Software as a disservice. What if you could run IT with one system powered by one source of truth? Meet Rippling. Rippling is the unified platform for global HR, payroll, IT, and finance. They helped millions replace their mess of cobbled-together tools with one system designed to give leaders clarity, speed, and control.

39:43By uniting your employees, teams, and departments in one system, Rippling can remove the bottlenecks, busy work, and silos your software might have created. Automated, perfectly in sync, and seriously simple to use, Rippling gives your company one source of truth for your people, their data, and everything they touch. With Rippling, you can run your entire HR, IT, and finance operations as one, or pick and choose the products that best fill the gaps in your software stack. And right now you can get six months free when you go to rippling.com slash verge. Learn more at r-i-p-p-l-i-n-g dot com slash verge.

40:24That's rippling.com slash verge for six months free. Terms and conditions apply.

40:35Millions of players. One world. No lag. How's it done? AWS is how. Epic Games turn to AWS to scale to more than 100 million Fortnite players worldwide. So they can stay locked in with battle-tested reliability. AWS is how leading businesses power next-level innovation.

41:05All right, we're back. Hayden Field is here. Hi, Hayden. Hey, happy to be here. Hayden wearing like a big red. It's just if you're listening to this, I just want you to imagine like devil wears Prada Hayden Field sitting in the studio right now. It's a testament to how cold this office is right now. So I am looking glam, but feeling extremely cold. I turned off the space heater in my basement before we started, and I'm going to freeze to death over the course of this segment. It's going to be great. You need the coat like this, clearly. So you're here because you just wrote a big piece about this thing called Model Context Protocol, which if anyone has ever listened to or watched the Vergecast, you know we love a protocol.

41:45And there's a real bet being made in the AI industry that this is not only just like a piece of the infrastructure of AI, but is like crucial to how we get to the future everyone is predicting and kind of needs to make happen. And the stakes for this seems super high to me. Is that fair to say before we get into it? Like this seems like a big thing that a lot of people need to go well. Definitely. Yeah, it's something that I've been hearing about increasingly over the past year, every month more and more. And then this is kind of the biggest milestone so far in its journey because donating it to a neutral body and having a neutral body to govern it actually is going to help it improve so much.

42:27because all these companies are going to make tweaks to it, improve it, make it better at security, other stuff, because they aren't afraid that they're secretly going to be eventually contributing to their competitors' bottom line. So now that it's all out in the open, no one can own it. It's going to help a lot with it getting better, and therefore AI agents getting better, which is something we really need. Right. Yeah, it's like if agents are going to happen, MCP kind of has to work. But let's just start at the very beginning here. If you are just a regular person in the world, you should not know anything about MCP.

43:02Exactly. You just shouldn't. And you probably never should. But I think it is a thing that your life is about to intersect with. So let's just start from scratch here. What is the model context protocol and where did it come from? Great question because I was doing some of my own research on this when I was writing this piece. I had heard about it. I kind of knew what it was, but it was something that I would have had trouble explaining to someone else. So it's a good thing I wrote this. Um, essentially the, the simplest way to think about it, and this is going to be oversimplified, but imagine you went to a resort and there were a bunch of activities you could do.

43:34Um, and you were handed a brochure of all those activities and how to book them. So before you got that brochure, you wouldn't know what activities were there and you also wouldn't know how to book them. Um, so that's kind of the situation with MCP for AI systems. It's allowing a model to know what tools are available to it or what databases, what contexts, just other things that it can use to make a user's experience better. And then how to use them. So it's not just how to use all this stuff, but also what's even available to begin with. So that's why it helps make AI agents so much better in theory, because instead of just browsing the web and trying to figure things out, you know, in terms of like a web made for humans to browse, it's able to talk to other systems on the back end and get answers like, you know, connecting to Slack, connecting to, I mean, honestly, every single thing that OpenAI announced in October with its ChatGPT apps was powered by MCP.

44:35Interesting. Yeah, I saw a diagram a few days ago that has stuck in my brain ever since. And it's basically like, if you think of the way that software has always worked, which is just through a series of what are known as APIs, you have, if I want to connect my thing to your thing, I connect my APIs to your APIs, and then we are connected. This is like standard software building stuff. This has existed for forever. but if I want to integrate my thing with a hundred or a thousand or 10 ,000 other things, I have to do each one of them individually. And it is like a gigantic pain in the ass. And for like the example I give to people all the time is like, if you've ever used an app that integrates with like Gmail, but not Outlook, it's because Gmail's API is really easy and Outlooks is really hard.

45:21And so it's like, it's just that simple, right? Like some things are easy to plug into, some things are not, but you had to do it one by one every time. MCP, the idea is sits in the middle and says, OK, all of the apps are going to provide a thing to me that says, here's how to talk to AI tools. And then the AI tools just go to MCP and say, what's available out there? And then, like you said, the MCP server says, here are all the things that are available to you and all the stuff that you can do. And it just says rad. And you get it all at once instead of having to build it every single time, every time you want to do anything.

