A

Sam Altman said this two years ago. He's like, right, if your product doesn't get better, every time our model gets smarter, we will steamroll you. Like our model. Our generally capable model will just do what your silly little SaaS company does, or we'll build it to do that. Right?

B

Right.

A

So they're not even hiding the fact that that's what they're doing. Welcome to the Artificial Intelligence show, the podcast that helps your business grow smarter by making AI approachable and actionable. My name is Paul Raitzer. I'm the founder and CEO of smartr Marketing AI Institute and I'm your host. Each week I'm joined by my co host and Smarter X 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. Join us as we accelerate AI literacy for all. Welcome to episode 224 of the Artificial Intelligence Show. I'm your host, Paul Raitzer, along with my co host Mike Kaput. We are recording Monday, July 6th after the holiday weekend. Mike and I didn't even literally talk before we started recording. So Mike, hope you had a great holiday weekend.

B

Same to you, Paul.

A

It's been a mad dash for me this morning getting everything situated. So yeah, it's good to be back after the holidays. We have a lot of interesting stuff. I think maybe some new models this week. We've got some stuff related to the government. This might be maybe, Mike, we can take a week off of Anthropic versus the government after this week, but we have one ideally.

B

I think Anthropic wants to do that too.

A

Yeah, I think it's like the final chapter of book one, maybe, and then like we're going to be into book two. All right, so quick question. When the last. When's the last time you actually clicked on a search result instead of just ranking? Reading the AI summary. That shift is happening right now and most brands have no idea how they show up in it. Site Improve is gentic content intelligence platform that helps marketing teams track, optimize, improve performance across both traditional and AI driven search. So we are thankful for Site Improve for continuing to sponsor the Artificial Intelligence show. And you can grab a free AEO check at siteimprove.com forward slash AIPOD. And this episode is also brought to us by AI for Business Boot Camp by SmartRx, which is happening in Columbus on July 16th. Mike, so you and I have one week basically to get ready for this, right? Yes. Oh, Man. All right, well, the good news is Mike and I run these workshops all the time, so it's constantly iterating on what we've done before, so we're not having to create from scratch. But Mike and I will both be busy this week finalizing our presentations for this live in person event. This is a single day event from 8:30 to 5:30. It is a single track, so you get all of it with one admission price. It's built for professionals and leaders who are ready to accelerate AI adoption and value creation. The day is going to start with a State of AI for Business keynote that I will present. It's then going to transition into interactive workshops. The first one is going to be led by Mike. This is before lunch AI productivity workshop.

B

And.

A

And then after lunch, I am going to lead an AI innovation workshop. So this is going to be an amazing day, not only to participate in the workshops, but to share with each other, inspire each other, learn from other AI forward practitioners and leaders. And again, all condensed into a Single day on July 16th in Columbus, Ohio at the Hilton Columbus at Easton. AI Academy Master members get discounted pricing, so make sure to take advantage of that. And then podcast listeners can get an extra $100 off of a ticket with the promo code POD100. So just go to SmartRx AI and click on Events and get registered for the AI for Business Bootcamp with me and Mike in Columbus on July 16th. Okay. AI pulse, if you're new to the podcast, every week when we do these weekly episodes, we start off with a Pulse survey. This is an informal poll of our listeners where we ask a couple of questions related to the previous week's episod. So last week we said, how should the US Government handle the release of powerful new AI models? This is going to be relevant to today's episode as well. 43% said, light review, but no release delays. Interesting. 25% said let companies release them freely. Okay, so that's 67% who are basically light to no government involvement. Yeah, that is noteworthy. 18% stagger or gate access for security as it's doing now. And 15% halt Frontier releases until safety standards exist. Wow. We would not have releases for a very long time. And that is 15% of listeners. That's. That's a fascinating one. That is. That might. That'd be a good one to ask a broader audience. I'd be really intrigued by that one. And then the second one is, how are you using AI agents like Codex or cloud code in your own work? This one's pretty balanced. 33% not using them yet. 33% for non coding work like research, analysis and writing. 20% they're now central to how I work. And 15% for coding or technical tasks only. All right, so again, you can participate. We'll do another pulse at the end of today with questions related to today's content, but you can go to SmartRx AI forward slash pulse to participate in those polls each week. All right, Mike, as I alluded to up front, we have the continuing saga of the US Government versus Anthropic. So what is the latest?

B

Yes, we do, Paul. So this past week, the Trump administration lifted the export controls it had placed on Anthropic's two newest, most powerful AI models, Fable 5 and Mythos 5, ending a roughly three week standoff that had knocked the models offline entirely. So if you recall, this all started back when the Commerce Department ordered Anthropic to block foreign nationals from its Claude Fable 5 model and the more powerful Mythos 5 over national security concerns. Because that order took effect immediately and Anthropic had no reliable way to verify a user's nationality in real time, the company ended up pulling both models offline for everyone. Now, the government did this because there was this report from Amazon researchers who found a way to bypass Fable 5 safeguards and get the model to identify software vulnerabilities. In one case, it produced code showing how to exploit them. That is basically exactly the type of AI assisted cyber attack that regular regulators had been most worried about. However, now the controls are lifted. They were lifted in a letter from Commerce Secretary Howard Lutnick, who said Anthropic had put appropriate safeguards in place. In exchange, Anthropic agreed to proactively detect and address security risks, to help the government develop standards for future models and to flag any malicious activity it detects. Reporting suggests the actual fix actually came down to them adding this new filter or classifier that blocks prompts aimed at finding and exploiting software vulnerabilities. Now, the company had originally argued these jailbreak concerns that Amazon raised were overblown and that guaranteeing zero jailbreaks was impossible. But they pivoted to reassuring the administration and building tougher safeguards. They even at one point changed the messengers representing the company, with co founder Tom Brown taking over White House negotiations. Frank from CEO Dario Amade, whom officials reportedly found harder to deal with. So Fable 5 is now back for users globally across Anthropic's platforms, Mythos 5 is still restricted to those trusted US organizations and partners, but they do have access to that as well. So, Paul, Fable 5 is back. It is worth noting that after July 7th, the day you'll be listening to this podcast, the consumer plans that are kind of flat fee licenses can only access it using usage credits only. So we could talk a little bit about that. But if at some point you're like, why don't I have access to Fable? Well, it's because you have to pay to play, to use it, and pay as you go. But Paul, my question is about kind of the bigger picture, broadly. Like, what have we learned from this, if anything, what does this mean for the relationship between AI labs and the US Government, or at least between them and the Trump administration moving forward?

