Ep 712: AI Agent Crash, Software Collapses and Non-Human Economies.

This is the Everyday AI show, the everyday podcast, where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

AI developments change by the hour, which means what we as business leaders are capable of is changing daily. And that's both extremely exciting and really hard to grasp, because as soon as you learn a new AI technique or model, it's already outdated. And then it's going to take months to get the simplest approval to use that thing at your company. So what's the solution? Well, you have to know what's coming ahead, and that's what you have me for. Granted, I'm not a mind reader, but I do spend almost all day, every day talking with the people who are building the technology and using it for myself and teaching others how to use it is quite literally my job. That's why each year I do my predictions and roadmap series. Because you don't have 10 hours a day to keep up. I do. So I can tell you what's ahead around the curve. Because I'm usually in the first wave of cars and no, I happen to talk to the person who's actually paving the road. Get it? It's an analogy. But that's also what we're kicking off here in what I think is one of our most important shows of the year, the 2026 AI predictions and roadmap series, volume one of two. All right, let's get after it if you're new here. Well, great show to tune into for the first time. My name is Jordan Wilson and welcome to Everyday AI. This is your daily livestream podcast and a free daily newsletter helping everyday business leaders like you and me keep up with everything that's happening in the world of AI. How to make sense of, to grow our companies and our and our careers. So starts here, unedited, unscripted live stream podcast. But if you want to be the smartest person in AI, our website is where you do that. Your everyday AI dot com. Each day we put out a free daily newsletter recapping not only that live stream podcast, but also everything else happening in the world of AI. So a little bit, yeah, if you don't really know what this thing is. Well, I have talked to hundreds of the smartest people in AI. That's what I do all year. Right. And I, I have this running series of notes, right, because there's a lot of conversations, both with guests and a lot of other people that I, I'm lucky enough to get to talk to I always have great ideas and I'm, you know, constantly connecting these dots. And, you know, I realized pretty quickly after doing this thing for like a year, I'm like, wait, I can sometimes understand what's coming before a lot of other people do, because this is all I do, right? So I've realized over some time that it's probably helpful for you all for me to share those internal notes, the off the record conversations, and, you know, me stealing some insights from some great enterprise leaders. So if you don't have hundreds of hours to go through and listen, listen to, you know, the past year or so of podcasts. That's why you got to listen in today. Also, as an FYI, I don't do these things and uncheck them. I think this is our third year doing it. So I always do a kind of a grading report at the end of the year. So make sure to go listen to episodes 674 and 676. So you can kind of go audit my predictions from last year. So I just did that audit about a month ago. So still pretty, pretty relevant. And I think it's a great kind of one, two combination between that one and this one, the 2026 AI roadmap and prediction series. And you know what? I. I know what you're thinking. You're like, all right, Jordan, it's, you know, Feb. Mid February. What are you doing this? Like, why aren't you doing it in January like everyone else? Well, number one, I don't know if you can tell, I'm still sick. I've been sick like five out of the last eight weeks. That's the first thing. But the second thing is I kind of felt and sense something in December, actually like late November. I'm like, I think we're coming up for a seismic shift here. And I think that shift did happen December and January. So if I'm being honest, my notes in December, you know, anyone's notes in December to where we are today, completely off, right? Everything has changed. Everything has changed. So, you know, a couple reasons, multifactorial. But regardless, here we are. And, you know, thank you everyone for your patience. I've been getting a lot of messages now almost every day. You know, people asking on, you know, LinkedIn, Twitter, text messages, you know, emails, like, when are, when is the, you know, 2026 AI prediction and roadmap series coming? So it's here. So today and tomorrow, two part series. Not going to do five parts. Don't. Don't have the time to dedicate that many shows right now, to tell you guys the truth, because a lot of our, you know, weekly programming has, you know, gotten pretty popular and people, you know, are used to seeing the, you know, AI news on Monday and the, you know, AI at work on Wednesday. So we're just doing two this time. All right, and FYI, a couple other housekeeping things. You're going to want to repost this one. Just letting you know. So if you are listening on the podcast, check your show notes. We always leave the link to this LinkedIn live stream. So go repost this on LinkedIn and I'm going to send you our exclusive bonus guide. Because I actually, my first list of prediction in roadmaps was like 150, and then I narrowed it down to 50 that I actually built out pretty in depth. So obviously we're only going to 26. It is the year 2026. So we're going over the 26 top predictions. So 13 today, 13 tomorrow. But we have a guide. It is amazingly in depth. Like I said, not even to have time to share all of this. I wish I did. So go repost this show and I'm going to go ahead and send it to you. It's ready to go. Copy and paste. It's on your way. All right, so I'm not going to wait, make you wait anymore. I'm going to read off the 13 we're going over today and then I'm going to go on each one just for about, you know, two, two or three minutes or so. All right, so, and these aren't in any particular order, FYI, Sometimes I do them in a particular order. You know, this, this year just kind of, you know, spaghetti on the wall. So first, number one, an AI lab pays universities for student data. Maybe an A. Number two, venture capital firms morph into venture studio firms. Three, grounded only modes or models become enterprise hot items. Four, Google wins the short game while building the long game. Five, internal agent to agent economies take hold. Six, software stocks and software ETFs suffer a sustained 2026 drawdown. Three, seven, messaging becomes the universal agent remote control interface. Eight, agent crash becomes a national headline event. Nine, shadow AI triggers a Fortune 500 data breach. Ten, agent audit logs become mandatory in regulated sectors. Eleven, skill marketplaces become the largest agent risk surface. 