45:55Exactly. That's exactly how it works. And yeah, like we put it like, you know, AI agents need new kinds of APIs and MCP is the standard those APIs will take. Exactly. Yeah, I love that. So where did this come from? Who built this? This guy named David Soria-Pera at Anthropic built it. And his co-creator was a former Anthropic employee named Justin Spar Summers. So they kind of built this as a pet project. They just felt like within Anthropic, people should be using Claude more for their work, like their everyday life. And they felt like people weren't. So they were like, let's make this actually more useful for the end user.

46:34And let's just figure this out on our own. So they both convinced their managers to let them spend like, you know, 80 % of their time working on this project. And luckily, the managers were like, yeah, go for it. You know, if it doesn't work out, we'll see. But, you know, try it out. Why not? That's such a funny way of like anthropic slightly telling on itself, right? That they're like, we had a bunch of people on our product team who were like, that's weird. We make an AI bot and no one wants to use it for work. Why is that? And I think they came to the correct conclusion, right? Which is like, well, because I can't, I have to do all of my work somewhere else.

47:06And so having it live inside of this chatbot is less useful to me than like, what I actually need is a thing that can make Excel spreadsheets for me. That's a very different projects to try and solve. And it seems like that is the thing they went out to try and do. It's like, how do we get the things that are happening in Claude into the software we actually use to do our jobs? Exactly. So they decided to work on this and they spent a couple months on it. I think they started in August of 2024. And then by October, they kind of introduced it at this internal hackathon. And everyone that was participating in the hackathon pretty much at Anthropic built on top of MCP for it.

47:46So that was like the first time they were like, oh, wait, maybe we really have something here. And then from there, it just ballooned. They ended up releasing it like right before Thanksgiving because they said that they figured people would need a break from their families. And they were like, let's just, you know, do it now. And then people have time to work on it, to play around with it. And yeah, then by, you know, early the next year, Sam Altman and everyone were tweeting about it. So there you go. I mean, that timeline is crazy, right? Like, that's so fast. Totally. I couldn't believe it.

48:17And neither could they, actually. They kept saying how surreal it felt and how, you know, it was something that, yeah, David Soriapero was, like, screenshotting tweets from tech CEOs, you know, and just, like, saving them because he couldn't believe this thing he created made it that far. I think it's because, honestly, I think the success is partly because engineers made this for engineers, you know? It's like the people that were on the ground doing this stuff are the people who made it, and maybe that's why it worked. Interesting. Yeah, it's just wild to see how quickly it has gone from a thing somebody built to a product that a company uses to a broader standard.

48:59And I want to get to the news that we have this week, which I think is a telling sort of next piece of the story. and you alluded to it a minute ago, but it's weird to me that all of a sudden Anthropic builds this thing that is working very well and makes a lot of sense. And rather than do what always happens in business, which is Google and OpenAI and a million other companies look at it and say, oh, cool, good idea. We should do that too. They all just pitched in behind MCP. Do you have any idea? How did that happen? That's such an unusual turn of events here that Sam Altman would be like, oh, good idea, Dario.

49:38We'll just throw in with you. I think it's because they knew that it was the best possible standard, the one that had the chance of going the furthest. And they still were a little bit cagey about it. I mean, they had a group of core maintainers at each company that would talk and meet in person and have a Discord chat and talk about ways to improve the protocol. But still, you wouldn't see like Google, you know, going all in on fixing MCP's security potential problems right now because they didn't, you know, if Anthropic ever decided to, you know, change its mind, right now it's open source. But if Anthropic ever said, oh, actually, no, we want to keep this for ourselves, none of these companies wanted to improve it too much because they would be helping Anthropic out in the end.

50:23So I think they would improve it and they, you know, talked about it and they thought it was the best possible standard for making all of their AI agents work better. So it was kind of like they were all getting something out of it. But that's why I think the news that is happening this week is so big, because now that it's going to be governed by a neutral body and overseen by the Linux Foundation, it's going to be a real standard that people can, you know, contribute to, tweak, make real improvements to without the fear of, you know, one company eventually benefiting by itself. Yeah, well, so then that brings me to the sort of flip side question, which is why would Anthropic release this to the world?

51:01You look at this like this is great. We've built a thing everybody suddenly wants to use. This is this is a way we can get MCP and Claude to be more powerful. Why would they give it to everybody? I know you talked to Mike Krieger at Anthropic, who I think is probably the person who made this decision. Do you have any idea why he made that decision? Yeah, he said that he just felt like it was really important and it was the way to make this an industry standard, that it would not really happen without doing that. And, you know, it was always open source, but a couple months in when, you know, the two co-creators were seeing the pickup, they started thinking, oh, should we donate this?