A

I'm not sure that we've actually learned that much, Mike. It's pretty opaque, actually, what's been going on. So I'll try and in their own words, I guess in the government's own words, like break down how the resolution arrived. So first, I think it's important to remember that Fable 5, which is the version that you and I, Mike, in theory, can go in and use today, is the Mythos model with safeguards. So Mythos is the more powerful one that the government and Anthropic are using with some trusted partners as well. Fable 5 is the model for the rest of us that was supposed to be safer. And then the government decided after it was out for a few days that it wasn't in fact as safe as Anthropic claimed it to be. So that that was kind of how we got there. Okay, so let's break down how we sort of arrived at the re introduction of Fable 5. So Tuesday, June 30, Susie Wiles, of all people, who is the chief of staff to President Trump, she tweeted. Now, she is not a super active person on X. That's why I say it's like kind of odd that she's the one that's sort of like doing the communications. But anyway, so she tweets at 7:26pm Eastern Time, Tuesday, June 30th. Under President Trump's leadership, the United States is the undisputed winner in the AI race. My gratitude to companies across industries who continue to work closely with the White House to implement the President's executive order called Promoting Advanced AI Innovation and Security. This includes excellent work around advanced model access and guardrail testing and security. The government and private sector have worked together in a way we have never seen before. And. And this foundation of America first is unprecedented. Our shared priority remains get the best tech deployed as quickly and safely as possible. Now, I read this as, like, damage control because there was a lot of allies of the administration who were questioning their approach to what they were doing. And so this was like, from the President's office basically saying, we are still all in on America winning the AI race, basically. So then Howard Lutnick retweets that Susie Wiles tweet, and he says, over the past two weeks, we have worked closely. And this is three. Five minutes later, he retweets this. Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US government and strengthen America's leadership in AI. And then Tom Brown, who you just mentioned, Mike, comments on that and says, thanks for your partnership on this, Secretary. And Sholto Douglas, who we have talked about numerous times, is an Anthropic researcher. Thank you for your leadership on this and governing AI with the gravity it demands. We're excited to get the model into people's hands. Then 20 minutes later, anthropic tweets, we've received notice that the Department of Commerce has lifted export controls on Claude, Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and we'll share an update soon. We're grateful to our users for their patience and to everyone who worked with us on redeploying the models. Okay, so then Howard Lutnick sends a letter to Tom Brown, not Dario, and this is like the formal. It says, since the issuance of my previous letters, dated June 12 and June 26, where he said he was not happy with Anthropic. Anthropic has taken steps in close coordination with the US government to address risks associated with missiles 5 and Fable 5. Among other things, Anthropic has agreed to proactively detect and address security risks associated with the models, which they were going to do anyway. To work diligently with the US Government on protocols and standards and releases for Mythos, Fable, and future models, and to inform the US Government of any malicious activity. Commerce reserves the right to reevaluate the decisions made in this letter and the necessity of reimposing. This is probably important line. And the necessity of reimposing a license requirement should circumstances change or should Anthropic fail to adhere to its commitments. Now, one note here, Mike, is this is a letter stating things that Anthropic and the government have agreed to. This is not saying OpenAI has the same level of agreement. It's not saying Google has The same level of agreement. And Meta, we know, does not have the same level of agreement, nor does Elon, Musk and Xai, should they get back in the game. So this is a isolated agreement based on whatever the relationship is today between Anthropic and the government. So then Anthropic posts a article on their site called Redeploying Fable 5, and I'll just highlight a couple of elements of this. So Again, this was June 30th. So then, you know, we're kind of going back in time here. So when they refer to tomorrow, they're talking about July 1st. So Fable 5 will be available starting July 1st to users globally on the Claude platform. Claude AI, Claude Code, and Claude coworker for Pro Max Team and select enterprise plans. Fable 5 will be included for up to 50% of weekly usage limits through July 7th. This is what you were referring to, Mike. And then viable via usage credits thereafter. So then they outline four key components. So they they in this post, and I'm not going to read all the details. You can go and read this yourself. They go back through the timeline of events that arrived where we are today. The second point is their general approach to safeguards. Now, this one, I'm going to call it a couple of excerpts because I think this is important. So, Claude, mythos 5 can be used to find and exploit software vulnerabilities more effective than any other model. And this is what got us into the position we're in to begin with, is that mythos 5 had these cybersecurity capabilities and all but the most skilled human experts. They say these prodigious cybersecurity capabilities make it uniquely attractive to malicious actors who wish to misuse it in cyber attacks. Fable 5, however, provides no such unique offensive capabilities. This is because we launched it with the strongest safeguards we've ever applied to a model. So, again, Mythos 5 is Fable 5, but Fable 5 has safeguards. The government didn't think those safeguards were appropriate or sufficient. That is why the model was taken away. So it is now being relaunched with these updated safety mechanisms that are supposed to make it more capable of defense. Now, the way they largely do this is something called classifiers. And so, in essence, what happens is they look at the prompts that you give it as the user and try and classify whether you're trying to use it for harmful purposes. If they determine that your intent is harmful, they will shut off access, they will turn off that chat thread and potentially remove your access to the model, period. So all they did, since the government objected and made them take it away, is they strengthen the classifiers to be basically more likely to be triggered. So if you, especially if it's really like biology or something like that that they consider to be Humber or security related, cybersecurity related, so they basically said, hey, this thing might get triggered with false positives way easier now. And that has led to some frustration from people who have actually played around with the model and unintentionally, with no harm meant, triggered these safeguards at Fable 5 prices. Yes. So you paid to be triggered.

B

Right. So it'll route your query to opus 4.8 in a lot of cases, which is having people freaking out.

A

Yeah, it's triggering people basically like, yes, it's not going well. So the reason I'm highlighting all this, Mike, is because this is the state of play moving forward. So like if we get GPT 5.6 this week, or if we get Gemini 3.5 Pro or like whatever the next generations of frontier models are assuming they're on par with the capabilities of Fable 5 Mythos 5, we're in essence entering a period where we don't even know what the government considers to be worthy of, you know, hitting these harmful intent triggers that shuts the model off or reroute you to a weaker model. That leads to the third point in Anthropic's post, which is about a shared industry framework. And that's where they're saying there's no consensus as to how to describe in objective terms the severity of an AI jailbreak, where someone is intentionally trying to get the Fable 5 model to do mythos 5 like things. They're trying to go in and get it to do something that these safeguards are supposed to prevent it from doing. And what Anthropic saying is we don't even agree as an industry as to like, what exactly deems something to be a harmful or high risk jailbreak. And so we need these common standards that we have to work with the government on. Which leads to their fourth point about deeper government collaboration, which is what Anthropic appears to have agreed to be doing now with the government, where there's pre release access and evaluation from whoever in the government is now going to be in charge of this stuff for the time being. Rapid information sharing on safeguards if they're broken. So if like something happens and these safeguards fall, you know, they're not working, then the government can step in again. Dedicated resource for joint research, which in large part I interpret to be Anthropic is going to share their internal expertise and researchers with the government to help figure this out and then work together toward a common industry bar. So then, a couple of final quick notes. Dean Ball, who we've referenced many times recently, he's become kind of a pivotal figure in all this conversation. He is joining OpenAI as of this week. He will be joining them, he said. Related to the June 30 news great news, but we have no idea what Anthropic did to make the models safe, what commitments Anthropic has made going forward, and and whether or how any of this applies to other frontier models in the government's licensing queue. We know GPT5 is in that queue, but it's fair to assume that other model developers are at least in early stages of submitting their models. The opacity will not lend itself well to a stable, investable, trustworthy industry over time. But the US Government needn't figure all this out in a day and a two week review. Timeline is not insane in the grand scheme of things. Status quo is not tenable. But we made progress. The Anthropic models are available. Progress is just a first step. Worthy of applause and then a couple of my personal final notes here is Impacts on guardrail of guardrails on model performance. So I touched on this already Users, if you're using it for more advanced reasoning, strategic planning, if you're using it in sensitive areas, regulated industries, things like that. The reality is we may be dealing with lots of false positives and frustrations when these safeguards are triggered with no course to figure out why or like get it resolved right? The government, Anthropic and their trusted partners have access and will continue to have access to more powerful models than the rest of us. That is going to be a fact of life moving forward. How this impacts approval release of future models. My guess is things are just going to slow down on the Frontier model release cycle. At least for us external users. The internally it might just keep going at the same pace. They may be on GPT 6.5 by the time we get GPT 5.7. Like who knows right? If the guardrails break the government has already alluded to and Anthropic reinforced in their own post, you could lose it again. So you you may become dependent on Fable 5. Pliny the Liberator on Twitter made like jailbreak the thing in a harmful way or someone may find a universal jailbreak and all of a sudden Fable five is gone again. That that also is just the reality. And then which kind of will lead us into our second topic, Mike, today, is will more enterprises just move off of these proprietary frontier models? Will they move to more open source models where they can train and control the models on their own with their own data, their own application layer? And so these are just like some of the many open questions that this has created. So, yes, we have Fable 5 back after three weeks of drama, but we don't really know much. We just know they made the safeguards more likely to be triggered due to these classifiers being more aggressive. That is like the gist of what we learned.