1212 hour autonomous task horizon becomes the new brag metric. And 13 agent ops become a formal enterprise function. There you go. So you have them all. Let's dive into them a little bit one by one. So first in AI lab, I'M not going to say acquires a university because that sounds crazy even for me. Although I do think that's essentially going to hire happen by the end of the decade. But I think in 2026, at least one Frontier AI lab will announce a structured partnership with a university centered explicitly on access to student generated data. Human data. Yeah, we need it. And I'm going to be talking a little bit more tomorrow kind of on work slop on why this is really needed. But I think this announcement is going to be framed as more of like a research collaboration. But the real asset is big AI labs. They need high quality human generated data because it is at a premium. And you might be wondering why, right? AI is not new. AI was around well before Chad, gbt, even generative AI that was just generating text, right. The GPT models started littering the Internet with poorly written copy in early 2020. So what we've seen is this, this cycle of slop that gets regurgitated, you know, and then future models have such a high, you know, it's, it's poisoned. Right? And you know, that's why, you know, there's all these conversations about scaling laws and, and all of these things and high quality data. But it is going to happen. There's no way around it. It's like the perfect storm. Because at least here in the U.S. well, for a couple of reasons, but number one, U.S. colleges and universities didn't know how to handle AI, so they wrote it off and they banned it, right? For two to three years. Many of them, which means aside from the enrollment cliff that was already coming, you know, universities are already at a, you know, all time low in terms of funding. So you have that, that was naturally happening, the enrollment cliff. And now job placement rates have gone down, not just because of AI and it's getting harder for entry level jobs, but also because the students coming out of most colleges are grossly unprepared because all employers want students who have experience in AI and know AI and learn AI. And well, colleges and universities banned it. So I think universities are going to go bankrupt. They're going to go out of business. Name brand ones, right? Quote, unquote, name brand ones, ones that we've all heard of. They're going to go under, not all of them, but I think many of them in five to 10 years. So I think what's going to happen is some of the more innovative universities are going to strike partnerships with big AI labs. I'm talking about the Microsoft's, the OpenAI's the Anthropics, the Googles, you know, there's probably two or three others on that list. You know, Amazon, maybe X, AI, we'll see. But it's going to happen. And here's what's, here's how it's going to work. Students and teachers, right? The value of what is created in the college classroom. It's the cognitive process, right? It's not the answers, right. It's not the, you know, that PhD paper that a student turns in, it's the red ink, it's the thinking, it's the conversations, it's the going back and forth. And I think that even though on the surface these things are going to be, you know, these partnerships are going to come off as, you know, oh, this is a research collaboration. All it is. It's for thinking, reasoning data. It is to collect high quality foreign. Moves too fast to follow. But you're expected to keep up. Otherwise your career or company might lag behind while AI native competitors leap ahead. But you don't have 10 hours a day to understand it all. That's what I do for you. But after 700 plus episodes of everyday AI, the most common questions I get is where do I start? That's why we created the Start Here series, an ongoing podcast series of more than a dozen episodes you can listen to in order. It covers the AI basics for beginners and sharpens the skills of AI champions pushing their companies forward. In the ongoing series, we explain complex trends in simple language that you can turn into action. There's three ways to jump in. Number one, go scroll back to the first one in episode 691. Number two, tap the link in your show notes at any time for the Start Here series. Or you can just go to start here series.com, which also gives you free access to our inner circle community where you can connect with other business leaders doing the same. The Start Here series will slow down the pace of AI so you can get ahead. Human generated insights and thinking because future models, they need reasoning data. And I think that unfortunately, right, you're not going to get that from the enterprise. I think where you're going to get that is you know, in the doctoral students and just, you know, in general, you know, different specialized programs. So this is actually like the data scarcity is increasingly cited as a competitive constraint for major frontier labs, one of them, if they haven't already, they're going to figure this out. I'm sure it's already been figured out. I'm sure it's already been talked about. We are going to see this and a lot of people are going to be shocked. But here's. It's, it's the reality. College is too expensive. So I think the students that participate are going to get compensated, right? Whether it's free tuition, reduced tuition, whatever. AI labs have money to burn, right? They, they are literally sitting, sitting on a stack of money so high they have to duck to fit into, to, to, to, to think, right? That's how much money they have, and they have to spend it somewhere. And colleges are going to go broke. Perfect storm. It's going to happen. Number two, venture capital firms are going to morph into venture studios. What the heck does that mean? Well, if you're not in the VC area, a lot of things are happening, right? Number one, I think a lot of VCs got burned, you know, in the early chat GPT days because they invested in, you know, millions or hundreds of millions of dollars of anything that had the word AI in it. Now they're getting a little bit smarter. But what's happened, right, traditionally, I would say in the last 10 years for most startups, right? A lot of them are, you know, SaaS adjacent or software adjacent. So think of, you know, Obviously the, from 20 years ago, you know, you had the, the Facebooks and Instagrams and, and, and all those. But you know, think of the last, you know, 10 