AI isn’t just shaping business anymore, it’s rewriting the economy.
In this week’s episode of The Artificial Intelligence Show, Paul Roetzer and Mike Kaput connect the dots on the rapid rise of the “AI economy,” from mass corporate restructuring and three-day workweek predictions to research forecasting trillions in productivity gains. They explore how people are really using ChatGPT, the future of AI-native organizations, and the latest breakthroughs—from Meta’s wearable launches to reasoning models acing elite coding competitions.
Listen or watch below—and see below for show notes and the transcript.
00:00:00 — Intro
00:08:48 — The AI Economy
00:31:03 — How People Use ChatGPT
00:38:49 — The Future of Organizations
00:47:49 — Meta Ray-Ban Display Glasses Launch
00:51:33 — Gemini, ChatGPT Achieve ICPC Gold-Level
00:54:55 — How Americans View AI
00:59:36 — Ongoing AI Lawsuits
01:03:02 — AI and Voice
01:06:42 — AI Product and Funding Updates
The AI Economy
AI’s economic impact is starting to show in everything from central bank policy to corporate restructures.
The Fed just cut interest rates for the first time since 2024, citing growing uncertainty around unemployment and inflation. Meanwhile, AI-driven disruption is ramping up across industries.
Fiverr laid off 30% of its workforce to become an “AI-first” company, streamlining management and restructuring around leaner, faster teams. Zoom’s CEO predicts a shift to a 3-day workweek, arguing that AI will soon make five days unnecessary.
Anthropic’s latest Economic Index notes that AI capabilities are already poised to transform productivity but adoption is uneven, often clustered by income and geography.
DeepMind’s Virtual Agent Economies paper goes further, envisioning an ecosystem where autonomous AIs create value with minimal human input.
And in Washington, 40 top economists are calling on the Labor Department to track AI’s effect on jobs in real time.
How People Use ChatGPT
OpenAI has released what it calls the “largest study to date of how people are using ChatGPT.”
In it, researchers looked at 1.5 million ChatGPT user messages between May 2024 and June 2025 and used AI to label what each message was being used for.
They found that most usage, over 70%, is now non-work-related. People use it for writing, information-seeking, tutoring, advice, translation, even creative brainstorming. The three biggest categories of usage, what the researchers call Practical Guidance, Writing, and Seeking Information, account for nearly 80% of all usage.
However, work usage is rising too, especially among educated users in professional roles.
Writing tasks dominate, but AI is increasingly used as a decision-support tool, not just a task executor. The study classifies most work use as Doing (like generating content) or Asking, where ChatGPT acts more like an advisor.
The biggest surprise? Only 4% of messages are about coding, and even fewer are about relationships.
The Future of Organizations
Replit CEO Amjad Masad says the future of business is being rewritten.
In a recent presentation at Y Combinator’s AI Startup School event, Masad envisions a world where anyone can build software just by speaking, with no coding required and AI agents handling the heavy lifting.
In this future, the role of the engineer shifts from technical execution to high-leverage thinking.
(Replit’s own Agent 3 is already working autonomously for multiple hours writing code, deploying software, and testing features.)
As a result, Masad predicts traditional SaaS will go to zero. Instead, every individual will spin up personalized apps or agents on demand. This will also, he says, have widespread effects on how companies are structured at a fundamental level.
Eventually, he sees organizational hierarchies being replaced by fluid networks of generalists collaborating with autonomous tools.
This episode is brought to you by AI Academy by SmarterX.
AI Academy is your gateway to personalized AI learning for professionals and teams. Discover our new on-demand courses, live classes, certifications, and a smarter way to master AI. You can get $100 off either an individual purchase or a membership by using code POD100 when you go to academy.smarterx.ai.
This week’s episode is brought to you by MAICON, our 6th annual Marketing AI Conference, happening in Cleveland, Oct. 14-16. The code POD100 saves $100 on all pass types.
For more information on MAICON and to register for this year’s conference, visit www.MAICON.ai.
Disclaimer: This transcription was written by AI, thanks to Descript, and has not been edited for content.
[00:00:00] Paul Roetzer: Years ago, I would say AI was eventually gonna be the operating system of society and business. And I feel like that's what's actually starting to happen, is like it, it really is just everything operates on top of the AI layer. Welcome to the Artificial Intelligence Show, the podcast that helps your business grow smarter by making AI approachable and actionable.
[00:00:18] My name is Paul Rader. I'm the founder and CEO of SmarterX and Marketing AI Institute, and I'm your host. Each week I'm joined by my co-host and marketing AI Institute Chief Content Officer Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career.
[00:00:40] Join us as we accelerate AI literacy for all.
[00:00:47] Welcome to episode 168 of the Artificial Intelligence Show. I'm your host, Paul Roetzer, along with my co-host Mike Kaput. it is, we're recording this on September 22nd, 9:00 AM it sounds like there's some stuff going on this week, so. [00:01:00] Again, as always, if something happens after this, we'll get to it next week.
[00:01:03] There's been a lot of chatter about some releases coming this week, and even Sam Altman confirming they're definitely entering a release season at openAI's, so lots to do. I don't know you Mike, but I got, put Notebook LM on Overdrive this weekend getting ready for this podcast. There was, yeah, a lot of like in-depth research papers, specifically related to the economy.
[00:01:27] so that's actually gonna be the opening topic. We're gonna kind of dive into some of those papers, but I , I mean there was no less than like five papers that I went through on Sunday night. Yeah. so yeah, notebook, LM was my friend. Gimme executive summaries, extracting key findings. if you don't use Notebook, for that stuff, it's awesome.
[00:01:45] And even separately, I've been using it for like, like flashcards and quizzes, like experimenting with some of those other features. So yeah, this wasn't intended to be a, a promotion for Notebook lm. I needed it to get ready for today. All right, so today's episode is brought to us by AI Academy, by [00:02:00] SmarterX.
[00:02:00] We've been talking a lot about this lately. Each episode we try and give you a little preview of some of the things we're doing. So today we wanted to touch on our Gen AI app series. This is brand new as of August 19th, and we rolled out this next generation of AI Academy. So the Gen AI app series is weekly, product and feature reviews.
[00:02:18] So we do like these 15 to 20 minute reviews. They're meant to be shorts. something you can pick up real quick and kind of learn about a feature or a product. So these drop every Friday morning. Claire on our team and Mike on our team, you know, Mike put, who you're familiar with, have been creating those.
[00:02:33] We're actually expanding, our plans for that. I'd love to do more of these. I think they're incredibly valuable. I love them. so I thought I'd just let Mike give a little overview here of what we've dropped so far. So, again, these are in our AI Academy by SmarterX. You can buy individual course series or you can buy an AI mastery membership, which is a 12 month me.
[00:02:54] These are AI Mastery member only, things that, that we feature. So, Mike, give us a quick [00:03:00] rundown of what we've released so far and maybe a little preview of what's coming up in the Gen AI app series.
[00:03:04] Mike Kaput: Yeah, for sure, Paul. So every week since launch, we've released, one of the Gen AI app series. We've dropped one on GPT five one on Gemini Deep Research and Chat, GPT Deep Research.
[00:03:18] there's one on Google Notebook, lm, which is relevant to what we just talked about. Claire did one on Google's Nano, nano, that's the kind of nickname of their, their image editor. And then we've also got one that just came out on GPTs, in general within chat JPT and how to get the most out of those.
[00:03:35] And what's really cool is we've got a ton of awesome video image and audio tools coming up. Mostly courtesy of Claire, like we've got in the next few weeks here, Google Image in four. Google VO three Hagen, openAI's Image Generator, and then video generation from PIKA and Audio Generation from Suno. So those are all once a week coming up here.
[00:03:59] Paul Roetzer: Yeah, it's, [00:04:00] it's, it's a fun series. And so this was like a big part of the shift in AI Academy was this regular creation of content. It's not just a collection of courses that, you know, over time, you know, we have to kind of rerecord. So the plan is for the major courses, we'll redo those probably every 12 months, but this plus our A Academy Live keeps the content fresh every week, in a really digestible way.
[00:04:21] So you can go to academy dot SmarterX dot ai, learn more about that. And I think POD 100, mike, I believe is for the AI Mastery membership, you can get a hundred dollars off the mastery membership. Alright. And then, it's also brought to us by MAICON. This is coming up really fast. Anyone who follows along on the podcast knows we are now, I think when this drops, we're 21 days out from MAICON.
[00:04:42] I've been teasing the main stage. We haven't really made these announcements. If you're subscribed to our emails, you've probably seen the announcements, but, there's some of these that actually got signed over the weekend. So this is kind of hot off the presses stuff. So this is our sixth annual marketing conference, MAICON.