B

I confess I wasn't even able to get excited about Fable coming back. Like, I used it a bunch, but I'm like, I'm not gonna pay for this on a usage basis moving forward. So it's like, I don't even want to get used to it at this stage. It's too good.

A

Yeah. I mean, and I was on PTO last week and I like, honestly, I was like, I don't even have time and it's going to be gone by the time we get back to work on Tuesday. So yeah, yeah.

B

And I realized we could still like, access it and use it, but it's like if you've planned historically around your flat fee license versus usage base, that's a whole different animal. If you're not doing that already. All right, so let's talk about that final point you mentioned about kind of, what do enterprises do about this? What about open source models they train and control on their own? Because in our second big topic today, we had Palantir CEO Alex Carp. He went on CNBC Squawk Box this past week and basically unloaded on the big AI lab. So he was on there to kind of talk about some news about a new Palantir and Nvidia government deal. We could talk about that in the context of this, but it kind of turned into this, like, diatribe, this, like, fiery argument for what he's kind of calling AI sovereignty. So this idea that companies and governments need to own their own compute data and AI models instead of renting them from the likes of OpenAI and Anthropic. So his argument is that when a business runs on someone else's model, he was saying, these people are stealing the weights and alpha of my business, meaning the labs are kind of quietly absorbing your data and your competitive edge while you pay them for the privilege. In his argument, he claims enterprise leaders are privately livid over this, convinced they're paying for tokens that don't create value on national security. He argued the US cannot hand its defense over to Frontier Labs. He said, quote, are we really going to outsource the battlefield of this country to the consensus view in Silicon Valley? That is effing insane. He asked point blank if he was angry about this. And he said, look, this is the voice of American business channeled through me, and I'm telling you it is absolutely a problem for this country. He challenged skeptics to call any CEO in private, saying, quote, they're twice as livid as me about all this. He also basically claimed that AI labs are charging enterprises way too much, roughly three times what they should, and that the models have been, quote, irresponsibly oversold. So he's talking about all this kind of in the context and we can get into this a little bit. Palantir has this product called Ontology, but Karp kind of uses that word almost more as a concept and he's basically saying, look, we're the software layer that sits between the raw AI model and your business, which keeps the model from caching your data, copying your business, or walking off with your ip. Now, he had some nice things to say about Anthropic's Dario Amadei. He called him a historic figure. He did not like undersell how powerful Frontier models are, but said there's so many problems right now due to this thing where the labs are essentially in his argument, capitalizing on the fact everyone's sharing their data and their alpha, as he puts it, with them. So, Paul, I'm curious, like Karp, let's be very clear, has an incentive to say all this. They sell the very thing that solves this problem, in his words. Palantir and Karp, we should note, are also quite polarizing. But I'm curious if you find like underneath his kind of shtick, his theatrics, like, is he onto something real here and a real concern?

A

I mean, he's definitely a figure where you, you do have to not set aside, you have to understand the context of who he is, how he behaves and talks, what his motivations are. So. But it's not unlike a lot of CEOs, once you get past the bluster and the energy of the interview and you have to try and say, like you're asking, like, what is there something relevant here though, like, is he making some valid points? Not every point is going to be worthwhile. So I actually just took a quick step back. I was like, who is Karp? And what is Palantir? Because it is a company that's very, very relevant to today that a lot of people don't know about or don't understand what they do. So just a little background. So if you go to their homepage, the. At the very top, above the fold is AI powered automation for every decision. So that's kind of how they position what they do. Their overall premise is that the models aren't enough. You have to have data integration, governance, permissions, workflow, design, logic, and deployment into real operations. Now keep in mind, they are in most cases discussing highly sensitive implementations and integrations of AI. They are not necessarily focused on the more standard or mundane applications like marketing, use cases, sales, customer success, you know, hr. Like more of our listeners, we think. I mean, again, as a podcast, you don't really know a ton about your listeners, but based on the people Mike and I talk to, they're generally more business leaders and leaders of like, those kinds of departments. They're, they're not necessarily often the people that are doing these highly sensitive infrastructure integrations of AI technology. So again, we have to keep our own audiences in mind as well. So, real quick background on Palantir. The company was founded in 2003 by Peter Thiel of PayPal fame. We've talked about Thiel many times on the show. Alex Karp, Joe Lonsdale, who's very outspoken on X, as well Stephen Cohen and Nathan Getdings. So Palantir's roots actually trace back to work on PayPal's fraud detection systems, which inspired the founders to develop software capable of identify identifying suspicious patterns without compromising privacy. So they say the firm gained prominence by working with U.S. intelligence and defense agencies, building tools designed to analyze vast majority volumes of data. Over time, Palantir expanded into commercial markets, applying its platform to everything from supply chains to healthcare. I didn't know this little fact, Mike, but maybe you did. Teal named the company Palantir after the mythical seeing stones in J.R.R. tolkien's Lord of the Rings that allowed its holders to see across great distances. For Thiel, the name symbolized the company's mission to enable its users to see hidden patterns across immense volumes of data. During its early startup phases, Palantir struggled to attract investors. And executive adventure. Capital firm Kleiner Perkins reportedly told the founders their company was doomed, a moment that has since become part of Palantir lore. The funding originally actually came through breakthroughs with the U.S. central Intelligence Agency. So again, keep in mind this is a company that was built to support intelligence gathering in the United States, which is why a lot of people have always been a little bit questionable about how much access Palantir has to data on human. On American citizens. Not even just American citizens globally, but mainly American citizens and how that data is being used. So part of why Carp? I mean, Karp's always exaggerated, I would say, in his comments, but Palantir has a market cap of 309 billion. This is a big company.

B

Yeah.