years, 10, 15 years, you know, Ubers, Airbnb, right? All those type of things. Well, they're software. So most startups need two things. They need distribution and they need engineering. Okay, Technically, venture capital firms have distribution, right? What's happened with engineering over the past two months, right? You've had, I haven't even talked about this one on the show because it happened very recently. One of the head engineers at OpenAI admitted that a lot of the new features that went into Codex and some other internal tools that OpenAI is using, 100% written by the models, right? Claude Code, the creator of Claude Code said the same thing. Now everything, every update is written by Claude Code. 0% humans. So the engineering cost is going down to pennies on the dollar. So what is a venture capital firm to do, right? They get pitched these same types of ideas over and over and over and they're like, unless it's an AI lab. I get it. I'm not saying that. I'm saying for anything. For anything. Whoa, slipped out of my chair there. For anything. Software adjacent, VC firms have the distribution and the cost of engineering has gone down to next to nothing. So why wouldn't a venture capital firm turn into a venture studio, right? Have their own in house engineers, right? And you know, your startup, you know, you get to come work with us, you know, we help you, we have all the processes down, cuts your engineering cost by 97%, your go to market is, is 3x4x faster and they have the distribution right there. And then instead of getting, you know, I don't know, 5 to 10%, the VC funds are getting, you know, 30%, you know, because they're venture studio. If I'm being honest, I think a lot it might not happen until 2027, 2028. But if I'm in the VCC right now and I'm seeing, but they're not seeing what I'm seeing yet, right? Go talk, go talk to the people building the tools like I have and you'll start to see it. It is a crazy, weird world in terms of what's possible with software right now. And you know, whether you realize it or not, it actually has a huge impact not just on the tools that we all use on a day to day basis. Right? If you're a knowledge worker, you sit in front of a computer and you use these tools that, you know, maybe they're have been around for 20 years or for two years or whatever, but they're going to constantly change. And almost everything starts with venture capital, private equity, venture studios, etc, so I think that's going to change drastically. All right, number three. Excuse me. All right, Grounded only models become an enterprise hot item. So what the heck is that? Well, here's the actual prediction. All right, so by the end of 2026, at least one major enterprise, you know, AI Lab, will ship a grounded only mode. Here's what that means. Notebook lm. Right? Notebook LM for the most part doesn't have any major competition. And maybe it's more from a marketing perspective because I do know that there's certain ways that you can accomplish this, you know, in different tools. Right. I know that, you know, Microsoft Copilot has had something like this, but it's almost like a hidden feature that no one knows how to use. So I do think that, you know, one of the big labs is going to lead with this. Technically, Google already has it via Notebook lm. Right. But Grounded is huge. So what this means is there's essentially, you know, three different sources that a large language model will draw data from. You know, first and foremost it's looking at its training data and then depending on your setup, depending on your custom instructions, what model you're using, et cetera, it's either going to go to the web or it's going to connect to your personal data. Those are the three sources. But for so many things, right, you don't want training data and you don't want the Internet in many cases, right? You just want your company's data. Especially if your company is a large enterprise and you have a ton of data, right? So I do think that specifically, either OpenAI or Anthropics is going to release this because like I said, technically, Google already has Gemini or, sorry, Google in Gemini, right? They already have access to Notebook lm. So there's kind of ways to already have this. But either Microsoft is going to lead with this and put some marketing around it, or OpenAI or anthropic, someone's going to do it. Notebook LM continues to just meteorically rise, right? It is still on fire in terms of the millions and millions of people who are using it weekly. It's growing at a rate in terms of percentages that is faster than the other labs, right? Notebook lm, if it was its own lab, it's been growing the most, you know, month over month compared to everyone else. Because consumers and enterprises want this. They want the option to have that grounded only. I've always said, like, I know it's harder than this, but there just needs to be a grounded button, right? Use my data only, nothing else. There are certain ways you can do this, right? Like OpenAI's new, you know, they just released an updated version of their deep research that you can just do like certain websites. So that's kind of a version of that, right? Like if your company is, you know, big company and you have four or five different, you know, entities or products or companies under your umbrella. So, okay, there's a way that you can kind of do it, but I think this is going to be a de facto mode pretty soon and also just with the long context windows as well. All right, our next one, number four, Google is going to win the short game while building the long game. So here's. This one's a little harder to measure and maybe this is more taste or judgment, but I do think that Google is going to, at any point just come out and win every single benchmark again and again and again. So here's what's, here's what's happening, right? Google Gemini 3 came out in November, okay? So now the release cycles are silly. They're so fast, right? So now, depending on what you look at, but for the most part you know, whether you're looking at arena, formerly LM arena, or any third party benchmarking sites or you know, just the scientific benchmarks themselves. You know, in November, Google with Gemini 3 Pro was untouchable, every category. It is still, I'd say, the best multimodal model by far, right? Just because, you know, it's, it's the only multimodal model by default that can ingest video, right? All those other things, but a lot of benchmarks now Google's been out of it, but at any point, at any point, Google more so than Anthropic or OpenAI, right? I think that Anthropic and OpenAI, when they come out with a model, it's, it's, it's going to do well in certain benchmarks or