[00:04:58] So October 14th to the [00:05:00] 16th in Cleveland. Dozens of sessions. I think we have over 40 speakers. you can go to MAICON.ai. That's MA ico N do AI to learn more for the main stage. This is the thing that I'm mainly responsible for. I used to oversee like the entire agenda. Now Tracy and Ashley and our team take the lead on kind of building out a lot of that.
[00:05:19] and my focus is on sort of these 10 main sessions for the main stage, envisioning what those should be and then, and kind of recruiting those speaker. So for these sessions, because it's everyone all together, so we'll, 1500 plus this year at the event, and everybody will be in the main room together.
[00:05:35] So I like to broaden the focus to more macro issues such as ai, advancement on the future of work and jobs, the economy, educational systems, society, science. so that's the gist of these is that they're, they really kind of broaden your mind about the impact of AI and what's possible. So I'm gonna give you a quick rundown of what those are.
[00:05:53] We have one more to announce that's not included in this, but this is nine of the 10. So I'm leading off with the move [00:06:00] 37 moment for knowledge workers. I'm very excited about that. I actually started really diving into that this past weekend, that's followed by becoming an AI driven leader with Jeff Woods, who's the CEO and founder of AI Leadership and the author of the AI Driven Leader, which is an awesome book.
[00:06:14] We have Thrive skills for the age of AI with Pat Yang pr, Y pr, chief Academic Officer for code.org. That's gonna be incredible. That's about the impact on AI and education. Empowering teams in the age of ai, how McDonald's is building an AI ready workforce with Mich, Michelle Ganley, chief Data and analytics officer for McDonald's.
[00:06:35] We have the future of AI marketing as Silicon Valley insider's perspective with Jeremiah o Yang, a VC investor, founder of Lama Lounge events and Blitzscaling Ventures. one I'm extremely excited about the human side of ai inside the leading AI labs was Shia Ma, who's a software engineer at Google DeepMind and Angela Fam, who's a UX content design lead manager of Gen AI at Meta.
[00:06:59] And that's gonna be [00:07:00] moderated by a former communications executive for Meta Google and Anthropic, who's kind of seen it from the inside. I'm really excited about that. We should be able to announce, her as well this week. Then we have backstage with big tech AI truce from the front lines of innovation with Alex Treitz.
[00:07:14] he's a reporter and founder of big technology, author of Always Day one and A-C-N-B-C contributor. Mike is gonna be moderating that one. That's gonna be awesome. Alex has interviewed all of the big names in ai and so we're gonna kind of hear what Alex is seeing and hearing from the front lines of that.
[00:07:31] one we just locked in last week that I'm extremely excited about. This was one of the ones I targeted for the main stage. from the beginning I wanted like AI and video or film. So we have the rise of the AI filmmaker from concept to 250 million views in 60 days with, PJA Satur, the CEO and AI filmmaker and genre.ai.
[00:07:52] This is gonna be incredible. He did the call see ad for the NBA finals. it is gonna be awesome. He's gonna do a walkthrough of like how to build an AI film. And then [00:08:00] the closing keynote is reimagining what's possible, a conversation with Dr. Brian Keating, who is a cosmologist professor, bestselling author of Losing the Nobel Prize and host of Into the Impossible Podcast.
[00:08:12] And I'm gonna actually lead that conversation. that is, I don't know which one I can, I can't even pick which one I'm most excited about like that. I wanna watch all of them, but I can't wait for that conversation with Dr. Keating. I love the podcast and I've followed his work for years, so it's, it's gonna be incredible.
[00:08:28] So that's the lineup for the main stage. Like I said, there's one more to go, but if you go to the Make con.ai, you can see the full agenda and check out all the speakers. yeah, it's gonna be awesome. So make on.ai and same thing, you can use POD 100 for a hundred dollars off registration there. We got three weeks to go, so we'd love to see you in Cleveland.
[00:08:48] Mike Kaput: Alright, Paul, we have a packed agenda this week, so we are going to kick things off with. Talking about AI and the economy because AI's economic impact is [00:09:00] starting to show in maybe everything from central bank policy to corporate restructuring. So a few kind of interwoven stories here. The Fed just cut interest rates for the first time since 2024, citing growing uncertainty around unemployment and inflation.
[00:09:17] And meanwhile, AI driven disruption is ramping up across industries. We had a story this past week from Fiverr, whose CEO we've talked about in the past that they laid off 30% of their workforce to become what they call an AI first company. They're streamlining management and restructuring around leaner, faster teams and some other news.
[00:09:37] Zoom's, CEO predicted that we're going to shift to a three day work week because AI will soon make five days of work unnecessary. At the same time, anthropics latest economic index says that AI capabilities are poised to transform productivity, but adoption right now is uneven. It's often clustered by income and geography.
[00:09:59] DeepMind [00:10:00] released a virtual agent Economies paper, which envisions an ecosystem where autonomous ais create value with minimal human input. And on top of it, all 40 top economists are calling on the labor department in the US to track AI's effect on jobs in real time. Now, Paul, I'm gonna turn this over to you here because you dedicated quite a bit of space and thought in your weekly exec AI insider newsletter from SmarterX to unpacking all these different signals we're seeing right now.
[00:10:33] Like what dots are you connecting here that we should be paying attention to as business leaders and professionals?
[00:10:40] Paul Roetzer: This was a. It was interesting. So the background here, if people aren't familiar with kind of our process, so throughout the week, we keep a sandbox of links and resources, research papers, articles, tweets, anything that we could potentially talk about on the show.
[00:10:56] And then, I go through on, usually it's, I try and do it [00:11:00] on Fridays. It never works. So usually on Saturday mornings, I wake up like 6:37 AM and I write the newsletter. So it usually takes about two hours. Mike, thanks to Mike, he, he gets it ready to go and sends it for us. and so at about eight 30 on Saturday morning, I message Mike, I was like, this is not gonna be ready at nine o'clock.
[00:11:19] Like, I'm sorry. So what happened was, I, we had close to 50 links in this week's sandbox, and as I'm going through Saturday morning to figure out what the editorial is gonna be, so I always lead off with an editorial. It's, I don't know, it's usually like 500 words or something like that. And then I just do a kind of a preview of some of like, usually it's about five to seven, topics that we're gonna talk about in the podcast this week.
[00:11:42] And so as I was going through these 50 or so links, I just saw this trend, like, you know, jumps out at, on the page of AI and the economy because a bunch of these were all related to it. And it was, it was across everything. It was as you were outlining Mike, it was job stuff, it was interest [00:12:00] rates that they weren't saying ai, but that was kind of the undertone of what was going on there.
[00:12:04] it was research papers. It was, I don't know, it was just like this crazy collection of everything. You're like, all of this happened and like a seven day span, like something is going on bigger here from an economic perspective. And so that's what I ended up writing the editorial about. It just took way longer than than normal to write it.
[00:12:22] So I think I finally got Mike the newsletter, like 10:00 AM or something like that. So I'm just gonna kind of wander through these a little bit and maybe add a little context as we're going because it, it's even hard. Honestly to like figure out the full story here yet. But what I think everyone will, will realize is we are entering this, this definite trend where there's way more conversation about AI's impact on the economy.
[00:12:47] so Mike, you let off with the Fiverr thing, you know, so Fiverr, if you're not familiar, they operate a self-service digital marketplace where freelancers can connect with businesses, or individuals requiring digital services like graphic [00:13:00] design, editing and programming. So they sort of have an inside track on kind of what people are hiring people for.
[00:13:04] So the quote from the, I think this was the CEO. Yeah. We are launching a transformation for Fiverr to turn Fiverr into an AI first company that's leaner faster with a modern AI focused tech infrastructure, a smaller team, each with substantially greater productivity and far fewer management layers. So that was in a letter to employees that was.
[00:13:24] shared and then they commented and basic, I mean, they, they didn't deny this at all. It was like, yes, we are getting smaller. AI first in their world means fewer people, prioritizing tech over people, which is what I've always said. It's like there's nothing wrong with AI first movement at all. but the implications are, it, it's tech over people and that's exactly what's gonna happen.
[00:13:42] So even things like this could end up giving that perception to AI first if it, if it wasn't already there. so Google DeepMind, this was a, this was a dense one. This was a notebook LM research paper for me for sure. So Virtual Agent Economies was the name of this paper. So [00:14:00] again, like if you really into this stuff, go read these things individually.
[00:14:04] if you just want a little bit better, drop 'em into Notebook lm and just ask for kind of like simple or summaries. 'cause some of these get kind of dense. but so this paper starts off with current tech. Technological trajectories could potentially lead to a global economy in which autonomous AI agents.
[00:14:21] Interact with one another to generate economic value independently of human labor. The paper concludes, certain domains may always require active human decision making for a variety of reasons. For example, human preference, culture risk sensitivity, et cetera. However, the rapid increase in AI agent performance, coupled with the development of scalable AI safety oversight frameworks and guardrails, is likely to result in an increasing number of use cases for autonomous agents.