A

But their stock has kind of gotten hit like other SaaS stocks over the last six months. It's down 28%. So in May 2026, just two months ago, he sent a letter to shareholders and I'm just going to read like three quick lines here. We believe it is not hyperbolic to say that nearly all that, nearly all AI workflows that actually create value, especially on the battlefield, are built on Palantir. So they're very confident in who they are and what they do. Our financial results now demonstrate a level of strength that dwarfs the performance of essentially every software company in history at this scale. We stand on the walls, sentinels of the inner sanctum against the assault of AI slop. Just to give you a sense of, like, how he talks. That was his words to shareholders. Okay, other context. If you'll recall, when the Anthropic's issues with the government started, it was because Anthropic questioned the use of Claude in the bombings in Iran. That was through Palantir. So Claude's the Claude model was being used by Palantir in the battlefield and potentially for other uses that Anthropic wasn't sure about. And the administration did not appreciate Anthropic asking. And that is where this beef all started. So they're very relevant to this. And Palantir's tools are deeply tied to the government. Defense, intelligence, law enforcement, things like that. Okay, so then a couple of quick notes on the interview. He talks about the Nvidia deal. It's 19 minutes, by the way. It's put it on one and a half speed. Although he talks really fast, watch out. Maybe like 1.2. He is very unhappy with the Frontier Labs. Now, again, keep in mind, his business is built on the top of the capabilities of the models that these Frontier Labs create. They talk about needing to make LLM safe, useful and precise. It talks about it being extremely important in government regulated industries and critical infrastructure. His big thing is the labs getting the IP based on the prompts that you're using, the outputs that are being created. That's what he's Worried about is that they're stealing your iPad, which is the alpha that that you as the enterprise user should control the compute, the models, the data stack and the competitive advantage. He said the enterprise leaders are, are livid. Was his quote, models are being oversold. And then as you said, Mike, he's the voice of American business being channeled through him. However, enterprises love them. Their FDEs, the Forward Deployed engineers and ontology is what's like making all this work. The ontology as you talked about is that operational layer that's on top of the models he talks about. The three fundamental components are model plus application, later ontology, and then compute. He says they are completely agnostic on models. They want more frontier models. So they're big fans of open source. They have more demand than they can supply, which is trying to get their stock price back heading in the right directions. He said what in. He literally said they asked him about their stock price. Markets are short term and very fickle. And then quote, what investors actually understand tech. That was pretty funny. Okay, and then just the final notes here. So Palantir then tweeted sort of a summary after this interview of trying to like consolidate his key points. So one, your AI sovereignty dictates your institution's future data retention is your treasures too. And they specifically said transferring data hands over access to your pre existing winning plays and yields the means of production for new ones. Meaning you're giving the labs everything you're doing token max. 3, token maxing hijacks your value orientation and decreases your institutional fortitude and intelligence. That's a lot of words. Four, controlling your weights is controlling your fate. Five, and this one I'll zoom in on for a second. There is no contradiction between sovereignty and alpha. So again, in their context, alpha means proprietary advantage. So in the sentence where they're saying this, like the best architecture is the one where an institution owns its knowledge, workflows, data judgments and operating logic. So sovereignty is that institution of control of the data, models, workflows, permissions, tribal knowledge they talk about is the informal expertise inside an organization. Alpha is the unique advantage. So these are like their words, the things that he keeps saying. Six is politicizing technical issues involving sovereignty is what your adversaries wants. So he's not a big fan of what anthropic is doing. Basically real expertise is existential. Number eight is learn from institutions that are winning or have consistently delivered. And the final one was only listen to institutions, countries and people who have proven record of being right, which is them basically. And so again, like my, my main takeaway here for our audience is there is relevance to what he's saying. There is an absolute argument to own more of this, to own more of the computer, the data, the application layer they talk about so that your prompts don't go back to the labs and they don't steal your ideas and things like that. But most of the time they are referring to highly sensitive technical applications of the AI. I don't know that like marketers and salespeople need to share these same concerns. It's a relatively commoditized thing, so it's just something to keep in mind. Super relevant, probably more like the CIO audience that is like thinking about this and highly technical CEOs in very regulated and sensitive industries I would say is like mainly who he's talking to, the people who are just trying to spin up marketing use cases. I don't know that you need to share the same level of concern and urgency when it comes to how you're using these models.

B

That's fair.

A

Mike. I don't know.

B

Oh yeah, I would absolutely agree with that. It's definitely much more of a specialized enterprise concern. One thing I did want to kind of emphasize here that I did find useful, which I'm saying this because I don't often always find things that David Sachs posts as particularly personally useful. But I will say he had a good summary of why this is such a concern sometimes because he said as part of a larger tweet, as Karp explains, technical customers want, quote, control over their compute their models, data, stack their alpha. They want to know they own the means of production and it's not being transferred to someone else. And he, and so Sachs then says, don't think that can happen. Just look at Figma, the design firm. Because according to the information, Anthropic blindsided its then business partner with the launch of Claude Design. Figma's founder said Anthropic had not been consistently honest with them. Anthropic's chief product officer had even served on Figma's board until three days before the launch of Claude Design. Figma's stock has fallen sharply this year while Anthropic's valuation has surged. He's got his own ax to grind with Anthropic, but just kind of this idea that some people see these model companies as stepping to the corner of a lot of SaaS businesses using perhaps some of the data or alpha, so to speak, that these companies have shared with them. I'm not sure how far that Is true. But that's one of the big concerns.

A

It's 100% true. And it's not even like, I don't even know why this would be debatable, because. Right, we've talked about this, that these labs for the valuations that they have, and they will have when the IPO and the amount of money they're raising that they have to go after all human Labor. So the SaaS industry is like 3 to 500 billion a year in revenue. The. The cost of wages in the United States for knowledge workers is like 5 billion or 5 trillion. Trillion.

B

Yeah.

A

So, yes, of course they're gonna go after all the viable markets now, but that's not different than what, you know, Apple's done to apps in the history or what Meta has done to everybody throughout its history. Like, big companies look and see what's working and they either try and buy up the companies that are doing it, or they just launch a competing solution. Now. It's just funny to me that the people in the orbit of this administration talk as though Anthropic is the only company doing this. Like, they're all doing it and they're all going to continue to do it. So, yes, it is. And I agree. I saw that tweet from Sachs and I was like, that actually makes sense. So, yes, I. It is the thing, if you, I mean, Sam Altman said this two years ago, he's like, right, if your product doesn't get better, every time our model gets smarter, we will steamroll you. Like, like our model, our generally capable model will just do what your, you know, silly little SaaS company does, or we'll build it to do that.

B

Right, Right.

A

So they're not even hiding the fact that that's what they're doing.

B

Not at all. All right, so our third big topic this week concerns something you shared, Paul, in one of your exec AI Insider newsletters a couple weeks ago. So you had started publicly sharing a project that you spent considerable amount of time, at least the last few months, building basically a new set of AI transformation frameworks and tools and organizations. So I wanted to maybe just turn things over to you to walk through what we're talking about here, kind of these pillars of business AI transformation you've been sharing about publicly now.

A

Yeah, so this is again, like, it was kind of an impromptu thing that I decided to share this. I. I write the newsletter on Saturday mornings usually, and Mike is gracious enough to take his Saturday morning and get it up and, and schedule it to send for me on Sundays, but I usually wake up at like 6:30 or 7 on Saturdays. And then I write the newsletter based on the key things from the week and then kind of what we're looking at to talk about in the next episode. So for the editorial. So I always write an editorial to start it off. I was like, I'm not sure what to write about this week. And then I thought, you know what, I'm just going to start sharing like elements of this idea. So basic, the basic premise. I'll just kind of give you the background. So I started Marketing Institute in 2016. We created our marketing AI conference in 2019. That was the first year we did that event. 2020 Covid hits and we couldn't hold the conference, but we already had all these speakers lined up. So we had like 30 speakers. So we got all of them to agree to turn what would have been their in person talks into courses for an AI Academy. So version one of AI Academy was launched in 2020 with the talks that would have been given at our conference that year. I'd always wanted to build an AI Academy. We thought it was going to be essential. We knew that lack of education and training was a, a fundamental roadblock to adoption within enterprises, which our own research has validated year after year. It is the number one roadblock. So we'd always intended to build this online education platform and that was just a forcing function. 2023 we launched the first certificate series piloting AI, which I've now recorded. We're on the third edition of that one. And then in 2024 I created scaling AI, which is a course series and certificate. So Academy basically had two fundamental course series and certificates, and then a collection of other resources, including some lives like trends briefings that Mike and I do, quarterly trends briefing, ask me anything sessions, things like that. So then in 2025 I spent the majority of the year reimagining what AI Academy was going to be kind of when it grew up, in essence that the pace of change in AI was accelerating so fast, so many companies were behind in terms of their ability to understand and adopt it. There was just a complete lack of AI literacy among knowledge workers that we saw as just a fundamental thing we needed to solve for. And so that was really where I spent a lot of my time. We relaunched AI Academy in fall of 2025 with a new AI powered learning platform. Our focus at SmartRx became this idea of accelerating AI literacy and transformation. So I would tell our team like, hey, we don't. Academy doesn't sell courses, we sell and power the idea of transformation. So, like, what do we do to do it? And so like, I would literally tell our salespeople, don't sell licenses to businesses until we know what success looks like. So the question became, okay, you want to come into Academy and buy 50 licenses or 500 licenses to level up your team and create these AI forward professionals, how are you going to know if it's working? Like, what is 12 months from now look like to you if Academy helps you? And so that's been my focus is what does transformation mean to individuals and organizations? Because we have thousands of individuals who are members and then we have thousands of Academy members who are business account licenses. And so how do we establish a baseline and then how do we create an AI transformation system that actually helps them achieve the true outcome, not just leading indicators, like people took courses and got earned some certificates. So the last few months in particular, starting mainly in February, March, I have been working on these frameworks and tools. Now I have, in a veiled way, a lot of the use cases that I've shared on our weekly, where Mike and I share what we're working on, was things related to this. I just wasn't ready to talk about it yet because I hadn't built all the things yet.