51:42Should we, you know, give it to a foundation that has done stuff like this in the past so that it can actually like go the extra mile and become a standard? And Krieger was apparently super supportive of that and one of the most vocal, you know, proponents of that. And the other reason is because Anthropic has shortcomings. You know, there are certain things that they're not going to have the resources or the knowledge to go full-fledged into when it comes to bettering MCP. So Krieger himself said authentication and security are two of those things. Like, you know, they don't hit those problems first, and they don't have as much knowledge as, say, like, you know, Google or another tech giant in fixing those things and those potential security concerns.

52:26So that's why, I mean, you know, it's kind of benefiting everyone if you just donate it. I mean, it's funny. My immediate instinct is to be like, well, there has to be something more cynical to it than that. But then I was like, well, they did just open source the thing and give it to the Linux Foundation. So I suppose it can't, there can't be but so much cynicism. They did then go and, you know, do the thing. And the fact that, I mean, it's like, it's kind of like a rising tide lifts all boats in this one situation. Because, you know, I mean, for Anthropic, if they ever do need it to be way better at security and authentication, it will be.

53:01And that, of course, it'll help their competitors, but it'll also help them. So I think, you know, basically it's like this would have been stymied if they kept it. You know, it never would have advanced quite as far as it could. So now that I think they just knew that this was the only next step if they wanted to ever get significantly better and actually become an industry standard for real. Interesting. And I would assume that this change, giving it to the Linux Foundation, bringing these other companies in, it is now officially or unofficially, it is an industry standard, right? Like this is have you talked to anyone who is like, well, we're still waiting to see on MCP like this thing seems to have essentially unanimous momentum at this point.

53:42Yeah, definitely unanimous momentum. I mean, I've talked to some people that are like, you know, we'll wait and see because, you know, the real market industry standard is just when everyone adopts it, which has already kind of happened in a lot of ways. So, I mean, some people are like, it basically is. Some people are being a little cagey about it just to see where it improves in certain areas. But it's certainly on its way, and I would say it already is. The other thing about it is, and I think this goes back to your point of, you know, what did the other companies get out of it? But Anthropic wasn't the only company to donate something to the Linux Foundation.

54:14So, you know, OpenAI donated Agents.md, Block donated Goose, its open source AI agent. So, you know, I mean, they're all kind of part of this. Block donated Goose is just a cent. I just I can't I can't just let that happen. I just had to look up what it was called again because it constantly goes out of my brain because it's so strange. But, yeah, the Agentic AI Foundation is what all these companies established. It was like Cloudflare, Bloomberg, Block, AWS, Microsoft, Google and OpenAI and Anthropic. So all of them came together to create this neutral body that's under the Linux Foundation. So technically, they all can have a hand in everything, but none of them own it.

54:54And so I think that kind of consortium helped with all of them feeling comfortable enough to donate also. That's fair. So what what happens when something gets given to the Linux Foundation? Like, what is actually going on here? What is now under the control of this group? So now all of these tools and this protocol are under their control. And, you know, it's just open for people to make it better. So, you know, Google could, for example, put like 10 or 20 engineers on this just to improve its security. You know, another company could be like, we got to fix authentication potential issues here.

55:31Let's put a ton of engineers on this and everyone's going to benefit from it. But the other interesting thing to me was that I talked to the CEO of the Linux Foundation. And he's been in the field for, you know, over 20 years. Like, he oversaw, you know, the expansion of, like, Kubernetes and containers. Like, he was the person that kind of, like, shepherded that. So for him to say that he'd never seen anything like this in terms of interest in MCP was crazy to me. And that's what he said. He said he could barely keep up with the number of inbound calls from organizations that wanted to be a part of this.

56:03And he said that oftentimes when a new protocol or, you know, a new thing that he's trying to push gets introduced, he has to scratch and claw his way and try to convince people to, you know, make it a standard. And in this case, he didn't have to convince anyone. It was just like already widely seen as being obviously the standard. And so I think that's part of why this is such a big deal, too. That is really interesting. And it seems to me that that, again, it's just like momentum begets momentum. Right. And it feels like even, God, like nine months ago, it was like, OK, MCP is the thing. Right.

56:38Somebody might try to build a competitor. We might end up with two weird standards. And it's like it just reminds me of all the sort of Fediverse stuff where it's like, OK, well, somebody built ActivityPub and everybody got really excited about ActivityPub. And then Jack Dorsey was like, I don't like it. I'm going to build Noster. And then they did AT protocol for Blue Sky. And I actually think that has been like hugely detrimental to all of those services. And it would have been better for everybody if they had just picked one. And I think the AI industry seems to have correctly understood that picking one is very important.

57:08And that what we need is for all of this to work. Like, I cannot emphasize the extent to which none of this works. And MCP is a way to get it there. And it feels like they made this call correctly in that sense. I think that they all have taken kind of an honest look on the fact that their agents are not doing what they want to do. And this is a way to hopefully bridge that gap. And they're like, at this point, you know what, let's just all help each other instead of because there's no other way. I mean, they need this stuff to work and they need to drive a profit, as we've seen from OpenAI's$1.4 trillion situation.