[00:14:50] Autonomous or semi-autonomous AI agents may potentially be able to achieve more faster adding substantial value to society. This will not come without significant [00:15:00] challenges requiring alignment and coordination, not only of individual agents, but perhaps more importantly, the alignment and coordination of agent networks across various scales.
[00:15:09] Okay. I'm gonna, I'm gonna pause on this one, Mike, because, we'll get into the next in a second, but there's a couple of things going on here. So one is. The labs themselves see a near future where autonomy becomes far more reliable. Mm. So right now, AI agents we have today aren't always super reliable.
[00:15:28] We've talked about, like with agent three recently, we talked about like the runtime and how they can, most of that is related to coding and software development basically. So this isn't like out into the real world of all the other industries doing legal work and marketing work and sales work and customer success work, things like that.
[00:15:44] But what they're seeing is the increased capabilities on coding that they're then translating and say, okay, this is gonna be a problem. Like as these agents become more reliable. The other thing it's addressing right up front is these agents interacting with each other. So it's no longer like human [00:16:00] sending its agent to go do something.
[00:16:01] Agent does the thing, comes back, and the human's like super involved. It's human maybe tells the agent, here's what I wanna achieve, here's what I want the outcome to be, here's what I wanna purchase, whatever. And then the agent goes and does something, but it starts interacting with the brand's, agents, and the humans just like get extracted from the loop except the very beginning and very end.
[00:16:19] And this can be in commerce, it can be information consumption, it can be anything. And so this is very transformative and also potentially very disruptive. And so I think what we're starting to see, the reason all these articles we're kind of talking about research reports, we're talking about are coming at the same time, is all the labs are seeing the same trajectory of agen capabilities and they're now investing the resources to figure out what does this all mean.
[00:16:45] Mike, before I move on to the next one, any thoughts from you on on, on those so far?
[00:16:49] Mike Kaput: I was going to ask you about that because my distinct impression, which I was going to mention is, look, maybe I'm just superstitious, but all these stories coming at the same time [00:17:00] just really made me feel like, what do they know that I don't know, or what conversations have been had behind closed doors?
[00:17:07] It does feel like this is in response to something. And so the AGI agentic piece of it makes perfect sense.
[00:17:13] Paul Roetzer: Yeah. And that's why even like the subject line for the newsletter, you know, I was trying to think like what, what I even say here. and honestly like, I just went with the simple, the AI economy.
[00:17:22] . Because that was what started jumping out to me is like, that's what this is. Like, we're now seeing the development of this entirely AI powered. Like it's, I I used to talk about years ago, I would say AI was eventually gonna be the operating system of society and business. And I feel like that's what's actually starting to happen, is like, it, it really is just everything operates on top of the AI layer.
[00:17:42] Mm. so, so the next one was Anth. this was a article from the information it said how Anthropic and openAI's are developing AI coworkers. This is a good read for people. the information we, we often cite as a source. I wanna say it's like $300 a year for a subscription to the information. If you love this [00:18:00] stuff, it's great.
[00:18:01] $300 to spend. they have incredible, articles and sources. This one becomes more tangible. I found this one really intriguing. So it said The AI models are being taught how to use everything from Salesforce customer relationship software to Zenex desks, customer support, and Cerner's health records app.
[00:18:19] The idea is to teach AI how to handle some of the complicated tasks white collar workers do. This training isn't like anything AI models have done before. Researchers give the AI fake versions of apps to play around with and hire specialists in various subjects to show the models how to use the apps.
[00:18:36] Such techniques aren't cheap. Anthropic leaders, for instance, have privately discussed spending 1 billion over the next year on these cloned enterprise apps, otherwise known as reinforcement learning environments or gyms. openAI's plans to spend around a billion this year in data related costs, which includes paying for human experts and reinforcement learning, or RL gyms.
[00:18:58] Rising to 8 billion [00:19:00] in 2030. one reason for the cost, the increase is the cost of hiring human experts. So they cited a company called Label Box, one of a half dozen prominent firms that provide expert experts to AI developers like openAI's. So these are people that are the companies that hire experts in different industries to train the models, to do the jobs of the people in those industries.
[00:19:22] So again, zoom out, people who, who don't think that these labs and these VC funded startups want to replace humans, you, you're completely missing what's happening. They absolutely intend for them to do that. And the way you do that is you train them to use the software we all use and to have the domain expertise we all have.
[00:19:45] And so the cost of that. So they're, they're literally hiring like mathematicians and consultants and doctors and like to train these models to do what they do in the software that they do it in. So. Turing, a firm that helps companies such as Google [00:20:00] Anthropic improve their model says that it has built more than 1000 RL gyms, including copies of Airbnb, Zendesk, and Microsoft Excel.
[00:20:08] the, in the recent months, Turing's rivals including Scale Surge mer, which stick with that name for a minute, we're come back to that one. And Invisible Technologies have also begun to offer RL environment services, including providing human experts who come up with tasks to run in these apps. And then the last piece I'll say here is, as the AI miles have gotten, better data labeling firms have gone through hiring student from hiring students, pursuing master's degrees and doctorates to working with professionals with multiple years of experience in niche fields where they're actually doing complete real world tasks using specific applications.
[00:20:43] So the, if you weren't aware this was happening, here you go. Now, if you want to know how significant this is and how it's shifting the economy, Mecker one of the companies I just mentioned that does this. Their CEO. Brendan Foodie, tweeted an article last [00:21:00] week called The Economy will Become an RL Environment, machine Reinforcement Learning the Gyms Basically.
[00:21:05] So building on this concept, he also tweeted Mercer scaled from 1 million to 500 million revenue run rate in the last 17 months. According to him, quote, making it the fastest growing company of all time. Mecker is an AI power talent acquisition platform that automates the hiring process by matching candidates with job opportunities, facilitating recruitment for companies.
[00:21:27] One trend he says is driving the meter of growth. Is the economy becoming this R Environment machine? He says, reinforcement learning is becoming so effective that agents can hill climb any benchmark, meaning pursue any benchmark, but humans need to define the rewards to automate everything. While everyone fears job loss, we're creating a new category of knowledge work faster than any time in history.
[00:21:49] The future of work, work converge on training agents. We're paying out over this, still quoting him. We're paying out over 1 million per day to people in our marketplace and hiring [00:22:00] experts rapidly across nearly every domain. Software engineers, doctors, lawyers, consultants, bankers, and many more, meaning million dollars per day for people like you and me to train models to do what we do.
[00:22:12] So, again, Mike, I'll pause there for a second. We've been explaining this for years, that this is where this was going. We now actually see the data, and it's just, this wasn't intentional. It's like, again, connecting the dots as you go throughout the resources of the week, you're like, oh, wait a second. The information says this.
[00:22:27] Oh, I had tagged this tweet from this guy, from Merck, and now all of a sudden it's like, oh, it's, it's all happening. Like you're seeing the pieces come together.
[00:22:34] Mike Kaput: Yeah. I think it was maybe easier before to ignore this trend as you're kind of really getting your hands dirty with AI and saying, okay, like some features aren't there yet, or some tools aren't as advanced as people would want to admit or want to sell you on.
[00:22:50] This could not be more clear now. You really, really, I just would implore anybody listening, if you haven't really taken a close look at the writing on the wall just yet, this is your [00:23:00] wake up call to do so. You're probably still far ahead of the curve, but this is going to be the story I feel like in the next several years.
[00:23:08] Paul Roetzer: Yeah. And I like this idea of the economy will become an RL environment machine. I get that like everybody in every field, people are gonna be willing to pay for expertise to train the models. But what happens when they've been trained, right? Like what happens when they're genius level at your job, and then what do you do?
[00:23:24] So I I think there's gonna be a window where people are gonna be able to make a ton of money. It's almost like, it's almost like the uber economy Yeah. For ai. Yeah. Like, it'll exist for a while until the autonomous vehicles take the role of the people driving the cars. And that's basically what this is.
[00:23:41] You're gonna have this window where there's like, tons of money's gonna be made, companies are gonna be built on this. People's careers may be this. You may be an AI trainer for the next five to seven years and like, make great money. Maybe you can make a couple hundred thousand dollars a year. Just train an AI in these models.
[00:23:54] then what? Right, that's my, that's my big thing. Which then [00:24:00] leads us to what will AI look like in 2030? So this is an EPOCH AI report. It's 119 pages. Another great notebook LM use case. This one was commissioned by Google DeepMind. So it provides a comprehensive forecast of AI development by 2030. So it basically looks at the scaling laws and say, okay, are they gonna continue?
[00:24:18] And if so, what does that mean? The report argues that the exponential growth in training and inference compute, which again is when you and I use the models, will continue requiring investments, potentially reaching hundreds of billions of dollars, which is what we've talked about. The training of these models is gonna cost billions or hundreds of billions of dollars.