certain, you know, kind of arena rankings. When Google comes out, it's going to win every single one, every single time. And I know that Google, whether they're, you know, going to come out with their Gemini 3 Pro GA, so they're generally available version, or if they go to a three one or a three. Five, whatever, at any minute they can snap their fingers and it's going to be done. Google is at that point now when they release a model, it is undoubtedly going to wipe the floor. But I think, yes, they're going to win the short game, but they've been building the long game silently. And here's what I mean by that, right? So for the most part I don't know the exact percentages because it's all proprietary information. All the weight and the trading data is technically private. It, not a lot of people know about it. But I will, I would guess that at least every single lab is using about at least 90% of the same training data. So it's like, okay, that's why you pay your, you know, your engineers tens of millions of dollars a year, you know, that's why you're entering into these exclusive, you know, partnerships with certain news publishers to get exclusive access to that for training data, right? But Google has always had something that no one else has had that is a gold mine, especially as models get smarter and that is YouTube, right? You can make the argument like, oh, you know, you know, X has Grok, okay, Grok has been shown year after year, study after study to be the most disinformation, heavy social network out there, right? Google has YouTube, which has an ungodly amount of hours of human video. And I think the long run, right, the long run has always been about embodied AI, right? And what helps with that, not just text, video, the understanding of the real world. I think that's why we've seen things like from Google, some of their, you know, their robotics models are really good. Came out of nowhere, you know, and now an instant, you know, tier S or you know, first tier competitor on the robotics model side. Right up there with, with, you know, what Nvidia is putting out. And then you had their, their Genie 3 world model. Absolutely nutty what that thing can do. It can, it looks like a very realistic video game that's rendered in real time, right? So Google is playing the long game and they have been playing the long game for a very long time. I think even in, you know, 2023, when I started the show, I wasn't privy to that information. You know, now I know Google is playing the long game and they've been, it's been quote unquote, long enough that they've been able to close the gap on the short game. But I still think that they're going to win because in the end it is about embodied AI and Gemini, you know, is consistently ranking at the top and they're going to come back and do it as well. All right, number five, agent to agent economies. That's going to be a thing. All right, here's the actual prediction. In 2026, enterprises will deploy internal marketplaces where agents can dynamically delegate and hire other agents. Here's what I mean. So essentially software that companies would normally make, right? And how you pay a monthly fee for a piece of software, but then you, the human still have to go do it or your team still has to use a piece of software. What think is going to happen is agents are going to hire other agents. So let's just say I work for, you know, Big Corp A. All right. In Big Corp A, we have been using, you know, Software A for 20 years and it's very expensive. But now Software A has an agent marketplace and it's more output driven. All right? And my Big Corp A, we've been ahead of the game on AI. We've been, you know, using multi agentic orchestration since, you know, late 2024. And you know, our agents are really, you know, we have more agents than human. So I know that sounds weird, but a lot of companies are in that position and those companies that are, this is what's going to happen. You are literally going to have enterprise agents out there. They're going to have budgets and they're going to go to, you know, big Software A that's now going to have an output priced agent and you're just going to have agent to agent commerce, but doing tasks that humans are going to do. And the humans at Big Software A aren't going to know that we're getting our job done. We will, but we're not going to really always understand how. Right? So this is going to like literally finance departments are going to require cost attribution per agent run, right? So you're going to have all these complex tasks that normally, you know, groups of teams do. They're essentially going to be, you know, decomposed into these subtasks that are then going to be executed by specialized agents. But it's going to be agents that are going to be negotiating with other agents because now we have these different protocols. I can't even list them, all right, but there's now a handful, right? One of the most popular ones is MCP, but there's a 2A agent to agent from Google, right? Now agents have languages that they can talk to each other, whereas you know, a year and a half ago they didn't. But now they also have their own commerce protocol as well, right? I know that sounds weird to think and like, oh no, that's you know, 2050. No, that's technically what's happening, happening at the end of 2025. But that is what is going to happen in 2026. You are going to see on the enterprise side, agent to agent economies. All right? Number six, software stocks gonna go down the. Down the tube. Down the tube. Sorry. I would especially look at ETFs, right? So these funds, right, these electronically traded funds that are groups of traditional software companies. It's already started to happen. We've seen some of these, you know, ETFs already go down like 20 plus percent year to date, which is again bonkers, right? If you don't cover, you know, if you don't pay close attention to the stock market or ETFs for something to be down 24% year to date. It's bad and it's going to continue to get worse for those companies. And one of the reasons why, right, And I'm not saying, I'm not saying that you know, someone's going to go vibe code, you know, the next, you know, Docusign or something like that. But those type of companies, public companies, huge enterprise companies, they have tech debt, right? They can't just, you know, snap their fingers tomorrow and become an AI native company. They have tech debt. So they, they have to start over or they have to spend months or quarters just to sometimes implement features that startups can implement in hours. Right. And a lot of times it will work and function better. Right. I like to think of it like houses. Right. If