B

Right.

A

So we basically got to the point week or so ago where I was like, you know what, let me just start sharing elements of this because I think it's going to be helpful to people as we're kind of building out the rest of this. So when I think of this whole transformation system, there's the upfront planning component where you're doing assessments, you're building blueprints that actually drive transformation based on where different people in the organization are at. So imagine like beginner, intermediate, advance, let's just say, Broadly speaking, take 500 employees who need to get upskilled, break them into these groups, and then you develop cohorts to do personalized learning journeys for these people. But it can't just be courses. Like, you need communications plans, you need handholding. In some cases, you need more interaction, like office hours. Like, it cannot just be on demand courses. There's already great providers who offer just on demand courses. We wanted to do something much more than that. So like our AI Academy lives become a really key component of this. Where there's trends briefings, ask me anything sessions, mastery classes, transformation spotlights. We're working on ideas to extend the podcast, like exclusive content just for members. So all of this has been kind of running through my mind. But I'll kind of get to the gist of today's main topic, which is how do you, at an organizational level, determine where the company currently is? So for our listeners or viewers, if you're watching on YouTube, think about your own company and say, well, how do I even know how far along we are? Like, how mature is our organization from an AI perspective? And what do we actually have to do to figure out what success looks like and then to get there? So that's like the basic premise of what I've been trying to solve for. And so I, I spent a. I don't know, this was probably like two months because this was like the core of the whole thing was organizational maturity. And then how do you actually drive transformation? And so I, I used Claude intensely for this. So like when I was, you know, talked a lot about like Sonnet 4.6, that was the main model I was like working with. But it was just these iterative strategic planning sessions, for lack of a better way of saying it. I didn't have AI have Claude write the stuff. It was like going back and forth and like, here's what I think and I would build an outline and I would like adapt it. So I don't know. Long story short, because we'll tell much more of this story as we move forward. I landed on eight pillars that define business AI transformation. And so imagine each of these pillars taking and saying, okay, where are we within this pillar? And then what do we have to do to get to the desired state with each of these? So I'll just go through these eight and then we'll kind of move on and we'll come back to, you know, future topics and other episodes. So the first is vision. AI transformation starts with leadership clarity. This pillar examines whether your organization's leaders have built the shared understanding, vision and cultural foundation that makes sustained AI transformation possible. The first criteria or the first element of these, because I have like 67 elements that all tie to, you know, throughout these eight things. The first is do your leaders actually understand the moment? Do, do. Does the C Suite actually understand the capabilities of these models and where they're going in the near term? Because if the CEO isn't AI forward him or herself, then you are going to go nowhere. Like, you can have all the department level innovation you want, but if the C Suite doesn't mandate this as one of the top priorities in the organization and understand why they're mandating it, then everything else is going to fall apart. So you have to have the vision. The second is strategy. It lays the groundwork to turn ambitions and vision into action. This pillar assesses whether your organization has translated its AI vision into a clear roadmap with dedicated resources, defined structures, and aligned partnerships that enable real progress. Third is data. Alex Karp would like this one. AI is only as powerful as the data behind it. This pillar examines whether your organization's data is ready for AI and whether the right guardrails are in place to use it responsibly. Fourth is technology. Access to the right tools is a prerequisite for transformation, but access alone isn't enough. This pillar looks at whether employees have the tools they need, the support to use them well, and the rigor to choose and evaluate them intelligently. This is one of the biggest flaws, Mike. You and I see it every day. Yep, companies just give people access to Copilot, ChatGPT, Claude, or Gemini. And they're like, here, figure it out. Here's a few use cases, a few prompts. So that is not enough. It is like, and that is more and more I see that where they just, they give them the stuff and they don't teach them how to use it responsibly and in an effective way. Fifth is governance. So responsible. AI is the foundation of sustainable transformation. This pillar assesses whether your organization has the policies, oversight, accountability structures, and transparency practices to use AI safely and ethically. Six is literacy. AI education and training are the foundation for success in every organization. This pillar examines whether employees have the knowledge, skills, behaviors, and mindsets to work effectively with AI, from everyday tasks to advanced applications. Seven is people Business. AI transformation is ultimately a collection of personal transformations. This is why we also have a separate assessment and pillars for individual transformation, which maybe I'll get into in one of the upcoming episodes. So this pillar looks at whether the organization is measuring, developing, incentivizing, and supporting its people in ways that make AI adoption impactful and sustainable. And then the final one is performance transformation must produce results. This pillar measures whether AI is generating tangible, measurable business impact, from operational efficiency to revenue growth to new forms of value creation. I have yet, just so like, people don't feel like, oh my God, I'm so behind. I have yet to talk to a single organization that would score highly across these eight pillars.

B

Right.

A

Like, maybe one or two of the pillars, but like, it's very aspirational at this point to think that an organization truly like, if someone rated highly across all these areas, then they're in the transformation phase, as we would call it. And so, again, like, we're, we're going to be releasing a lot more capabilities and tools and frameworks related to all this stuff. But because so many people are struggling through what does a transformation actually look like, I wanted to at least share some of the foundational work we're doing here because I think it'll give people a little bit of a head start. And then if you're one of our AI Academy business account customers, we actually have already built some of the tools related to this and we're going to be opening up beta access exclusively to AI Academy business account customers. So reach out to us if you are one of those, or if you want to get early access and you want to become a business account customer, you can just go to Academy SmartRx AI and kind of learn more about business accounts. But we're going to do a soft rollout of this stuff in the near future to some of our business account customers and then, you know, full public rollout later this summer, early in the fall.

B

Yeah, and one small thing I'll add here, Paul, is that obviously I'm biased, but I really like how seriously this takes the concept of transformation. I think it's really easy if you're in kind of the AI bubble on X, which is where a lot of the news breaks, a lot of the commentary happens. It's easy to see these people talking about like, well, you just need to move faster. Maybe you need to fire half your people and get the right people. Maybe you need to just, just be lying awake all night dealing with agents so you can learn how to transform and accelerate. And like, that all might be true, but it doesn't take very seriously the concept of how much goes into actual transformation. It's not just about posting on X that, hey, we're moving faster, firing up 30 agents. It's like organizational change. Change management is a whole animal.

A

Yeah, I think that works in AI native companies that are smaller and nimble and able to, to be super disruptive at all times to all phases of their company. Yeah, enterprises can't do that. Like, it just, it doesn't work that way. There's too much friction. There's too many legacy systems, there's too many guardrails in place for usage. You have to take a methodical approach now. You can still have that innovation happening. You can still have pockets of the company that are moving really fast, centers of excellence that are experimenting labs, things like that. But for a big enterprise to do true transformation, there has to be a system in place to do it. And that's what I've been trying to build. And I'm we're close. Like, I'm really excited to get some of this stuff out. And I've, you know, sort of had some initial conversations with some large enterprises and showed them what we're working on. And I think it's going to be really impactful. But I, I also just see it as an essential way to think about the future of work and business.