[00:24:35] The byproduct of these investments could be a significant increase in net productivity in GDP. Across the economy worth trillions of dollars. that then led to the, the other one from last week, the Anthropic Economic Index, understanding AI's Effect on the Economy. This is a 48 page report, as we've talked about before.
[00:24:54] I think this is the third edition of their report. Claude is dominantly used for [00:25:00] coding. So you always have to look at their uses and understand that their user base is, is heavy coding, so they don't have the totality of coverage across, you know, the world of, of diversity of use cases, but will actually get into that in openAI's research report in the second main topic.
[00:25:17] So in, in the Anthropic one, they said existing capabilities of clawed and other frontier AI systems are already poised to transform economic activity. Given how broadly applicable the technology is rapidly advancing AI capabilities only in reinforce the conclusion that immense change on the horizon and yet early AI adoption is strikingly uneven.
[00:25:36] Usage currently clusters in small sets of tasks with strong geographic variation that is highly correlated with income. So this gets into some much larger societal issues. People with money, people in privileged areas are the ones that are benefiting most, and this is the first study that breaks this into like regions.
[00:25:54] They say AI is being used to fully or directively, automate more and more tasks over [00:26:00] time. In just nine months, we saw directive automation jump from 27% to 39% of all conversations. For enterprise customers, that figure is 77%. So this is a really important distinction. They say users are entrusting Claude with more autonomy, directive, quote unquote conversations where users delegate complete tasks to Claude jumped from 27% to 39%.
[00:26:22] That's 12 basis points, which is. I dunno, quick math, 40% ish or whatever increase not insignificant. So again, the difference between 27% and 39% isn't 12%, it's 12 percentage points, which is much closer to like 40%. So there's a major jump in a very short time where people actually relying on these things to do the work for them.
[00:26:44] So they then break, everything into automation versus augmentation. And it, so I kind of like where they're going here. Again, their data is somewhat limited, because it's predominant coding. But I like the direction and I like the fact that they're really committed to these [00:27:00] economic impacts studies.
[00:27:01] And then that leads to Mike, which you had, highlighted, the leading economist of 40 of them signing a letter calling for the Department of Labor data AI's impact on jobs. Good luck. Yeah. The current administration is hiding every relevant piece of data that would show any of this. So. I, this is great that the leading economists are calling for this.
[00:27:23] They're not gonna get it like any, any indication. So this is where we have to rely on these like private companies and third parties to do the research. The US government is not going to tell you if they're seeing trend line data that says AI is going to take jobs. It's not coming from the Senate administration.
[00:27:39] So, again, regardless of political views, if you're new to the podcast, we, we keep this as middle of the line as humanly possible. I don't care who is in office, which side is in office, all we're gonna talk about is the implications on ai. And I'm gonna tell you right now, this administration is not going to tell you if they see an impact of AI on jobs.
[00:27:58] So you have to look at the third party studies [00:28:00] to see what's actually going on. Alright, I'll stop there 'cause I know we're gonna kind of work into the openAI's one, but you, you make the call. Mike, if there's anything else you wanna talk about on this one before we go to open.
[00:28:09] Mike Kaput: No, that was an awesome connecting the dots here.
[00:28:12] I think it's super valuable. Just one kind of final point here, and I don't wanna be too. Down in the dumps. But I think you have to, as a professional, kind of in charge of your own journey and your own success. Like I do think you kind of, I applaud very much the company is putting out more research. This is all needed.
[00:28:29] But like you kind of have to sit back and realize whether it's the government or the company's building this stuff, like nobody is coming to save you on this issue. And I realize that can be really depressing in some ways. But I also think the sooner we kind of get to that as individuals trying to kind of make our way in the world and navigate this stuff, I think the sooner you get to that idea, the more you can at least start empowering yourself using these tools, using the research out there, the education being put out by us, by plenty of others to kind of chart a path forward.[00:29:00]
[00:29:00] Because if you think to your point about the administration, if you think the data is coming, the policies are coming to fix this or to address this fast enough, I think that's deeply unlikely.
[00:29:10] Paul Roetzer: Yeah. And I do think. You know, when I look at this over time, I just see a very extended runway for people who are highly competent with AI tools and have domain expertise.
[00:29:22] Yes. Like I just feel like the diffusion of this technology is still going to take years, no matter how advanced the tech gets, no matter how good these models get. People who work in corporations know nothing happens fast, nothing happens overnight. That's, so there are some companies that are gonna be disrupted very quickly, but you know, a lot in like the tech space and stuff like that, like disruption is coming, but you becoming AI literate, you being highly competent in using these tools, being the one in your company that can apply it, you, you have a, a wonderful runway ahead of you.
[00:29:53] Yes. and to be more valuable in your company. So, yeah. and I think like over time, maybe we do figure it out and maybe the [00:30:00] cliff doesn't happen as fast as, you know, sometimes we think it might. and maybe new jobs do emerge and like, maybe we solve all of this, but sitting on the sidelines and waiting is not the answer.
[00:30:11] Yeah. You, you will get disrupted as a company or as an individual. So your best chance is just stay on the forefront of this stuff and like connect the dots before everybody else. And I feel like that's always been true in business. The more proactive you are, the more intrinsically motivated you are.
[00:30:27] The more knowledge you have, the more power you have, and the more leverage you have in your own career. And I , I feel like that's going to continue to be true. so yeah. It's, yeah. Yeah. Like that lens of kind of hope and optimism around this Yes. While also being realistic that we're running out time to get started if you're not started yet.
[00:30:45] Yeah. It's extremely exciting. Yeah.
[00:30:47] Mike Kaput: Yeah. I mean, I feel like roughly, I kind of, in my head I'm just like, there's a golden opportunity decade at least. Yeah. Right now with this stuff, doing everything you just said despite, you know, how starkly I frame this, [00:31:00] I think there's a huge upside. I'm excited.
[00:31:02] Yeah.
[00:31:03] Mike Kaput: All right, let's get into our second main topic here, which is a study released by openAI's. They have released what they call the largest study to date of how people are using chatgpt. So in IT, researchers looked at 1.5 million chatgpt user messages between May 20, 24 and June, 2025, and they used AI to label what each message was kind of being used for, and they found that most usage over 70% is now non-work related.
[00:31:34] People use it for writing information, seeking tutoring, advice, translation, creative brainstorming, and they kind of break this down into three big categories of usage. The three biggest ones are what the researchers call first practical guidance, which is basically like getting advice. Writing and then seeking information.
[00:31:53] And these three broad categories account for nearly 80% of all usage. However, work usage is rising [00:32:00] too, especially among educated users in professional roles. And in that area writing tasks dominate. But AI is increasingly being used as a decision support tool, not just to execute tasks. The study actually classifies most work use as what they call doing.
[00:32:15] So things like generating content or asking where ChatGPT acts more like an advisor. Now an interesting surprise here is only 4% of the messages they looked at are about coding and even fewer are about relationships to kind of big topics of AI usage. So Paul, I'm curious like what surprised you if anything, about this data?
[00:32:35] I definitely thought that last point was interesting.
[00:32:38] Paul Roetzer: again, another good one to analyze, throw into notebook alarm is 64 pages long. They dropped this I think the day after Anthropics maybe, but like, again, both labs working on big economic data, big research studies simultaneously. the one thing like I will note is, so this was done in part with National Bureau of Economic Research.
[00:32:57] Yes. working paper by Pennet economic research team. [00:33:00] It covers consumer plans only. So when we're talking about the use of commercial or, or, you know, work versus personal, that's only for these consumer plans. It does not, based on my interpretation, what they mean by consumer plans only does not mean team and enterprise accounts.
[00:33:17] So that's extracting the millions of users who have team or enterprise accounts. And we're only looking at individual plans here. So the fact that even then 30% are still Yeah. Work related. Right. Like, so 70% is, that makes sense. Like it's a personal account. one thing I thought was interesting, Mike was the age breakdown.
[00:33:35] Mm. So they said nearly half or 46% of all messages sent by adult users were from users 18 to 25. Wow. So that's one that you, again, when we zoom out and we think about the impact on society, the impact on the economy and the future of work, the people who are getting the value from these tools are the younger generation.
[00:33:54] It's the generation we've talked about recently that are, have 13% unemployment, right. Like the 22 to 20 [00:34:00] fives. And so the people with the capabilities are really starting to, to centralize in that young age group. And that's gonna be that AI native generation, like they won't have known education or life without an AI assistant on call.
[00:34:14] And that's fascinating, like I think the downstream impacts of that. The other one I would just mention there was an excerpt said A key way that value is created is through decision support. Chad CPT helps improve judgment and productivity. Especially in knowledge intensive jobs, and as people discover these and other benefits usage deepens with user cohorts increasing their activity over time through improved models and new use case discovery.