57:42I mean, they all need to make money and they need to do it quickly. So this is kind of their best bet. The next piece of the explainer I want to talk about here, and I want you to just sort of cast out a little bit, like, let's assume this works. What happens next? Everybody is like, all right, sick. MCP is the thing. Agents don't just magically get better, right? Like, what happens now that this is a thing everybody is sort of, in theory, willing to, like, bet on and believe in? What are people going to go spend the next 12 months doing? I think that this basically gradually is going to make agents better in a lot of ways, just because now that everyone can take a deep breath and say, okay, I'm not going to be contributing to Anthropics IP secretly in the future.

58:26You know, I can just make it better and make this work. Everyone's going to spend the next 12 months, you know, working on MCP and integrating into stuff and hopefully making agents actually better. Like, here's an example. agents traditionally have been like pretty bad at buying you stuff or like if you're like oh you know let me find a pair of um you know white sneakers from this brand in my size for pickup in my city um as i think v wrote about recently it has trouble doing that and it has trouble finding that and putting in your cart and all the things it's gotten better but it still struggles a lot.

59:01So I think in this way, that type of stuff, like actual consumer-facing end-user tasks are going to be more useful with agents. And we won't have to know anything about MCP or even know the name of it. Hopefully, this stuff just gets easier and better and more successful over the next 12 months. And we don't even need to hear anything about the back end of it. It just happens. So for example, like, you know, let's say you ask Chachi Petito, do what I just said for you. Find those shoes. You know, maybe with a Macy's integration or whatever, I can talk on the back end to Macy's instead of just like browsing the Internet like a human would, which is not that efficient for an AI agent.

59:40So, yeah, that stuff I think is just going to gradually get better. But then again, you know, agents have always moved slower than companies said they will. So we'll see what the actual timeline ends up like. Well, to your point about the Macy's thing, it's not lost on me that not that long ago all of these companies launched features that would just go use the computer for you right like this was computer use was the thing that that actually chat gpt would open a chrome window inside of chat gpt and click around the web browser to do stuff on your behalf and then it would it would you know dump stuff into your cart and again this didn't work.

1:00:18But it was kind of the idea there for a minute that like, OK, what we can do is solve this problem in literally like a brute force way. This is just the end of that, right? This feels like everybody, I think, must have correctly surmised that was never going to be possible and it was always going to be problematic. We need a better way. So I floated that. And whenever I floated that to an expert I was talking to for this piece, They were like, no, it's not the end of it. I was like, well, what do you mean? I mean, clearly it is. But their caveat was that they still really need that for a bunch of areas of the Internet that aren't going to have MCP.

1:00:58There's going to be some areas of the Internet that humans, they're built for humans and they're going to stay built for humans. And that's just the way it is. And so that was an important milestone and it's going to keep needing to get better. But MCP is going to shortcut a bunch of the stuff that you would have had to use tools like computer use for and make it way faster and easier. So I think you're right in that it's going to diminish a lot. Like, we don't need to use computer use for everything now or in the future. But it's apparently still going to stay around for a bunch of areas where, you know, MCPs— I also feel like it might just be a euphemism for saying we still need to do it in case companies don't want to give us access to their data.

1:01:36Yeah, I think that's true. Yeah. I mean, this is the thing we talk about all the time, right? Like, Neelai calls this the DoorDash problem. Like if I'm a platform provider, why am I going to just happily offer AI models a perfectly mapped set of my data so that no one ever comes to my website again? You might not. Right. Like there are a lot of reasons it's useful and valuable to do so. Like flight tracking is one that I'm sure came up a bunch in your conversation. Yes. Because this is like the canonical one. Right. You can go and you can try to scrape airlines' websites to get their flight info, but that is inefficient and bad and wrong.

1:02:12And also, all of these companies have set up this incredibly accessible database so that the kayaks and booking.coms of the world can access this stuff. And if they just point that at MCP, all of a sudden, LLMs will actually be able to go in and get good data in order to buy you flights. Sure. As far as that goes, sure. And I think the airlines may be more likely than most to just be happy selling you flights, even if you don't go to their website. But there are going to be a lot of them that don't want you using LLMs to do their thing. And so maybe that's the computer use angle is like, yeah, it's going to still have to click around the DoorDash website because not everybody is going to want to let the LLMs access DoorDash.

1:02:56Totally. And I think about the flight thing a lot, especially because, you know, when the scandal broke that a lot of airlines were doing, you know, per person pricing or like pricing solo travelers higher than group travelers. You know, I mean, I'm sure that that's going to be a thing. They may want to just hold on to the power to do that stuff. And, you know, that's how it is. And, you know, I even saw a story the other day about, like, you know, some grocery app testing algorithmic pricing. So, I mean, I even saw something about Instacart potentially, you know, introducing, like, AI pricing tools.