B

Absolutely. All right, before we move on to our rapid fire topics, this week's episode is also brought to us by Macon, our marketing AI conference. This is an AI conference for marketing and business leaders happening October 13th to the 15th in our home base of Cleveland, Ohio. Macon is a leading industry event with three days of keynote sessions, workshops and conversations built specifically for marketing and business leaders who are actively figuring out how to adopt, operationalize, and scale AI across their organization. So go check out the agenda at Macon AI M A I C O N A I where you can also register. And if you do want to register, now is the time. Prices go up in a couple weeks. You can use the code POD100 at checkout and save a hundred dollars on top of locking in your best rate available. So go to Macon AI. All right, Paul, let's dive into some rapid fire this week. So first up, the Financial Times reported this past week that OpenAI has floated giving the US government a 5% ownership stake in the company. So at OpenAI's recent post money valuation of $852 billion, that stake would be rough, worth roughly $42.6 billion. CEO Sam Altman has reportedly been pitching the idea directly to the Trump administration. He's framed it as the best way to share the upside of AI with the public. According to the reporting, he has raised it in conversations with both Trump and Commerce Secretary Howard Lutnick, as well as Treasury Secretary Scott Bessant. Altman has also suggested each of America's leading AI labs, potentially including Anthropic, Google and Meta, hand over a similar 5% slice to a government vehicle modeled on something like the Alaska Permanent Fund, which actually invests the state of Alaska's oil wealth and pays out dividends to residents. It is not clear that any of the others would actually go along with this. CNBC has reported that the administration and Anthropic, for instance, have not actually discussed such a stake. Now, obviously, AI labs are facing a much tougher environment in Washington. There's growing public and political concern over the data center build outs, AI impact on jobs and of course, cybersecurity. Obviously, Anthropic and OpenAI have recently had their own models held back by the US Government. So, interestingly, there is maybe some precedent for something like this. The Trump administration took a 10% in intel last year after an $8.9 billion investment. They've also taken stakes in IBM and several quantum and critical minerals companies. Trump has actually called the U.S. taking ownership stakes in AI labs, quote, a beautiful thing that would make Americans partners in this revolution. For now, though, these talks are just described as conceptual and early stage. Any actual deal would probably, people speculate, require an act of Congress. So, Paul, I'm curious, like, what do you think of this idea? How likely is this to actually come to pass?

A

I think we're going to see a lot of ideas floated in, you know, the coming six months to 12 months. I don't know that something gets done. You know, I. We've seen some success with his Brad Gerstner and what became known as Trump accounts that, that Gerstner created. The basic premise is to be able to, like, give or gift you know, money to American youth and they. That money grows over their lifetime. I. I don't know. I mean, I think when I think about what is a viable model, I keep going back to what Gerstner has led the charge on. I mean, we just saw SpaceX this morning announced that they were contributing stock to that. That was like 2% of SpaceX stock or something was getting contributed to the Trump accounts. Dell has given money to the Trump. I think that there's something there that we may eventually get to a point where there is some model established. It seems to make sense. I have no idea what it's going to end up being and how you're going to avoid

B

fraud.

A

And what works for one administration maybe doesn't work for the next administration that comes into power, and that's where it gets really messy. So I don't know. I mean, I think this is going to be a recurring topic on the podcast for a while, and I do think eventually something gets put in place. I just don't know what it's going to look like.

B

Well, something would probably go a long way towards alleviating some of the public's outrage about some of the AI decisions being made, like we've talked about.

A

Right. Well, I mean, right now it's just all going to happen through super PACs. You're going to have leaders of certain labs personally don't personally donating, you know, 20 million here, 100 million there. And their models magically get fast tracked over other models. Like that's how it's going to work now. It's probably how it is working now. So I would rather there was some more transparent process for how this is happening versus some labs being favored over the others because they're better at, yeah, working with the government or give more money to the administration, things like that.

B

All right, so next up, the information reported this past week that OpenAI engineers have actually found a way to more than halve the cost of inference, which is the cost of actually running models to answer user queries and do stuff for them, as opposed to the training costs when they actually go and train models. So this is interesting because it was a pure software breakthrough, apparently. So the engineers were able to get far more out of the same Nvidia chips OpenAI already has, rather than buying or building more chips. Interestingly, when the new techniques that they have developed were applied to Power Chat GPT for logged out visitors, which is people using it without an account, the number of GPUs needed to serve all that traffic at one point reportedly dropped to just a couple hundred GPUs, which the reports have called a shockingly small number. Now, OpenAI hasn't said how this pulled this off. They probably will never say how they pulled this off. But the report from the information and some analysts point to a mix of well known efficiency techniques. There's quantization, quantization which trims the numerical precision of a model's calculations. There's smarter caching so the model doesn't redo work. It's already done. They batch many queries together to keep chips fully used and route simpler questions to lighter, cheaper models. Now, Paul, the reason we're talking about this is we've talked a lot in the past several weeks about token usage, AI budgets, constraints on compute. This is one of the things we've mentioned that if it's it works, as the reporting seems to indicate, is how the bottlenecks of compute get solved. We make the models dramatically more efficient at getting more out of the chips we already have. How are you looking at this?

A

Yeah, so first, the information is an exceptionally reliable source. We cited a lot on this podcast. They very often break news from inside the labs of what's coming. Back in the day when you know, the reasoning models were being developed, they were, you know, constantly talking about Strawberry and the inside information about what was going on. With an open eyes lab, we know that one of the ways to make these models more efficient and capable is algorithmic breakthroughs. Like, doesn't you have to necessarily train a bigger model all the time? Sometimes you can just get these breakthroughs with better algorithms within how the models are served and how they function. And then the other thought I would have is like, if OpenAI has done this, maybe they got there first. But the talent moves so rapidly between these labs. And all of these researchers hang out at the same parties in Silicon Valley and like nothing stays a breakthrough within one lab for a long period of time. And often what's happening is the labs are simultaneously working on similar breakthroughs. Like, you hear something showing promise and it happens to get to some, a friend at another lab and all of a sudden that lab's also working on it and three weeks later somebody else launches the same thing. So the good news for all of us is more efficient means more, you know, cost effective, which means greater intelligence in our businesses at a lower price point, which is what everybody's so aware of at the moment with the cost of tokens. So, yeah, interesting thing to track. Would not surprise me at all if this is all true and they did find a breakthrough. And it also wouldn't surprise me if other labs are close behind them on this or ahead of them and just weren't talking about it.

B

Yeah, maybe.

A

Yeah.

B

I wonder also if or how they'll press the advantage. If you're open AI, it might make sense to slash pricing just while anthropics kind of got their back against the wall with Fable and the government. So interesting.

A

And we are expecting 5.6 this week. And I agree. I would, I would say that it would not surprise me at all if to take advantage of Fable 5 being kind of pulled back and only available at the usage costs that OpenAI doesn't try and like undercut them with cheaper models.

B

All right, next up, we got a new study from the finance company Ramp and Revelio Labs that has quickly become kind of highly cited piece of evidence for all the people arguing that AI is not killing jobs. So in this research, RAMP used its corporate card and bill pay data to see which of more than 21,000 US firms are actually paying for AI and then match that against their headcount. And these findings start to challenge this narrative that AI is leading to job loss. Because RAMP found that the companies that invested Most heavily in AI grew total headcount by about 10% in the two years after adoption, and they grew entry level headcount by 12%. While lighter adopters saw no statistically significant change, the gains showed up gradually, roughly six to 12 months in and they were broad across engineering, sales, administration and customer service service. Now, the authors of this study are just upfront about the fact these heavy adopters are a selected group. They're already larger, more technical, faster growing more likely to be venture backed. Interestingly, this optimism just runs straight into the debate we've been talking about. There's a wall of data also pointing the other way. A New York Times analysis this past week laid out how by several measures, AI already appears to be driving up unemployment among recent college grads. It may have already destroyed tens of thousands of jobs. We've talked in the past about one of the starker warnings out there from Stanford economist Eric Brynjolson. He used a few months back ADP payroll data to find a roughly 16 drop in employment for workers age 22 to 25 in the most AI exposed job since Chat GPT launched. He called that a canary in the coal mine. So Bloomberg also separately reports that the tech and finance sectors are now shedding around 28,000 jobs a month. So Paul, just curious on your take here. Definitely interesting research like cool to see how widely applicable is this in your opinion?