[00:34:37] The one thing I thought about here, Mike, that just keeps coming back, every conversation I have, whether it's, you know, basketball on Thursday nights or just meeting with friends who are in corporations or talking at, you know, when I go out and do presentations, meeting with executives, the lack of awareness and usage of reasoning models Mm.
[00:34:55] Is, is shocking to me still. Like people who have never done a deep research [00:35:00] project. Yeah. Nine times outta 10 when I'm talking to somebody and they're like, oh yeah. Because they know I do AI stuff, so everybody always talks to me about AI stuff. it's like, oh yeah, I'm using ChatGPT all the time at work.
[00:35:08] I was like, are, have you used deep research yet? No. What's that? Every time, and I'm like, dude, you gotta go do it. Like, and I'll give 'em like, Hey, run this prompt. Like you have to try it. It'll change your perspective on all of this. So this, this research is really fascinating and this, this, the contextual stuff you and I hear all the time, Mike, about how little people still know about what these things are capable of.
[00:35:32] It's part of why I love the Gen AI app series we've been talking about. It's like, get that stuff to people and like bite-size bits, like notebook lm. I'm not intentionally like, you know, talking about notebook, I'm nonstop this, but that's the kind of thing, like you go in and you use that, you try the guided learning.
[00:35:45] I said this to some parents, so I think I've mentioned this. I paid BA basketball on Thursday nights and so it's all a bunch of dads from my kid's school. And so I was, you know, sitting on the bench like talking with a few guys. Like, have you tried guided learning with your kid yet? No, no. What's that?
[00:35:57] Mm, I mean, it changes the way you help your [00:36:00] kids learn. Like, and so there's all these incredible features that I think help people connect the dots of the value these tools can create. and so, you know, to watch this data over time is, is really interesting. But I think we're kind of hitting that escape velocity point where.
[00:36:15] I mean, OpenAI's is what, 700 million? So about 10% of the world
[00:36:19] Yeah.
[00:36:19] Uses chatgpt on a monthly basis. The question is, are they just using it to do quick prompts and they don't know about image generation and, you know, reasoning models and guided learning and all these other capabilities. My guess is, yeah, and that's what kind of the data is showing.
[00:36:34] It's like the vast majority then they say the vast majority of use cases is for writing. Like it's a dominant use case. It's the most obvious thing to use 'em for. And so I think once we catch up a society of all the other capabilities, I think Google's doing a really good job with Gemini right now to surface those things.
[00:36:50] Like when you go into Gemini now, they're kind of showing you prompts to help you use nano banana. . ChatGPT, we talked last week about like the hundred sample prompts for college students. Like [00:37:00] I think the labs know it. They need to make the uses more tangible for people so they understand the value they can create faster.
[00:37:06] And I think we're heading that way.
[00:37:08] Mike Kaput: Yeah. The. We barely are even scratching the surface on using these things as true like cognitive co-pilots, like we've discussed many times using a reasoning model for deep strategic work is like gaining a multi, you know, double digit IQ bump. Yeah. And I wonder what the effects of that will just be once people, if people are doing that at scale, you know?
[00:37:29] Yep. one final thing that just jumped out to me is they actually wrote in this report, quote, this widening adoption underscores our belief that access to AI should be treated as a basic right. And like, honestly, I like can't disagree with that at this point. I feel like it's roughly where internet is for me as well.
[00:37:47] Like you have to have access to this to do anything,
[00:37:49] Paul Roetzer: which is interesting 'cause it ties back to the philanthropic thing that looks at the disparity of income levels and access. Um. Yeah, I agree. I mean, it, it is truly like a [00:38:00] utility. It's like electricity, you know, it's intelligence is the new electricity in essence.
[00:38:03] . Like, everybody's got access, running water, you know, electricity intelligence like that. I think that's how the labs think about it. And it's probably a good mindset where I was saying at the beginning, like, AI is truly the underly underlying operating system to society. I think that's how it becomes.
[00:38:18] And anyone who doesn't have access to that intelligence on demand, whether it's for helping your kids at school or helping you with work, or helping you find a job, or helping you plan your life, like it, it is going to change the way we do everything, the way we buy, the way we consume information, the way we build companies.
[00:38:33] and yeah, access is, is fundamental, I think.
[00:38:36] Mike Kaput: Yeah. Just flagging for our future selves. There's at least a couple keynote ideas in there of what you just said. Yeah.
[00:38:43] Paul Roetzer: I don't know what I just said,
[00:38:45] Mike Kaput: but intelligence as the, as the, as the new electricity, you know? There
[00:38:48] Paul Roetzer: you go.
[00:38:49] Mike Kaput: All right. Our third big topic this week is a talk given by Rept, CEO, Amjad Mossad, in which he said The future of business is being rewritten thanks to AI in [00:39:00] some really unique and interesting ways.
[00:39:01] So he gave this recent presentation at Y Combinators AI Startup school event, and in it he envisioned a world where anyone can build software just by speaking with no coding required, and AI agents doing all the heavy lifting. In this type of future, the role of an engineer, a software engineer, for instance, shifts from technical execution to high leverage thinking.
[00:39:22] And as we discussed last week, rep's own agent three is already working autonomously for multiple hours, writing code, deploying software, and testing features. As a result, Mossad predicts traditional SaaS will go to zero because every individual will spin up personalized apps or agents on demand. And he says this will also have widespread effects on how companies are structured at a fundamental level.
[00:39:44] He sees eventually organizational hierarchies. Being replaced by fluid networks of generalists collaborating with autonomous tools. Now, Paul, that last bit is kind of the money quote here, money idea. And I know you have unpacked [00:40:00] that a bit on LinkedIn and you've been thinking about that idea a lot.
[00:40:03] Paul Roetzer: Yeah, so I don't remember what day this was that I listened to this, but you know, we obviously talked last Monday, came out on Tuesday, episode 1 67, about agent three and their pursuit of autonomy and how they felt they'd get 10 x increase in the autonomy of their agents and their ability to do like 200 minutes of runtime without human intervention.
[00:40:19] And so that was like on my mind and the way I do my podcast listening is I'll, I'll go through and like, you know, once a week I'll scan all the latest episodes, I'll save ones to listen to. And then depending on how much time I have, whether it's in the car or out for a run, I try and listen to one that I can, like consume as much as possible, like in that, that short segment.
[00:40:36] And so I listen to this one on the way into work and it's like 32 minutes long, but one and a half speed. You know it, you get through pretty quick. And so I got to the office and I was like, damn it. Like, I shouldn't have listened to that because I had a lot to do that day and my mind was just sort of racing with what he had said.
[00:40:52] And it reinforced a lot of the things we'd heard about and we've even talked on this podcast. But yeah, the four things I called out in the LinkedIn post was, [00:41:00] and on his vision for the future of business is ideas are the greatest resource. So pairing with knowing how to use AI to bring them to life.
[00:41:06] So again, these things are gonna be capable of basically everything. If we just assume the scaling laws to be true, we look out, you know, two, three to five years, you have to assume they're gonna be genius level at every profession, every domain, and they're going to be able to execute most of the tasks that humans do.
[00:41:22] So the greatest resource he's saying is the ideas of what to use them to do it. So if we assume we all have access to this genius level, intelligence on demand for anything we want to do. What do you do with it? . And so, I mean, you could look at deep research right now as like a prelude to that.
[00:41:38] You, and I know deep research is capable of it. I could come up with 15 things to do every day with it. If we had the capacity to do it like all day. I think about things I could be doing with it. I think that's what happens here. It's like, okay, we have the ability, but how many people actually understand it?
[00:41:52] . So this idea that moving forward ideas become the greatest resource, which requires experience and expertise, which bodes really well for [00:42:00] senior level people who can envision the application of this intelligence to their businesses. The second, everyone is empowered to create apps and software in real time to solve problems and increase productivity using language.
[00:42:10] No coding needed. I don't think people comprehend the significance of that.
[00:42:15] Hmm.
[00:42:15] just anything you, you want to spin up. You can literally just go and say, Hey, build this thing for me. Build this app, build this software. You know, do this thing more efficiently for me. The one you mentioned, org charts look more like fluid networks versus rigid structures.
[00:42:29] As someone who is actively hiring and planning to scale very quickly, our business, this is the one that I lay in bed thinking about. . Like, I don't wanna structure the business in a traditional rigid way when we have the opportunity to, to reimagine this as an AI native company. And so I want to, I want to get that right and I want to like position us to be able to adapt quickly and not be stuck in some traditional org charge structure.
[00:42:52] And then the fourth optimal teams are collections of generalists, working with agents to achieve a mission. I love that idea. And similar concept, it's like, [00:43:00] do we really hire salespeople and customer success people and marketing people? And like, do you do it in this traditional way or are you just hiring really intelligent people with great ideas and the ability to work with agents who can maybe work cross-functionally?