1:03:34So, yeah, I mean, a lot of companies want control over who's searching and when and what their history is like so that they can charge you different prices. So we'll see how that works out here. I hate that. So I would love for MCP to overhaul that and just like, you know, we all have equitable insights into what something costs. I know. Right. On the one hand, MCP is sort of a great user experience in the sense that you can just sort of describe what you want to have done and it gets done. Like I was talking to Amir, the CEO of this app called Todoist the other day, and he is all in on AI and MCP.

1:04:15And the thing that he really likes about MCP is he's like, well, we get a lot of these people who have really specific feature requests for us. They're like, I have my to-do list, but I want all sorts of things to happen. Whenever I, if I complete a to-do list, I need it to also go to Jira and market complete, or I need it to send an email to my boss saying that I've done this thing. Whatever. Everybody has these sort of specific automated or semi-automated workflows that they want. And Todoist can't build them all. But the idea that through MCP, Todoist can just say, here are all of your tasks.

1:04:46Here is their state of completion and their due dates and their projects and their priorities. Have at it, right? And then Claude or ChatGPT or Jira or whoever can tap into that. And suddenly you can build these automations essentially just by asking for them. It's very, very cool. And there's just like little pieces of that that it's like, Like if we can put that stuff in front of users and just say, I want to whenever I book a flight to L.A., I want it to also book me a hotel automatically. Like that's a thing you can do that MCP makes possible in a way that has not been possible before. It probably still won't work, so I don't recommend trying it.

1:05:23But it is like theoretically MCP is what helps us get there. Do you know what I mean? Exactly. Yeah. Yeah. But there's just it really, really, really relies on everybody being on board. And there are still enough reasons to not be on board that I'm just skeptical. Right. I know. I totally agree. And that's why I think we'll have to see what happens over the next year with companies addressing some of their concerns in MCP, like authentication, security, stuff like that. There's a bunch of areas that I think could be improved and that'll get more companies on board. But we'll have to see how that actually shakes out.

1:05:58Plus, you know, I've been burned enough by AI agents not working that, you know, I'm too cynical. We'll have to see how it actually plays out. Like, you know, do some tests now, do some tests in a year, do some tests in six months. But yeah, it'll be interesting to see how much this actually overhauls the landscape for consumers. Because, I mean, you know, enterprise users, agents are working a little bit better because it's a predictable, controllable environment. But when it comes to like, you know, booking flights, booking travel, figuring out a restaurant that everyone can go to and putting it on their calendar, stuff like that is still kind of a pipe dream in some ways.

1:06:32because it just doesn't work well enough for everyday people to be using agents for this type of thing. What are the things about the protocol that the people you talk to are still working on or thinking about or paying attention to? I mean, this is 14-month-old software that we're just going to stick at the middle of the AI industry. Are people concerned about pieces of it? Pretty much just security is the main concern I've heard. You know, there are things that Anthropic just isn't equipped to, you know, beef up on MCP. And so that's part of also why they're donating it. They're like, you know what, you guys go ham and just fix all this stuff they're concerned about.

1:07:10Great. So, yeah, that's the main thing I've seen. Are we talking about security in the sort of normal AI agent kind of security where it's like, if I'm allowing an LLM to go talk to a bunch of tools and talk to a bunch of services and access a bunch of my data, there's just a lot of potential points for like everybody talks about prompt injection. And there's a lot of places where my data becomes available to a lot of different services. And there's just like this whole system needs to be kind of hardened against releasing my data and my work in the wrong places. Is that the security stuff you're talking about?

1:07:45Yeah, that's a big part of it. The other part is like authentication and just making sure someone is who they say they are. So that's another thing Anthropic wasn't really fully equipped to handle. So they're, you know, saying, okay, you guys do this. So yeah, those are the things like when it comes to payments, stuff like that, sensitive data. So yeah, I mean, we'll see how that stuff shakes out. I think there's going to be a lot of work done and certain companies are just going to put teams on this and just say, hey, like you guys for the next couple months, just work on MCP and try to fix this aspect that we're a little concerned about.

1:08:16Do you think every big and small company is going to spend the next year doing that, like pointing somebody at MCP and saying, OK, let's stand up a server and see what's possible here? Is this going to start to happen that fast? I don't think it's going to be every company. I think it'll just be the main players that have a vested interest. I mean, this is just kind of anecdotal, but I think smaller companies are just going to wait for the bigger companies to do it. They don't need to put the resources on this. And they know that bigger companies have you know, a lot of the same concerns they do.