there's a house. But let's say software is more like dog years or 10 to be easy. Right. So a piece of legacy software that's been around for, you know, 15 years, that's like 150 year old house. Right. Good luck in the pipes on that one. Even if it's a great house. Right. A house now that's built the right way, even though some stuff is built cheap now. Right. You know, old man Jordan on my porch. No, but you'd rather have the new house. Always. Always. Because everything is changing. Right. The foundation, the type of concrete. Right. Oh, now these bricks get moldy, whatever. Right. The same thing can be said with, well, enterprise software. And it's going to get gobbled up. Like I said, I think the winners are going to be those that own the data and the distribution and the losers are those that just sell expensive UIs that agents can bypass. That's the other thing I think we're going to see in 2026 that a lot of the software, I won't say a lot, I think we're going to start to realize that even just software in general is going to be less needed because it's going to be agents interacting with, with each other again. Right. If you're able to set up guardrails and have expert driven loops of multi agent or multi agent orchestration, software becomes less and less needed. It becomes outcomes based, quantifiable, traceable, observable. When those things are happening, it's the process or the journey doesn't necessarily matter anymore. Right. We've just become so accustomed to over, oh, we the human use this piece of software to get this outcome. But we the human, okay, well now there's these models that are smarter than us software. Well, what's the point? If agents can talk to each other and bypass the software, why is the software even needed? Right. That's what we're seeing. And I think the most essential, you know, I'd hate to throw out like DocuSign, but you know, that's, that's a great example. Right. They're, they're, you know, a huge company, inexpensive and it's kind of like, you know, Comcast, Xfinity. Right. It's, it's one of those companies that everyone complains about. It's like, oh, this is the simplest piece of software. You know, you go in and sign something, you know, but companies are paying, I don't know, thousands of dollars a month, you know, to use it. And everyone's like, wait, why? Right. Things like that are going to be extremely, extremely and quickly, I think, disrupted. All right? And related to that number seven, messaging becomes the universal control. I think one of the reasons agents, I, I'll say general agents because I think narrow agents did really, really well in 2025. But I think one of the reasons that general agents didn't take off like we thought they would in 2025 is, well, the interface. You have to rework your daily processes. I think what's going to happen is how we talk as humans day to day. That's going to become the interface. Specifically, I'm talking about messaging Slack phone calls. I think that's how we're ultimately going to be communicating with agents in the same way that we would communicate with other humans. How do you talk with humans? Right. If you're not face to face, how do you talk to your co workers, to your family? You text them, right? Your co workers, you slack them. Right. You know, maybe, maybe something like WhatsApp. Right. Telegram, maybe Discord. This is how we're going to be talking to agents because now that agents kind of have these different languages that they can speak and things easily translate, we don't have to use these clunky, you know, or not clunky. Right. But we don't have to fit our messaging and our processes into a new whole. We can just do the same thing that we've always been doing. Right. Bust out your phone, text. Done. Right. Have your agent call you, pick it up. Yep. Go ahead and do that. Done. This is in the future. Literally what I said there, it's available, right. A lot of people have heard of, you know, the Open Claw or Clawbot. Right. It's a good example. You know, this is an open source agent that you can connect to a lot of your data. I think dangerous if you don't know what you're doing, but if you do know, if you do know what you're doing. And as these, the open source project becomes more and more supported and you know, with harder guardrails. I think it's a little more, it's a little safer. Right. But how do people communicate with it? Well, text, it imessage people have, you know, figured out to use 11 labs and their agent calls them what it needs, you know, their, their call on something. But I think that's what's going to happen. I think a Lot of people don't see the value in that. They're like, no, I'd rather sit down and you know, click some buttons. But why, Right, why? We've been communicating more or less the same way, you know, phone call emails, you know, since the 90s or you know, phone calls since forever, you know, emails since the 90s. And then we've, you know, messaging slack message text messaging for the last 20 years. Right. Clearly something works about it, even if you think it's inefficient. Right? It works and it's going to continue to work. But for agents. All right, next there's going to be an agent crash. Not what you're thinking, but here's the prediction. So in 2026 there is going to be a high profile autonomous agent misaction incident that triggers national regulatory scrutiny. All right? So I'm not talking about hacking or a bad prompt. I'm talking about something that is going to be like, you know, when those data leaks, like the, you know what, like the Home Depot data leak that was, you know, millions. It's, it's going to be something that is going to grab headlines. Not in the AI world. And I don't think it's going to be, you know, an agent that hacks model weights or anything like that. It's probably going to be as simple as prompt injection and tool permissions. Right. An agent's going to go rogue. Right. We already saw a small, very small micro, micro versions of this with the, the Molt book. Right. The social, the, the AI social media social network site for open claw bots. You know, we already saw some small examples of this, but it's going to be big. And I think the media coverage is just going to focus on kind of the autonomy gone wrong. And I mean we're going to see, I think the congressional hearings might not happen in 2026, but I do think it's going to get to that point. I think we are going to see a big incident. Right. This is, think of how much, you know, my going back to my old journalism lingo, how much pub, right, how much pub that it got when the, the lawyers, I believe they were from New York, you know, early on they just were copying, pasting stuff from chat GPT. They