[00:43:15] Like, I don't know the answer. And that's why I loved this podcast, is it actually like forced me to step out and given where we are as a company ourselves. Just think about what would I do if I could do anything? If I could reimagine how to build a company, what does it look like? And to me, as you were saying earlier, Mike, the optimism and the excitement of this moment, like to get to do that .
[00:43:35] To like not have in our case, thousands of employees that it might impact. but where we we're a smaller team and we can like, try and do this right from the ground up. I don't know that there's ever been a better time to build a company when, when you think about it in this way. we're still planning on hiring a ton of people.
[00:43:53] Like we are a people first company. Like we wanna hire a bunch of people. Yeah. But I'd rather hire 50 or a hundred versus a thousand. [00:44:00] Like if I can do this the way we want to go and pursue our mission of accelerating I literacy and transformation and not need a thousand people to do it. That'd be amazing.
[00:44:08] Like I'd love to do it. I want those 50 to a hundred, but I don't want more than I have to have as an entrepreneur and as a CEO. and I think like I know executives at major companies with thousands of employees who are thinking the way I'm thinking. . But they have a bunch of legacy staff and technology and that's why I'm convinced disruption is coming.
[00:44:26] Like, they have to get where we're trying to go, but they have to first get through what already exists to do it. So that's the idea of AI native versus AI emerging companies that we often talk about way, way easier to be an AI native company right now. Hmm. To be the ones building smarter from the ground up.
[00:44:44] Mike Kaput: Yeah. It's interesting to think about. I don't have any of the answers either at this stage, but you can almost see at least one possible pathway forward where when you go to hire someone in this kind of AI native organization, it's like. I don't even [00:45:00] know if I'd care or even look at their background.
[00:45:02] It's like, just get me in the room with an hour with them for an hour to kind. It's like carry your vibe. Hiring almost. But you're kind of testing that.
[00:45:09] Paul Roetzer: Yeah, we're doing it right now. We've made a couple hires like this already. Yeah. Where I just wanted people who I trusted, who I knew were highly intelligent, who are adaptable, who are resilient, and who could solve problems.
[00:45:19] . And bring ideas to the table. And it's like, I don't even, like, I 'll create a new role for you. Like, I just, like, that's the kind of people I want. And I think that, you know, it's a change for businesses, but it's also a change for professionals. Like people who've maybe built their career and become successful in one track.
[00:45:34] Like, well, I was a marketer, I was a salesperson. Or like, yeah, but you don't have to be like, you can be anything now like that. And I think that's a, again, to like wrap your mind around that. Yeah. And then how do you hire for that and how do you train people for that when it is this gen? But I'm, I've always, Mike, you know, we worked together at my agency all those years.
[00:45:52] I always hired generalists, right. Like I was a huge believer of like. Companies hire us to transform their businesses, grow their businesses. You need diverse knowledge across a [00:46:00] number of areas. You can't just be a marketer and be a consultant. Like you have to have diverse knowledge. So I 'm very comfortable in the building a business around generalist idea.
[00:46:08] That's what I did. But I, that's not a comfortable thing for people who have made their careers in, in one specific department or domain
[00:46:16] Mike Kaput: Yeah. Topic for another time. But kind of interesting as we have these conversations to just see like all these almost serendipitous things that you've learned in your past businesses and career, that it becomes super, super relevant in the age of AI and build, it's like, almost like if you're meant to build this type of business.
[00:46:34] Right?
[00:46:35] Paul Roetzer: Yeah. And I would say like, just, again, I don't, I don't talk about the personal stuff too much, but like, this is, it's, it's wildly exciting. It's wildly frustrating. . Um. Because like, there's a part of me when I listen to this podcast that's like, I just wanna go, I wanna go hide for five days and solve this.
[00:46:51] Like, because it, you, you hear something sometimes and you're like, oh, I can figure this out, but I need brain power and I need quiet. And like, you just wanna [00:47:00] like leave and go on a retreat for five days and do nothing else but figure this out. But then the reality smacks you. It's like, oh wait, MAICON's in three weeks after four presentations, we're hiring a bunch of people I need to make hiring decisions.
[00:47:11] Like, and so reality comes back to where like, we all have jobs, we all have other things we have to do, but stuff like, this is what I want to do. Like, I wanna do all of it. But like I love trying to figure this stuff out. And I think the answers are there. And I think in part it's because we now have reasoning models that you can work with as a business leader to like bounce things around at midnight when you're thinking about it, which we didn't used to have.
[00:47:35] So yeah, with my CO C-E-O-G-P-T and me, I think we could. Do things like this. I just, I have to find the time to do it and I think we all struggle with the same challenge.
[00:47:45] Mike Kaput: Yeah, for sure. Alright, let's dive into this week's rapid fire topic.
[00:47:49] Mike Kaput: First up, meta has launched their Ray Band display glasses.
[00:47:54] This is a wearable that features a built-in screen and it's paired with a neural [00:48:00] wristband for gesture control. So these are not just camera glasses. the right lens now shows text messages, video calls turn by turn to directions and visual answers from Meta's AI assistant Instead of tapping the frame, you control these glasses by subtly moving your fingers.
[00:48:17] Thanks to the EMG powered meta neural band that reads your muscle signals, mark Zuckerberg calls this the next step towards super intelligence positioning the glasses as the start of a larger platform ship shift, and they ship September 30th in two colors and cost $799. So Paul, we've talked about wearables in different formats, like plenty of times.
[00:48:41] but there's a good reason for this. Like they do seem like a necessary, but as of right now, immature or even missing piece of kind of AI being truly everywhere, right? Not just on your phone or in voice mode or whatever.
[00:48:54] Paul Roetzer: Yeah, I mean, I think Apple's gonna make a play here. Google's gonna make a play. I think openAI's, while they're saying it's not [00:49:00] glasses, their initial one, they also are looking at all user interfaces of the future.
[00:49:04] So I think we just have to come to grips with whether you like these things or not. They're probably gonna be a pretty commonplace thing, you know, in the coming years in society. I will say, just on a separate note, if you've ever given a presentation, especially a live demo and it didn't work,
[00:49:18] ah,
[00:49:19] just go watch Zuckerberg, like it wasn't as bad as what he went through.
[00:49:23] It was so painful. Their two live demos just crashed and burned, and Zuckerberg ended up on stage for a minute trying to get. A live demo to work where God, one of his executives was calling him and he was gonna like, you know, FaceTime or whatever they call it with, with meta, just brutal. And after the fact, I did see one meta executive say the problem was when he was saying, Hey, meta or whatever, it was activating every meta device in the auditorium.
[00:49:50] And so it was like crashing the system and that's why it wouldn't work. But it's so uncomfortable, Zuckerberg did as best of a job as you can do, where he's probably thinking like, [00:50:00] how many people am I firing when I get off stage? Oh, right. You can see every like, thought going through his head while it's happening.
[00:50:05] Yeah I , again, like glasses, wearables, the form factor TBD, like I feel like a lot of 'em do feel like glasses is a play. I think that Apple will probably input, cameras into AirPods eventually. I think Apple believes AirPods are actually, may, maybe even a bigger play for this. Always on, always aware of your surroundings, where you can get a 360 view from, you know, multiple cameras in your AirPods.
[00:50:30] I think they'll make a big play maybe next year around that. So yeah, I, wearables are gonna be a thing. I think it might take five to seven years maybe before there's a clear winner, as to which form factor it is or maybe it ends up being a collection of form factors. But the thing I am confident is by 2030, just assume everything you do, everything you say is always being recorded by someone.
[00:50:53] Yeah. Wow. Like if you go to a conference, especially a tech conference, like everybody's gonna be recording everything, probably already are [00:51:00] in some cases. So again, like it or not, just be prepared for a future where this is standard, through different devices.
[00:51:08] Mike Kaput: Yeah. We also, we can't talk about onstage fails.
[00:51:11] This close to make on that was making me extremely anxious. You have to connect
[00:51:14] Paul Roetzer: that dot. I wasn't even thinking about that. I'm not doing any live demos, so. Right, right. No, I just,
[00:51:19] Mike Kaput: even just hearing about it, I'm like, I feel so bad. I mean, obviously it's your job to make sure this works, but I just feel bad no matter who it is.
[00:51:26] Actually, I
[00:51:26] Paul Roetzer: understand that. I will probably be doing some live demos in the workshop. Oh,
[00:51:29] Mike Kaput: no.
[00:51:31] Paul Roetzer: Always have plan B.
[00:51:32] Mike Kaput: Yeah. All right.
[00:51:33] Mike Kaput: Next up. AI has aced one of the toughest coding competitions in the world. Both Google Deep Minds, Gemini 2.5 and open AI's reasoning models have competed in the 2025 ICPC World finals, which is basically the Olympics of collegiate software programming.