1:08:43So they're like, yeah, just you guys work on the infrastructure part of it. We'll use it when it's ready. So yeah, I mean, I could see all the companies that are involved in this AAIF foundation, you know, putting their teams on it, you know, AWS, Google, OpenAI, a lot of the same companies that have representatives involved in this kind of like discord group chat and core maintainers group where they were already kind of talking about ways to improve it but now they're going to actually make all those changes yeah it does feel like we're we're six to 12 months away from all of those companies just making it like plug and play mcp tools where it's like aws is just like click here and you have an mcp server it's like great and that's that's when it'll start to really take off when it's like i don't even have to figure out how to do this relatively easy thing from scratch it just comes free with whatever piece of software i'm using and that's if you want mcp to win that's what it's gonna have to look like but we'll see 100 we'll see i really do hope i both am very excited to talk about mcp and i also hope this is the last time we ever talk about mcp do you know what i mean that's the same that's the correct answer it's like we don't talk about tcp ip on this show for a reason it's just like yeah it works it does the job and i don't ever have to think about it ever again and it feels like that's what mcp needs to become um all right hayden will You stick around for five more minutes.

1:10:03We have a hotline question. I want you to answer. Oh, yeah, 100%. All right, we're going to take a break. We'll be right back. Support for the show comes from Udacity. Much has been said about the surge of AI, especially in the workplace. It can feel pretty intimidating, no matter what industry you're in. That's where Udacity comes in. Udacity is an online learning platform with courses in AI and tech, including generative AI, Gentic AI, Python, data science, and much more. When you learn with Udacity, you're not just passively watching videos or reading articles. You're doing practical exercises and projects that prepare you for the job you want.

1:10:39That's why 90 % of Udacity graduates' surveys say that they achieve their enrollment goal. Udacity just launched a master's degree in AI. You heard that, right? And when you have a certification from Udacity, recruiters and employers take notice. So for a better job, better salary, and better skills, you can check out Udacity today. The tech field is always evolving, and you should be too. You can try Udacity risk-free for seven days. Once again, that's udacity.com backslash verge for 40 % off, and make sure you use the promo code verge so they know we sent you.

1:11:18Support for this show comes from Zapier. The list of tasks AI can handle these days is so extensive, it can be confusing and, frankly, overwhelming. And sure, there's a lot of potential to talk about. But you don't need AI to do everything it could potentially do. You need it to sync up to your workflow right now. Zapier says they can help. Zapier helps you deliver on your AI strategy, not just talk about it. With Zapier's AI orchestration platform, you can bring the power of AI to any workflow. So you can do more of what matters. Teams already use Zapier to streamline their rev ops, marketing, sales, IT, HR, and more.

1:11:58You can connect top AI models like ChatGPT and Cloud to the tools your team already uses. So you can add AI exactly where you need it. Zapier is for everyone, tech expert or not. And teams have already automated over 300 million AI tasks using Zapier. Join the millions of businesses transforming how they work with Zapier and AI. Get started for free by visiting zapier.com slash verge. That's Z-A-P-I-E-R dot com slash verge.

1:12:48Once it's iPhone 17 and at T-Mobile, you can get it on them. That center stage front camera is amazing for group selfies, right? Mrs. Claus. I'm Mrs. Claus's much younger sister. And at T-Mobile, there's no trade-in needed when you switch. So you can keep your old phone. Or give it as a gift. And the best part, you can make the switch to T-Mobile from your phone in just 15 minutes. Guys, my side of the tree is slipping. Timber! The holidays are better at T-Mobile. Switch in just 15 minutes and get iPhone 17 on us with no trade-in needed. With 24 monthly bill credits for well-qualified customers, plus tax and$35 device connection charge.

1:13:22Credits end and balance due if you pay off earlier. Cancel finance agreement, 256 gigs,$830. Eligible port in a new line,$100 plus a month plan with auto pay, plus taxes and fees required. Check out in 15 minutes or less per line. Visit T-Mobile.com. Support for this show comes from Darktrace. Defenders and cybersecurity are always there when we need them, keeping the systems that run our daily lives safe from bad actors. At Darktrace, they think defenders everywhere deserve a bigger spotlight for keeping the lights on. They deserve blockbuster movies about breaches that never happened. And they should get a parade every time they block a novel threat.

1:13:55But most of all, they deserve AI cybersecurity that can stop novel threats before they become breaches. Across email, clouds, networks, and more. That's why Darktrace is going beyond for cyber defenders. Such as IT decision makers, CISOs, and cybersecurity professionals to give them the power to see across their entire attack surface. and stop the threats other platforms miss. Darktrace is a cybersecurity defenders deserve and the one they need to defend beyond. The world needs defenders. Defenders need Darktrace. Visit darktrace.com slash defenders for more information.

1:14:38Hey, how's it going? I'm Eric, one of the Vergecast producers. And before we move on with the rest of the show, I just wanted to mention that this week's hotline is presented by AWS, how leading businesses use AI for next level innovation. Now back to David. All right, we're back. Let's do a question from the VergeCast hotline. As always, you can call 866-VERGE-11. You can email VergeCast at theverge.com or you can do what we're about to do, which is you can just post it to me on social. I will lose the post and I will continue to think about the question you asked me for so long that I bring it back up on the show.