didn't look it up. You know, it hallucinated these sources, whatever. That has still been an international story for years. And that was just people, you know, copying and pasting something that wasn't true. Right? This is going to be bad. This is going to be bad. Remember When I talked about those, you know, agent to agent communications and enterprises are going to do this, even though the risks are high because the reward is higher. And sometime the risk is actually when enterprises are looking at this, they're like, well, our competitors are going to do this. And if our competitors have thousands or millions of agents working around the clock, they're going to gobble us up alive if we don't match them. So I think it is this not like a fomo, it's a rush to not get lapped because now what, like I started this show saying what individuals, teams and enterprises are capable of. It's not an exaggeration to say that we're going to see the, the smart teams. It's not doubling, it's not tripling. I mean, you're technically able to 5x your output. Right. With the same resources that you're putting in, if you know what you're doing, if your data is in a row, if you've already trained your people on AI basics, I think in theory agile companies are going to be able to 5x their outputs without doubling their inputs. Right. Keeping their inputs the same, which is going to create this, this rush to go as agentic as possible, which is going to lead to, I believe, a huge and noteworthy agent crass crash in 2026. And it's going to lead to backlash that I think the backlash is actually going to be good because it will finally, you know, nothing is going to get, you know, codified, solidified from a regulatory standpoint. But I think it is the least going to get the regulatory, you know, conversation actually going. Right. Because regardless of what your feelings are on President Trump's stance on AI, he's made it the wild west, which a lot of people really like. Right. But I think what that's going to lead to eventually is it's going to lead for an opportunity like an agent crash, and that's what we're going to see. All right, number nine, very similar here. Shadow it is going to trigger a Fortune 500 data breach. So here's the difference between what I'm calling the agent crash in prediction eight versus the shadow AI data breach in prediction nine and number eight, the agent crash that's going to be fully signed off on. Right. This is going to be, no, we intentionally put this agent out in the wild and it went terribly wrong. Right. And this is a brand new technology and it, you know, didn't go how we wanted it to. Right. Unfortunately, if you think back to the earlier days of aviation, there's a lot more plane crashes in the earlier days of aviation because it was brand new technology. It changed how people live their lives. And there was a rush to, you know, make that technology better and to give it access to more people. So I think with number nine here with shadow IT triggering a Fortune 500 data breach. Shadow it is unauthorized, right? When the agent crash is going to be an authorized agent that just goes off the rails or people enterprises didn't build the right rails here. I think by the end of the year I'm calling it, it's going to be a Fortune 500 and it's going to be a public data leak. And again, not to keep looking at like mult book, right? The example of the, you know, Claude bot turned malt bot turned open claw bots, their own social media network. But a research company came out and said, oh wait, there's millions of, you know, exposed data entries here. I think we're going to see something very similar, but from a Fortune 500 company, which is what's going to make it kind of noteworthy. And here's why companies aren't training their people on AI, right? That's number one. Number two, they're locking down access, right? Those people, I think for the most part, you know, enterprise companies here in the US they have access to Microsoft Copilot. I think smart teams have been choosing their AI operating system. But for everyone else, you know, they're like, wait, I don't have Copilot where I need it or I don't have access to this. So, all right, on my second computer, or I'm just going to open a, you know, incognito window here and just, you know, use Chat GPT or whatever or use, you know, even worse, right? I've seen enterprise leaders, like, ask me. I get messages all the time and they're like, oh, Jordan, what do you think about this piece of software? And I'm like, no, it's garbage. Don't use this, right? It's all these like people literally in the enterprise, people are looking at like rappers still, right? Not like, you know, Snoop and Eminem like rappers with W, right? Just these, these pieces of software that pop up, you know, and it's like, this could be one dude building this and. And you're, you know, thinking about moving your company over to this piece of software versus like chat GPT or Claude, like, no, don't do that, right? But that's what's also going to lead to a data breach is people don't know any better, right? They're going to say, oh, this is like chat gbt, but it's for construction. Oh great, I'm in construction. I'm going to go ahead and paste all my stuff in here. No, you know, don't. It is going to lead to a data breach and that is, well, if this is also multifactorial, it's training and education, number one, but then it's, it's access. Right. Companies aren't giving their people access to the right models that they need. So yes, this is going to, it's, I think it's going to reach a point where even copy and paste might get locked down. At some companies, I think it might get that bad, right? Not permanently, but I think, you know, companies might get to that point where they're going to lock down copy and paste because of shadow it. Right. So many, you know, people here, knowledge workers in the US are working remote or hybrid companies, you know, smart companies, know the shadow AI that's becoming a problem and companies. Well, I think a huge Fortune 500 data breach and a lot of people are going to have a freak whiplash reaction and make some crazy, some crazy decisions based on that. All right, number 10 agent audit logs are going to be, become mandatory because of. Yeah, I put these ones in order for a reason. When you have something like an agent crash that's going to happen and become a national headline and you have a shadow AI that is definitely going to trigger a Fortune 500 breach. Yeah. All of a sudden, especially in regulatory, regulated sectors, finance, legal, accounting, etc. They're going to require audited logs of every single action an agent takes. Right. So as an example, if, oh, if