[00:51:51] They delivered stunning performances. Gemini solved 10 of the 12 problems that they have to complete in the five hours of the competition, [00:52:00] including one. No human team crack. openAI's solved all 12. And this is a perfect score that would've secured it first place if it was actually a competitor in the competition.
[00:52:10] what's interesting about this is that these problems they give during this competition, they demand things like abstract reasoning, multi-step logic, and creative solutions under strict time limits. So, you know, both Google and openAI's are saying, look, this is more than just like a party trick when it comes to coding or, or when it comes to mass.
[00:52:30] It is signaling this general purpose reasoning that ais are now able to do at or above elite human level in high pressure problem solving. So this could translate, they say, into all sorts of breakthroughs in fields like drug discovery, chip design, and logistics. 'cause all these problems are like algorithmic.
[00:52:50] So this definitely seems like a noteworthy milestone, Paul. I mean, I realize it's like, it's a kind of one competition. But it just really feels like even more if I'm a, [00:53:00] if I'm a programmer or a coder of any type, I'm looking over my shoulder what AI can do now.
[00:53:06] Paul Roetzer: Yeah. You can't. I get that it's like abstract and maybe these are like competitions you've never heard of and things like that.
[00:53:13] These are huge deal within, like Silicon Valley, within the tech world, within AI world, these math competitions are massive. Yeah. but to the average person, it's like, I , I have no idea what, what I would say is like, as you explain this, I kind of think about this. So, for a long time, decades, the pursuit of language was like the cornerstone of AI development because they, they believed that that language was fundamental to human intelligence.
[00:53:38] . So understanding and creating language was needed to break through and build general intelligence. That's what the Transformer Mormon moment in 2017 gave us. Was that the architecture to now pursue what ended up becoming the language models we see today. Yeah, math and reasoning is sort of the next frontier to [00:54:00] solve what they think to solve, to truly reach human level general intelligence, right?
[00:54:06] And so math, these breakthroughs, these advancements are fundamental stepping stones to building the agentic AI that we talk about and that they all envision. And to building the kind of economic impact we let off with. These are fundamental steps to, to doing that. and so, yeah, I would not sleep on these kinds of breakthroughs.
[00:54:31] They can pass by real fast. And you, you know, you're never gonna hear about the stuff in mainstream media, I don't think. Yeah. Again, I don't watch too much mainstream media read too much, but this is not stuff that the average person thinks about or cares about. But when you look back three years, five years from now on, the milestones toward AGI and beyond, things like this probably appear on that timeline, I guess is a good way to.
[00:54:55] Mike Kaput: All right. Next up, a new Pew research report finds that Americans are far more [00:55:00] concerned than excited about AI's growing presence in daily life. Half of US adults, they surveyed, say they're worried about the impact AI will have compared to just 10% who are ex or are more excited about the impact it'll have.
[00:55:13] And the bigger fears here are that AI will erode people's ability to think creatively and form meaningful relationships. Only 16% of people in this research surveyed believe it will actually improve creativity. However, at the same time, most Americans are open to using AI for everyday tasks, especially in areas like weather prediction, fraud detection, and drug development.
[00:55:36] But when it comes to personal decisions that they asked about, like faith or matchmaking, most say AI should play no role at all. Interestingly, 76% of people said it was really important to know whether content was created by a human or ai. But more than half admit, they can't reliably tell the difference.
[00:55:56] So Paul, it's always good to see some solid data from a reputable [00:56:00] polling outfit about AI attitudes, like what jumped out to you here in these responses.
[00:56:04] Paul Roetzer: Yeah, it's, it's a great point, Mike. The, we talk a lot about research on this podcast and we always encourage people drill into the methodology, figure out how was it done, who, who was part of it, what did they ask?
[00:56:16] So this is over 5,000 adults from June 9th to June 15th. So recency is great. large sample size done as you mentioned by Pew Research. There's very few, if any more reputable, research arms. So that's all great. But I was curious, it's like, okay, so we're, we're doing, sharing all these findings about ai .
[00:56:36] From who, like, who are the 5,000 adults? And I get that it represents, you know, the us base and it does, it's like a broad spectrum. But I actually was, I wanted to see the questions. And so I wanna see how are they defining ai. So if we're assuming that a large portion of US citizens don't know what AI is, other than maybe what they've seen in ads or heard about or seen in movies, how relevant are the rest of the responses is [00:57:00] kind of what I'm getting at.
[00:57:00] if you're asking people who don't even know what it's, so I found the questions they ask, which Pew does a great job of, like, Donald, the report view the question. So it's super clear how they did this. The first question is, AI is designed to learn tasks that humans typically do. For instance, recognizing speech or pictures.
[00:57:17] So assume it's your grandma, your mom, your your brother, like your coworker who knows nothing about ai. That's the question they were asked. And the, your choices are a lot, a little nothing at all. Don't know or just refuse to answer it. So right off the bat, nothing at all. I know nothing at all about this AI you are explaining as recognizing speech or pictures.
[00:57:38] That's a pretty. Broad and incomplete explanation of what AI is. If someone is like, I don't even know what you're talking about, but Oh, okay. Recognizing that's now what I think AI is, is kinda what I'm getting at five. 5% said nothing at all. 48% said a little, so 53% of people surveyed under the definition of [00:58:00] AI as recognizing speech or pictures, doesn't know anything about it.
[00:58:04] Mike Kaput: Yeah.
[00:58:04] Paul Roetzer: So you then have to segment the responses by like, okay, the only people whose answers I even care about are the 47% of people who actually claim to have read a lot about it, because now we're gonna get like a little bit better sample. And again, to pew's credit, they break this stuff down 10, 10 ways tomorrow.
[00:58:24] Like there's 46 pages, tons of data, they can throw all the charts. So I would say it's, it's really interesting stuff. There were definitely some points that jumped out to me. you know, the creativity one you mentioned, 76% say extremely, are very important to be able to tell if pictures, videos, and texts were made by ai.
[00:58:40] Like that, you don't need to know what AI is to know that that's should be true.
[00:58:43] Mike Kaput: Yeah.
[00:58:44] Paul Roetzer: so there's, there is some good data in here. The attitudes, 50% say they're more concerned than excited about the increased use of AD in their daily lives. So, good stuff. Just always go into it before you start throwing, you know, clickbait up and things like that.
[00:58:59] Know a [00:59:00] little bit more about the research, to, to make sure that you, you know, what you're sharing. But this is a super legitimate study. Lots to be learned from it. very digestible. This is not a highly dense, research like some of the other ones we talked about today. So I would say it's a good read and it's a good resource for people wanna understand how we're at.
[00:59:17] I would imagine there are a lot of politicians reading this research. . Trying to gauge. Again, how much does AI play into the midterms next year?
[00:59:24] Mike Kaput: I was going to literally say I would bet a substantial amount of money. There is some private polling in the field trying to figure out what the wedge issues are here and how people feel about them, you know?
[00:59:34] Yep.
[00:59:36] Mike Kaput: Alright, so next up, the lawsuits against AI companies keep coming. So periodically we're going to dedicate a rapid fire segment to recapping some of the lawsuits that are happening and their significance. So this week we have two new ones that have been filed against AI companies. First, Disney Warner Brothers Discovery and NBC Universal are suing the Chinese AI company Minimax for what they call massive scale piracy of copyrighted characters.
[01:00:03] So this is filed in federal court and it alleges that Minimax is image and video generator, which is called Hell. UO AI routinely produces high quality content featuring Hollywood ip. So, for instance, one prompt they said can generate. Videos of Disney, Marvel, DC characters, all with Minim Max's branding on it without any licensing or permission.
[01:00:26] And the studios also claim Minimax ignored multiple cease and Desists, and they continue to profit from this kind of theft. So they're seeking damages and an injunction to halt what they argue as blatant copyright infringement. Second Penske Media, which is the publisher behind Rolling Stone Billboard and Variety, has filed a landmark lawsuit against Google, accusing the tech giant of using its journalism to power AI generated search results without permission.
[01:00:51] The suit actually targets Google's AI overviews saying that they scrape and summarize Penske content while siphoning away traffic [01:01:00] subscriptions and ad revenue. Penske says Google effectively forces publishers to hand over content in exchange for visibility. It is the first major US publisher to sue over this issue specifically.
[01:01:11] So Paul, another lawsuit against AI video generation companies. Curious how you see us playing out, just because it seems so blatant and obvious these tools have been trained on this stuff. And then would also just love your thoughts overall on the AI overviews lawsuit.
[01:01:25] Paul Roetzer: Yeah, I don't know. I mean, I, again, they're just, there's going to be dozens if not hundreds of these.
[01:01:30] There probably already are dozens of them. I think we just get the one surface that we see in the media. But, yeah, I mean, like we've talked on before, they, they did, they, they obviously trained on this stuff. Yeah, there was one. I was just trying to find it real quick because I don't think I put the link in for this week.