1:15:11So to the person who sent this to me, please know that I'm very sorry that I cannot find this post to give you credit. I'll put it in the show notes if I can find it. But Hayden, the question was essentially, why is it that every AI company is trying to get me to shop? That somehow everyone has simultaneously decided that the main thing you want to do with an LLM is buy something? I have a couple of theories on this, but I'm curious what springs to mind for you. That's a great question. I think partly it's a really good foil for like a complex multi-step task that involves common sense and logistics.

1:15:50So in a way, it's just like kind of a good tell for whether an AI agent is actually good-ish or not. But, you know, I mean, common sense of which things to highlight or put in your cart. You know, can it understand the instructions of what you are even looking for? So let's take the shoe example. Let's say you wanted like white Nikes size 8.5 for pickup in, you know, Manhattan. And that's a lot of different steps. And AI agents are designed, they say, to do really complex multi-step tasks. And they need to be able to reason. Okay, well, this shoe is white, but it's not Nike. So, okay, I'm not going to put that in her cart.

1:16:32Things like that. It's just kind of a good foil for what an AI agent can actually do. Plus, yeah, time. We've already learned that chatbots can't really tell time. So, yeah, like, you know, is it available for pickup today, tomorrow? Manhattan, geographical, you know, limits there. So, yeah, I think it's just kind of a good foil. The other part of it is I'm sure, you know, all these companies want to make money. So why is TikTok shop a thing? Why are influencers, you know, peddling their special codes for you to buy something? It's because e-commerce is a great way to make money. I'm sure that if you buy something in ChatGPT or in Claude, eventually they will have a certain cut or a commission of that, you know?

1:17:14I mean, it's like easy money. Why not? And the other part of it is they'll learn more about you. You know, data is valuable and what you buy says a lot about you and is really lucrative to these companies. So, yeah, I think those are the main reasons that come to mind for me. I'd even thought about the data piece of it, that it does seem to be true that 2026 is going to be the year of ads in AI tools. And one, A, one really great thing to advertise is products people can buy. and B, one really great way to get better data is to watch people go through shopping flows. That is where all of the money is in advertising.

1:17:56And so if you want to do that, having one-click shopping stuff inside of your AI tool goes a really long way. I think the only thing I would add to that is that it is like a sort of obvious and benevolent use case, if that makes sense. Like I think, I'm sure you hear this too, But a thing I hear from AI companies all the time is they're not very good at explaining to people what you can do with their product. And this idea that they're just like fun to talk to and sort of act like your best friend gets you a long way. But ultimately, the tool has to do something for you. And a thing that the Internet until now, frankly, has not been very good at for all the reasons you just described is shopping.

1:18:41Shopping is a thing everybody does. It's like a universally appealing demo. Almost everybody likes shopping in some way, shape or form. And it's a non scary and totally mainstream use case that you can just point people at and be like, yeah, we can find you better deals than you would otherwise. There was this Alexa announcement last week where you can have Rufus, Amazon's shopping AI, which again is like clearly named to not seem scary. I don't like when they do that. But anyway, it will actually monitor prices on an item for you and will buy it automatically when it hits that threshold. This is like, that doesn't need to be a generative AI use case, but it is also like a perfect and understandable use case for this kind of tool in a way that I think there are not a lot of those for AI.

1:19:37Yeah, especially because you're absolutely right. I mean, I think I can't think of anyone that wouldn't use that. You know what I mean? Exactly. Which is very funny that that is a thing that Amazon has to build on top of Amazon. Like, what a funny way to admit that your pricing scheme is insane and doesn't make any sense to anyone. Again, algorithmic pricing. Yeah, truly insane. But yeah, I think it's funny because these things are built, I think, a lot of times. Honestly, a lot of AI tools in general are just built, I think, sometimes without an end use case in mind. They're just built to build them without really solving a problem or without thinking about what it's actually for.

1:20:15So this is a great example of something that, yeah, people would use and want to use. Booking travel, shopping, saving money. Do you remember when like Google Chrome had that extension, Honey, where it would like try automatically every coupon code when you were going through a cart? I mean, that's a great use of technology. So it's like people want to save money. They want to, you know, be. And also, here's another example. Honestly, Instagram ads, a lot of people like them, right? They find them useful. They buy stuff from them. That's one of the only ad experiences that I've seen people actually enjoy.

1:20:50And so, I mean, yeah, I think that's what ChatGPT eventually is going to offer. When I was at Dev Day for OpenAI in October, Sam Altman was talking about ads within ChatGPT, and he said specifically that he wanted them to operate kind of like Instagram ads in that, you know, he wanted to recommend products that people did want or would find interesting. So I think that's what we're going to see happen next year, honestly. They put it on pause for a little while during their Code Red, But I think next year they're allegedly going to release adult mode in the first quarter, and I think they're going to release ads soon after that.