A

Listen, there's everyday articles on both sides of this debate. It's becoming increasingly polarized. The one side who are the jobs aren't going away side, let's say that like it's only going to create more jobs. It seems like there's some unwritten rule that you have to start a tweet with narrative violation like, like before you share what you're going to say. So normally if I see narrative violation, it's sort of a tip that they're very polarized in one direction and I don't really put a lot of weight behind the rest of the tweet. But the basic premise of narrative violation is that the narrative is that jobs are going away. And a study like this violates that narrative and it's the opposite is true. Whatever. Okay, so I did not spend a ton of time on this research. I will just state what seems to be quite obvious to me, which is fast growing firms hire people. That has always been true. So if you're a fast growing firm, whether you're using AI or not, you're going to hire people. The question I have, and I don't know that the research addressed this, Mike, is are you hiring at the same rate you would have three years ago?

B

Right.

A

So, okay, you're growing 40%, you've added 30 employees, 50 employees, 100, whatever it is, three years ago, would you have added 300 employees instead of 30. If we don't have the answer to that question, then this research is basically meaningless. Like it doesn't tell us anything. It, it just confirms that fast growing company is hiring people. Okay, great. Now I'll add one other piece of context because this is a rapid fire item. Today I'm in the midst of listening to this brilliant episode of 20 VC with Harry Stebbings. This is like one of my, my favorite podcasts and it's with Nikesh Arora, who's the CEO of Palo Alto Networks. I'd never actually listened to an interview with the guy. He's brilliant. Like, it's a great interview. They're a cybersecurity company. Big, big publicly traded company. He has an insane resume. So I'm just going to zoom in on like the one time Harry actually asked him this question about is AI affecting work? Like, what are you doing at Palo Alto Networks? How is an actual CEO of a tech company that is growing? How do you think about this? And so he said the challenge right now is 90% of the enterprise employees are not AI savvy, which is what we talked about. AI literacy is very low within these companies. They have to learn. I can't send them to university. There's no course you can take at any school anywhere to, to learn to be AI forward, basically is what he's saying. They have to be able to learn on their own. I think we're back to a Darwinian moment where everybody has to figure out who's really good. Now you have some people like Brian Armstrong, Coinbase, Jack Dorsey go out and say, I'm going to decimate my organization. So existing companies, like, I'm just getting rid of everybody. So he's saying, I'm going to start building from scratch. They've gone to some version of 30, 40% less people because they figured out there is no redemption. I can't train these people. So what he's saying here is some CEOs are like, screw it, I'm just going to get rid of all the people who don't want to use AI or aren't any good at it and I'm going to cut. And that's why we, we hear about these massive reductions. He then said, I'm just going to find people who are going to come in and help me do this stuff. That's one model. The other is more gradual. So then he talks about his experience at Palo Alto. We've been hiring people only through hackathons. Now we see natural attrition of 2%, give or take a month. We just replace them with people who actually are AI savvy from these hackathons. So they're using hackathons as a farm system to find AI forward employees and then they just hire those people. Give me 12 months, that'll have transformed 20 to 25% of my team. Give me three years, I'll hopefully have enough AI savvy people working at Palo Alto. There's two different ways to get there. So what he's saying, I don't know if this is a narrative violation or not, is like, I'm only going to have AI savvy people in my company right now. If that means I have to get rid of 20 to 25% of my workforce, I'm going to go hire AI savvy people. Now. They're going to be able to do the work of three to five people or ten people or whatever. So I'm going to hire fewer of them. But three years from now, I will only have AI savvy people. And so this is what I've said time and time again for years. If you are not AI savvy, if you're not, as we would call AI forward, you don't have a job in three years. Right. I don't know how else to say that, other than to be very blunt, that the only people that will be working in enterprises in three to five years will be AI forward professionals and leaders. If you choose to not participate in that economy, you won't have a job. And so if that means we net out positive because growth happens so dramatically across the US that or the world that everyone has jobs and we create millions more jobs, wonderful. I hope that's the case. I don't see that happening as fast as the displacement that will come from companies led by CEOs like this who are saying, I only can function with AI savvy people, but I'm going to be, I'm going to do it gradually, I'm going to do it in a human centered way. I'm going to give people time. We'll do it through natural attrition. But three years from now, only AI savvy people work here. That is like when I go back to my business, AI transformation pillars. Yeah, that is the prototypical way to do this. Give people time, provide them with training and education, give them the technology, personalize use cases, let them self select that they want to be part of the future or they don't. And I will stop now because as you Know, Mike, I could talk about this for hours and it's only a rapid fire item.

B

Well, one last thing here, Paul. It does seem like two stories can be true at the same time. On one hand, you can have new AI native or AI focused jobs that are compensating very well, create huge opportunities and value for organizations. But also that's not, not a lot as many jobs as the ones that are lost due to massive productivity gains or at least reorganized or people that don't want to use AI. So it's really interesting to kind of see these almost divergent paths, I would say.

A

Yeah. And again, like, think about like, so these, the, in the US the enterprise companies account for. I don't know, I remember the percentage off the top of my head. So I'll just say, let's say we'll just make an assumption. 50% of all employees work at these large enterprises.

B

Yeah.

A

Okay, so let's look at the ramp data and let's look at like VC funded companies and let's say you'd like getting to a billion dollar valuation, but they're doing it with a hundred employees instead of 1,000 or 10,000 employees.

B

Right.

A

So great, they created 100 jobs. You're going to look at the data and say, hey, they went from 20 employees to 100. Like this is great. Like, look at all this growth. Meanwhile, take an enterprise with 50,000 employees that lays off 20% of the workforce over the next next three years. That's 10,000 employees. How many VC funded startups do we need to create to offset that? And then assume that the people that got laid off are the non AI savvy people. They're not going to work for the VC funded startups that are only hiring AI savvy people. I don't.

B

Yeah, yeah, absolutely. Yeah. It's pretty simple when you lay it out.

A

The math just isn't mathing for me that this is all going to work out great over the next three years. Yeah.

B

All right, so next up, Meta had a kind of a messy week on AI. Its own leaders sent some pretty different signals about how the company's enormous AI bet is actually going. So this started with a report in Reuters which obtained audio of an internal town hall this past week where CEO Mark Zuckerberg admitted that Meta's push into AI agents, which is automated systems that carry out tests on their own, quote, hasn't really accelerated in the way that we expected over at least the last four months, according to Reuters, and that the company's restructuring bets haven't yet come to fruition. Zuckerberg also conceded the company's AI reorganization, which we've talked about several times, this included cutting roughly 10% of Meadows workforce, moving about 7,000 employees onto AI teams back in May. He admitted that this was not as clean as it could have been and that executives had misjudged the timing. He said he expects more significant benefits from the AI spending they're doing within the next three to six months, as Meta is projected to spend as much as $145 billion on AI infrastructure this year. Now here's the interesting kind of tension here. Meta's AI chief, Alexander Wang, quickly pushed back on this reporting in public. He posted on X that Zuckerberg quote was clearly talking about the industry's progress on agentic capabilities on the whole, not Meta specifically. And he also in the same town hall, according to Business Insider, told employees that met his next model, codenamed Watermelon, currently in training, has caught up with OpenAI's flagship GPT 5.5 on benchmarks, though he did not say which ones. He added that Watermelon uses an order of magnitude more compute than its predecessor, Muse Spark, which is a model that Meta released in April. So, Paul, a lot of dispute from Wang about where Meta is actually at when it comes to AI. Some bold predictions on how they're catching up. Like, do you believe what he's saying here? How. How are you looking at Meta right now?