[01:01:43] We'll touch on next week, but Sora, I think a new version of Sora is right around the corner, video generation from openAI's, maybe, maybe this week. certainly soon. And it was blatantly trained on all of this stuff. Like all of it we know took these things. I have no idea how this plays [01:02:00] out in the courts.
[01:02:00] we know a bunch of, you know, penalties are gonna be paid. Whether like I , I don't imagine they're gonna eventually have to like, retrain the models. Like, I don't think anything like completely disruptive to the industry happens. Yeah. But there has to be a reckoning at some point. And there has to be a resetting of the values that go into building these models by these companies.
[01:02:21] Right now. It is this, they did it, so we're gonna do it. And it, that's basically how this started in 2022. Like, well, they did it. Like, we gotta keep up. I mean, we have quotes from internal memos at Meta that said this exact thing. Like, we, we know this is what happened. So I don't know. I mean, we'll just like you said, we'll kind of keep following the space, but it's, it's challenging and everybody wants to just do a deal.
[01:02:44] I think, like at this point, just like, get their money from these labs and accept that they've done it and, try and, you know, get some revenue out of it or get some, some fees paid. I don't, I don't know. It's, it's a very complicated space and I just think it's gonna take years to [01:03:00] play through all this.
[01:03:02] Mike Kaput: Next up, Reid Hoffman, the co-founder of LinkedIn, has in a new thread on X says he is quote, voice pilled. and if you've ever spoken to ChatGPT, instead of typing, you might be voice pilled too. So he just kind of coined this term in this viral thread on X, where he argues that the next major leap in AI interaction won't necessarily come from bigger models.
[01:03:23] It'll come from how we engage with them. Voice input, he says it's faster, more natural, and more flexible than a keyboard. You can do things like fumble, rephrase, ramble, and today's models will keep up with you. This makes voice an ideal way to interact with ai, especially for creative or exploratory tasks.
[01:03:42] Because you're not compressing your thoughts to fit a text box. You're more like thinking out loud. So he thinks that voice will reshape hardware and even office design, and it may also make AI more accessible, lowering the barrier to AI for people across cultures and literacy levels. So soon enough here as you, [01:04:00] if you've ever experimented with voice mode and how much it's gotten better, talking to machines may just feel more human than typing ever has.
[01:04:08] And Paul, I mean, I genuinely have to say I think this reflects my own experience. I'm not like coming down on one side or the other, but my gosh, I've gotten so much done using voice mode. I mean, with the caveat, like it craps out like five times on my morning commute every morning, but it is really incredible what it, that it, it fundamentally unlocks some different stuff, I think.
[01:04:29] Paul Roetzer: Yeah. I 'm with you on the, you know, it stops working. Probably you and I drive the same route to work, tell you where it drops for you. I can tell you
[01:04:38] Mike Kaput: exactly where. Yeah.
[01:04:39] Paul Roetzer: On the way there and on the way back. I think that's my biggest problem with it at the moment is that you don't know when it stops listening.
[01:04:47] And if you get going, like you're in the flow and you say a bunch of stuff and then you're like, hello, are you there? and nothing. You're like, oh man, every, all those thoughts are gone. I think that's solved with local [01:05:00] running models. you know, so let's say Apple solves it by having a, a small language model on device, and it doesn't have to be connected to anything.
[01:05:08] It don't have to be a wifi cell signal, things like that. And then you, you know, it's working. and the whole idea is hands free, eyes free. Like, you know, when I'm having these conversations, well, if eyes are free, I don't know if it's transcribing what I'm saying or not. So I think you, for me, that has to be solved.
[01:05:25] I get so frustrated when those thoughts are just gone because it, it dropped and you didn't know it. So I think that that is an infinitely solvable thing that like I, as a non AI developer know how to solve that. I don't know technically how to do it, but it seems like an obvious way to solve. It's just to have it on the local device, whether it's your glasses, your AirPods, your phone, whatever it is.
[01:05:47] It's just the model lives there, not up in the cloud. So, yeah, I don't know. I. I do think, like at the office though, it'd be kind of weird if like everywhere I walked, everyone was just talking to their machines all the time. [01:06:00] So I do think there's some nuances that I think I don't, I don't know. Like I think I'm still gonna type, like I don't, I don't think everything's just gonna become voice, because you always get that when you walk by.
[01:06:09] Like people are talking on their air pods and you don't know that's what's happening and you stop and look at them like, oh, they weren't talking to me, they were talking to someone else. That would get weird if everyone in the office is just like constantly talking to their machines and no one knows if they're on a call or what they're doing.
[01:06:22] Mike Kaput: Right, right. Yeah. And I will say too, I'm running an AI productivity workshop at MAICON, and voice is not the focus. I'm it, but that's definitely one strategy we're going to cover. It's like exploring these different modalities that help you get more out of AI tools because you could see a very near future where you're not only typing and prompting and working with ai, but also you've got voice fired up just to log any thoughts or work through any issues.
[01:06:47] Yep. Things like that. Yep.
[01:06:42] Mike Kaput: Alright, Paul, to wrap things up, I'm going to run through some quick AI product and funding updates to kind of bring us home here. Sounds good. So first up, Elon Musk's AI startup Xai is raised over $10 billion according to Bloomberg. This gives it a $200 billion valuation and makes it one of the most valuable private companies on earth.
[01:07:11] Google has given its chrome browser major AI upgrade. Gemini. Google's flagship AI model is now deeply embedded into Chrome. So you can ask it to summarize a webpage, coordinate across tabs, schedule a meeting, or even pull up a YouTube video all without leaving your browser figure. The robotics startup building general purpose humanoid robots just raised over a billion dollars at a $39 billion valuation.
[01:07:36] This puts it among the most highly valued robotics companies ever. This money will help them scale up manufacturing, expand robot deployments, and grow the AI system that powers its bots. openAI's has launched GPT five Codex, a specialized version of GPT five, designed to act as a true coding teammate can independently work on complex software tasks, refactor or code bases, [01:08:00] debug, and even review poll requests with a level of depth that rivals senior engineers.
[01:08:06] Luma is a startup behind a tool called Ray three, which is the world's first video generation model, they say with native reasoning and studio grade HDR. So Ray three generates photorealistic 4K HDR video with real world physics, preserved anatomy, complex motion, and even interactive lighting. But what really sets it apart is this reasoning Ray three can think in visuals and language, interpret sketch, annotations, and follow complex directions.
[01:08:36] Paul Roetzer: And then one more Mike to add that I saw on Sunday nights, Sam Altman tweeted, over the next few weeks we are launching some new compute intensive offerings, which, would mean video image, reasoning would be the three things that jump immediately to mind as compute intensive because of the associated cost.
[01:08:53] Some features will initially only be available to pro subscribers. Which 200 a month, Mike? Is that the, that 200 a month?
[01:08:59] Mike Kaput: Yep.
[01:08:59] Paul Roetzer: [01:09:00] Yep. And some new products will have additional fees. That'd be interesting if you pay your 200 a month and you pay additional fees on top of it.
[01:09:06] Yeah.
[01:09:07] Our intention remains to drive the cost of intelligence down as aggressively as we can and make our services widely available.
[01:09:14] And we are confident we will get there over time, but we also want to learn what's possible when we throw a lot of compute at today's model costs at interesting new ideas. Hmm. so stay tuned. It's gonna be a busy September, October for openAI's and I would imagine they're not the only ones, but. I think we're gonna see video, image reasoning, maybe some new audio stuff from people, but those would be the things to watch for as we, and agentic would be the other thing.
[01:09:42] extended runtime of agents, kinda like we saw with agent three, that that would be the other, compute intensive. So agents, video image reasoning, that's the things I would expect in some forms from openAI's. yeah, that'll be, that'll be fascinating. So stay tuned. It's gonna be a, a busy [01:10:00] period here as we enter the fall.
[01:10:03] Mike Kaput: Awesome, Paul, well thank you for as always, breaking everything down for us this week and super packed, super exciting week.
[01:10:09] Paul Roetzer: Yeah. Thank you Mike, and thanks everyone for listening. We'll be back with you next week. Again, check out MAICON.ai. If you wanna join us in Cleveland, October 14th to the 16th and academy dot SmarterX dot ai.
[01:10:21] If you want to jump in on AI mastery membership and start checking out these Gen AI app reviews, core series and certifications. And all the AI Academy Live stuff that's gonna be coming this fall. So thanks everyone. We will talk with you again soon. Thanks for listening to the Artificial Intelligence show.
[01:10:37] Visit SmarterX dot AI to continue on your AI learning journey and join more than 100,000 professionals and business leaders who have subscribed to our weekly newsletters. Downloaded AI blueprints, attended virtual and in-person events, taken online AI courses and earned professional certificates from our AI Academy, and engaged in the Marketing AI Institute Slack community.[01:11:00]
[01:11:00] Until next time, stay curious and explore ai.