an agent denies a loan now. Well, you need a log of exactly why there's no more black boxes, there's no more, you know, opaqueness in your AI processes. So I think as business leaders, and this is why this is part prediction, part roadmap, Right. This is going to come especially if you work in anything that, you know, where discrimination can be determined. You know, finances, loans, healthcare, etc, and those are the industries now obviously, you know, trying to take advantage of AI. You have to be thinking ahead. You have to say, okay, how are we going to be able to audit our logs? Right. Are we able to see, do all of our agents have id? Who's, who's observing who's, you know, going back and, and tracing all of this? Right. How often are our experts going in and reworking our processes? Governance all of a sudden has to become embedded in the architecture and it's no longer just the competitive nature, right. All of a sudden these buzzwords that we've been throwing out to sound like good, smart AI people like, like governance and, and guardrails and responsible AI, all of a sudden those become more than buzzwords, they become lifelines for your company. Right. Otherwise, you know, you. I'm not going to go this far and say that there's going to be companies that are going to be shut down. Maybe that might help happen in the health care space, but it could definitely happen. All right, number 11, the Skill Marketplace becomes the largest agent risk surface. So here's the prediction. By late 2026, at least one major Asian ecosystem experiences a high profile security breach via malicious skill distribution. So yeah, there was actually a study again, I know, I keep going back to the whole. Claude, sorry. Openclaw. Right. So a researcher came out and said the most downloaded skill for openclaw. Right. Again, this is an open source project. I think it's great, it's fantastic. I still think it's not quite safe enough for the majority of our listeners, which is why I haven't devoted a lot of time, right. But the number one skill on kind of the Skill Marketplace was malware, right. And everyone's just rushing to do it. I do think skill marketplaces are going to take off, right. When you start looking at, you know, even as you start using programs like OpenAI's Codecs more, you know, which now supports the skills. As you start looking at, you know, Anthropic's Claude code, Claude Cowork. You know, there's skill support now across, you know, some of the Google Gemini products as well. And then essentially skills are these their folders and markdown files, right? So do you know how easy it is with a large language model to be like, hey, here's 10 of the top skills, make me 10,000 more. But inject malware into all of them and then make them all industry specific because people are going to love those, right? And they're going to over, you know, they're going to like, oh, this is just like that popular one, but for construction, yay. I'm going to use this skill for Claude Cowork. Oh, now all of a sudden, Claude Cowork, which if you don't know what you're doing and you just give it access to your entire computer and it has terminal access, it can read and write your files and all of a sudden you just blindly give it a skill that is injected with malware, it could go ahead and take every single file on its desk, on your desktop and upload it to a public server and goodbye. Everything that you have on your computer, right, it is now public in the hands of someone else and you may not even know it. We are going to see something like that. I don't know how big this one is going to be, but it is going to be a huge risk. So it's kind of like downloading a virus from an app store, but it's for your corporate agent that is out there autonomously making decisions for you. That's the biggest thing, right? Think of this happens all the time, right? I'm sure your parents or in laws or someone, right? If they're in their 50s, 60s, 70s, you can probably relate to this. Someone's had a virus on their computer, right? The difference is that person probably knows and they're like, oh, man, virus. And they turn their computer off and hopefully nothing bad happens, right? When an autonomous agent, if you give it a skill and you don't fully know what's in that skill, that tells the agent what to do. And most people want their agents working 247 and the more and more access that you give to that agent, things could go bad. Yeah. So like I said, that was open claws, claw hub, that there was already some hacks on there. But I think it's just going to be malicious skills that are disguised as productivity tools and they're going to silently just ruin lives, ruin departments, ruin careers, ruin companies. It's going to happen, right? That's why this is part prediction, part roadmap, y'. All. I am telling you, you have to be aware, like, I'm not going to go off into, like, things are getting weird. I always try to have a little bit of a positive, you know, spin at the end of the day, even though I do think AI is going to take more jobs than it will create, things are going to get very weird in 2026 in terms of capabilities, in terms of human purpose, right? You're going to see some AI that's going to do something that you've been doing for 30 years, way better, way faster than you, and you're going to be like, what's my purpose here? I'm not going to get into that. But when it comes to safety of, you know, if you're a decision maker, if you're a department head, if you're an entrepreneur, you have to be safe. You really have to understand what you're doing. Don't rush in, right, to, you know, give your Claude code or give Your codex, all these skills, especially if it is a computer that you're giving it access to different file folders, different structures, right? It has terminal access. You have to be extremely careful. All right, two more number 12. 12 hour autonomous task horizon. So I am saying the meter metric. All right, so we've talked about this a little bit on the show before. So meter is essentially, it's M A T R. They're a non profit and they essentially measure how long large language models can do long work tasks that humans would do at a certain pass rate level, right? And it's been growing significantly. Like a year and a half ago it was like 45 minutes. Now it's multiple hours. But I think by the end of this year I think we're going to see at least 12 hour, right? 