1:21:24Yeah, it's really it's going to be a banner year for it being a fun product for humans to use. No, I agree with all that. The only the only other thing I would add is just to to put an even finer point on your this is how they make money piece of it. This is how they make money, right? Like it is so easy to make money from shopping because this is like a huge industry that everybody understands. The commissions are well established. The pipeline exists. And if you can be the one who is like, oh, now we're the place that people go to find product. products you you can just take part of that for yourself like i've been following google shopping for years because i think google shopping is fascinating it's like what if you had all of the data in the universe and could see the whole internet and still couldn't do better than amazon fascinating but the the google shopping's whole thing has been okay people are searching anyway if we can just prevent you from going to their website and then clicking to buy something and we can just Google pay you straight through that whole process, A, that's a pretty good user experience because you found the product you wanted faster, you pay faster, you get the thing faster.

1:22:28But also Google now owns and gets to extract value from more parts of that process. So you can think even like if ChatGPT gets access to my credit card number and can just be like, oh, I found the Nike shoes you were talking about. Do you want me to just go ahead and buy them? And you can say yes. A, excellent user experience. B, ChatGPT gets to just run around a whole bunch of people who want to take commissions and just take it for themselves. 100%. And I can relate because one of my first jobs in New York was a personal assistant. And yeah, that's the type of thing that I would be doing for people.

1:23:01So there you go. I mean, it's useful. People want it. Were you a good personal assistant, would you say? I think I was pretty good. You know, I booked the travel. I kept the calendar. I did a myriad of random chores. worse. You know, I think I was pretty good. I was way more organized about the person I was being an assistant for his life than my own, because I think you can only manage one person's life at a time. I managed her life, mine. I fought by the wayside a little bit until I was done being a personal assistant. I also had a separate phone with a separate ringtone that I had to keep on at all hours.

1:23:35So there you go. This is why ChadGPT is going to totally collapse as it starts to take over more and more of our lives. ChadGPT is just going to get like weird and sad, but it's going to be very helpful for us. Exactly. All right. We got to get out of here. Hayden, thank you as always. Thanks so much. All right. That's it for the show. Thank you to Hayden and Joe for being here. And thank you as always for watching and listening. If you have thoughts, questions, feedback, feelings about prediction markets, or guaranteed winners on Polymarket and Kalshi, I can't do anything about it, but I want to hear about them anyway.

1:24:04Email vergecastattheverge.com or call the hotline 866-VERGE11. We absolutely love hearing from you. It is the single best thing. I have a lot of inboxes in my life and the only one I like is the Vergecast hotline and the Vergecast email. So keep all of your stuff coming. We love hearing from you. This show is produced by Eric Gomez, Brandon Kiefer and Travis Larchuk. Vergecast is Verge production and part of the Vox Media Podcast Network. We're going to be back on Friday and then next Tuesday, those are our last two shows of the year. Next Tuesday is the Vergecast holiday spectacular, which is one of my favorite episodes every year.

1:24:35And this one is extremely fun. So stay tuned. And then we are all going to just fade into the holidays and I'm very much looking forward to it. We will see you next time. Rock and roll.

1:24:48300 sensors. Over a million data points per second. How does F1 update their fans with every stat in real time? AWS is how. From fastest laps to strategy calls, AWS puts fans in the pit. It's not just racing. It's data-driven innovation at 200 miles per hour. AWS is how leading businesses power next-level innovation. Support for this show comes from Amazon Ads. Every business owner has been there. You put a significant amount of money into an ad buy and then wonder, did those ads actually have an effect? Luckily, there's Omnichannel Metrics from Amazon Ads. Omnichannel Metrics helps advertisers understand how their Amazon Ads campaigns drive sales both on and beyond Amazon while campaigns are still mid-flight.

1:25:40OCM measures performance across streaming TV, video, audio, and display, helping you understand what's driving results across the full funnel. Using Amazon Shopper panel data plus third-party signals, you'll be able to see beyond Amazon product sales, units sold, and ROAS. Whether customers buy on Amazon or at a brick-and-mortar store, you'll understand the full impact of your campaigns. measure comprehensive sales impact to better understand purchase behavior and drive greater efficiency, effectiveness, and ROI. Tired of guessing where your ads are actually driving sales? Capture the full impact of your media spend with Amazon Ads Omnichannel Metrics.

1:26:20Head to advertising.amazon.com to learn more. That's advertising.amazon.com. If you're tired of database limitations and architectures that break when you scale, then it's time to think outside rows and columns. MongoDB is the database built for developers by developers. It's ACID compliant, enterprise ready, and fluent in AI. That's why so many of the Fortune 500 trust MongoDB with their most critical workloads. Ready to think outside rows and columns? Start building faster at mongodb.com slash build.