12 hour at that 50, 50% success rate, maybe even 24 hours, right? So this means that an agent can successfully complete work more than 50% of the time. A task that would take a human more than 12 hours. Do you know how crazy that is? Just the models that we use every day, right? I'll probably do a show on meter because there's also a like a 90% pass rate. This is a little bit different than GDP val, which we're going to be talking about on tomorrow's show, right? But this is the ability for a out of the box model to do 12 hours of work. It's going to happen this year because it's already getting close, right? It's already up there in The I believe, 7, 8, 9 hour range with some of the more recent models. But I think that AI is going to move from a tool to a functional employee, right? Maybe a junior level employee in 2026. But undoubtedly it is going to be able to long tasks, these long task horizon times and the capabilities are just going to increase. So an example, right. I think we've been thinking of AI as giving it a tasks, right? Giving it a task or a series of tasks. This is giving it hard projects, right? That would take someone a day or a couple of days or it might take a team an afternoon, right? So an example, right? Maybe researching competitors, building a competitive analysis website scraper that scrapes your competitors, you know, pricing page updates over the course of a year and then it debugs the software and then it sets up a database with everything that it scrapes and then it writes a memo based on everything it found, right? If I were to tell you that you probably need a small team and they probably, you know, spend a good, you Know, day or three on something like that. An agent can go out and do that now, right? Go do research, go scrape some data, go build a tool with all that stuff that you did, go run the tool. If it breaks, go fix it, and then, you know, write me a couple briefs and, you know, make some decisions based on that. And that's all you tell it. And it's going to go out and do all those things autonomously, right, for many, many hours at a time. All right, Last but not least, number 13, Agent Ops. That's going to become a formal enterprise function. So here is the. The actual prediction. By late 2026, large enterprises are going to establish dedicated Agent Ops teams responsible for monitoring and maintaining AI systems. So I actually saw some, some tweets about this recently. I think it was actually from Greg Brockman, president of OpenAI, said something along the lines of, like, not that he had fomo, but it was almost like if his agents weren't running 24 7, he felt like behind and felt like he's missing out. And I think that's going to be for enterprises on the cutting edge, right? Companies, huge companies that have gotten AI, right? It's going to be like that. They're going to have, if they don't already, thousands or tens of thousands of agents that are running 24 7, which means a new breed of work is going to come. Yes. AI is going to create new jobs that didn't exist. And I think a lot of that is going to be Agent Ops, right? I'll even get crazier. I would say by 2030, many of us are going to be in some form of Agent Ops in the same way right now, knowledge workers, you know, we're in some form of. Well, we all just use the Internet and software and we make decisions, right? It's going to be the same thing. All of our. Right. This is more in like the 2000 and 30s. We're all going to be an Agent Ops, just different types, right? We're going to be, you know, front end, back end, I call it. You know, we are, we are becoming the buns for the agent sandwich. We give it right directions and context on the front end and then feedback and kind of iterations on the back end. But the agent does the majority of the work. It is the, it is the meat, the cheese and the condiments in the middle, right? But I think that agent uptime is going to be one of the most important KPIs for a lot of enterprises. And, you know, drift, you got to Be able to detect drift. Right. Prompt changes must be version control. You know, incident responses must include agent behavior. Right. Agent Ops is going to become as normal as DevOps and probably more commonplace in a couple of years. But in 2026 agent op teams, it's going to be a very normal thing. Right? It is the full industrialization of AI. Everything from audit logs and change control meetings, everything. Right. You are literally going to have large teams of people that all they're doing is, you know, investigating the operations of agents and coming up with ways to optimize them, improve them and scale them. That's all they're doing. Large teams of people just watching dozens, hundreds, thousands of agents and making sure they're operating right. That's it. That's all these teams are going to be doing. You know, going beyond, you know, simple orchestration. This is like large devop but with Agent Ops. All right, so that is a wrap for today's show. Let me go ahead and again give you the quick, quick 2 second rundown here in case you fell asleep halfway. So number one, an AI lab essentially is gonna not aqua hire, but kind of aqua hire universities for student at us. Two, VC firms are gonna morph into venture studios. Three, grounded only mode is gonna become a huge thing in enterprise. Four, Google is going to win the short term game while building the long term game. Five, internal agent to agent economies take hold. Six, software stocks and software etf' messaging becomes the universal agent remote control interface. Eight, Agent crash going to become a national headline related to number nine, shadow AI triggers a Fortune 500 data breach. Also related to 10, agent audit logs become a mandatory thing in regulated sectors. 11, skills marketplace become the largest agent risk surface. 12, I think we're going to see 12 hour autonomous task horizon becoming the new metric that companies go after. And 13, Agent Ops become a formal enterprise function. All right, that is a wrap for volume one. All right, so make sure you join us Tomorrow for volume two of our 2026 AI predictions and roadmap series. I hope this is helpful. And like I said, go find Today's show on LinkedIn. Go repost it. You are going to want this exclusive bonus guide that we put together. I didn't even get to half of my notes for these. Thirteen, I'm handing it all over to you. Right? This is one of those things where I'm like, yeah, I could charge money for it. No. All right, if you find any support in the show, if you found any help over the course of 2025, take, you know, 10 seconds 20 seconds out of your day to go, you know, find that LinkedIn live stream, just click it. If you're listening on the podcast, like many of you are, go repost this and I'm going to send this to you. All right? So thank you for that. Thank you for tuning in. Hope to see you back tomorrow and every day for more Everyday AI. Thanks, y'. All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.