A

Seems like some pretty desperate. Yeah. Positioning to try and save face right now. I mean, what is catching up to 5.5 do? Like they're, they're building six internally. We're going to get 5.6 this week. They're already like a model or two ahead. I don't know, it just seems like a, like Meta. Yeah. Grasping at straws here. They've spent 16 billion on the talent. It's not working. They've spent, I don't know, 100, 200 billion on the infrastructure. So now you're going to sell the Compute to try and like make some of that money back and they're going to basically take the XAI route, which is we're not going to make the money and compete on the frontier models. Let's like sell more ads with. I, I don't know. It's it. I'm not. If I had to force rank the labs. Meta is very close to the bottom, if not the bottom right now in terms of relevance to the discussion, which is why we, we don't really spend a ton of time. Plus they haven't done anything in right six to eight months of note. Like it's just not. They're not up there as a noteworthy frontier lab at the moment.

B

So it seems like XAI is essentially admitted they're not competing, but Meta is not yet Based on these comments, it's has not admitted it yet.

A

Not yet. I think it's coming. But you know, they spent a long time doing with the virtual reality, augmented whatever. Like, you know, Zuckerberg makes big bets and he sticks to them for a long time. So who knows, maybe they show up and they do have an incredible model, but right now it's not trending in a great direction.

B

All right, next up we have our AI use case spotlight. So every week we give people a quick look under the hood at some real AI use cases we're exploring. So Paul, I'm going to share a quick one. If you have anything to share, can talk through that too.

A

Sounds good.

B

So for me, I actually wanted to reference this week's use case as a personal project, but I think it's actually illustrative of something that, you know, I think we've talked about a little bit through these topics and I've basically been in my own AI usage trying to become increasingly model agnostic. I've gotten pretty gun shy with everything we've covered recently, worrying about models getting pulled offline, government taking stakes in providers. So I basically want to try to train whatever tool I have available to work really well for me. So this kind of connects a little bit at a personal level to what Karp was talking about. But basically where I've arrived is that your context, your data, the way you work, that is your alpha when you get value out of AI tools. So I wanted to have my personal context be as portable as possible so I can move it between models. So basically what I've been building in my own personal life in one form or another since first experimenting with CLAUDE code late last year, is basically my own personal context layer, like a set of personal accounts data documents that I can point hopefully any reasonably capable AI model at. And the goal is you could hand me a good model. It would work for whatever I need pretty consistently. So just quickly I'll just go through kind of the pieces I've got so far. Again, this is all experimental. I'd love to hear from people if I'm doing this right or wrong, but the first piece is basically given how heavily I use CLAUDE code and codecs is basically an agents file. So it's like essentially your CLAUDE MD or Agents MD file, which is a readme file for AI agents. So I have one that like points to all the right places in this system and tells it, here's where to find all the stuff you need to do everything we do. Second piece is a folder full of skills which are essentially playbooks to do things step by step just in markdown files. I created these using Claude code, but Codec seems to be able to read them fine. I think any agentic system could use them pretty well. So that's cool because then you can just say, here's how to do the things I do. Please go learn how to do them. And then one other piece I want to just mention really quick is like the data layer. So this was kind of a cool one that I did where these are just like connectors that give AI access to information. So I have have quad code and codecs connected to my personal Google Drive, which is critical. I actually use a service called Simple Fin Bridge, which costs about a buck fifty a month and that gives AI read only access to my bank and investment accounts. So basically I can get real time intelligence on my money and it can't touch anything, which is good. But basically, long story short, this is like a bigger lesson of trying to figure out, okay, what about my life can I make legible to AI and then how can I make sure most models out there can use this effectively if one gets taken away? And I think it's probably something businesses might want to be thinking about too.

A

I think it's something our business should be thinking about. That's awesome. Yeah, I need, I need a demo. I need like a. Do a one hour walk me through all that. All right, well, I'll stay on the personal trend here. So this is what I wasn't sure I was going to talk about, but I figured there's a good moral to the story here, so I'll share this one. So. So I'm pretty private about stuff that happens in my life. So. My dad passed away recently. He was our biggest podcast fan and ironically the day Mike and I were talking about the Ben Sass interview and Ben Sasse dying of pancreatic cancer, the day we were recording that episode, I took my dad to the doctor and he got diagnosed with stage four pancreatic cancer. So he was given weeks to months to live at that point and he didn't have an estate in order and I, I had to figure all this out. So he treatment wasn't going to be an option. So he chose hospice care and spent his final three weeks at home. He passed away on May 29th. And it was. It was good. So from just a. I guess stay on the personal side, like, it's okay. Like, he. He went peacefully, so that was good. But to manage that process was very hard for me. And so in the midst of dealing with everything, I turned to Gemini. Like, it was my go to understanding test results, figuring out how to explain to him what was happening, what was going to happen. So, like, I used AI intimately throughout this process, could not have made it through on the medical side and the personal side without having AI there. I also had to race to get his estate plan in order to get a will and testament to do power of attorneys. Like, none of that set up. And we just didn't know how much time we had. We. We thought it was going to be months. We were optimistic that there was going to be time. He. He wasn't in pain. He was living. You know, I'll use the term normal lightly here. But, like, he was at home and got to cut his grass. Like, the first week home, and it looked like it was going to take a little. A little while. And so I got everything in order the second. So two weeks in, I got all of the stuff in order, got everything notarized, following a lot of the procedures that Gemini and I discussed. I would vet things, check stuff. But, like, again, I was. I was racing time here, and I. There was a couple of things I prioritized, like the will and testament, the power of attorneys, and then there was some secondary things that you learn about when you do this kind of thing for the first time, like transfer on death certificate certificates and stuff like that for, you know, personal assets, trying to avoid probate court. So. So I race. I do all this stuff. I, you know, it seems like I got it all in order. And then there was a few key components. We were trying to finalize the end of that week. And then he, you know, he passed during the night that. That Friday, I guess. So basically three weeks from diagnosis until he passed. And I thought I had most of it in order. I had. I had followed the procedures that Gemini had guided me on with some additional verifications, but I didn't have time to get an expert. I didn't have time to get a probate attorney. So I was just trying to do my best to, like, solve this as a smart person with access to a really smart AI assistant. And the lesson I learned and the reason I'm sharing this on the podcast, other than, you know, he was a huge fan Gemini and AI got me 98 of the way there in the end. So the estate process, the estate planning, the. The last 2% ended up being the most important, and it screwed that up. So the process it told me to follow for. For one element of the. The transfer process was completely wrong. And it cost me all of that work. Like, I might as well have done nothing with the one component of the estate. And so that was. It was that lesson to me. And again, why I thought it was worthy of sharing is it's like almost. My dad's final lesson was like, the humans are still needed in a way. And so I told Mike about it that, like, the day this happened, because I was so angry, and I know what to be angry at. Like, I was this after he passes last week. And I was like, I don't know if I'm like, I can't be angry at myself like this. I was just trying to do something in a really short time period under immense personal challenges to solve for something I wasn't an expert in. I did my best. Didn't work. I couldn't be mad at Gemini because it's an AI assistant. Like, I can't yell at Gemini. I did. It didn't make me feel any better. So I actually went into Gemini after this happened. And again, I'm. I'm saying Gemini because I use it. And it was great throughout. I mean, this could have happened with ChatGPT Claude. I don't know. And I went in there and I was like, hey, the process. You told me that I came back to you and said, are you sure this is the process? And you said, yes, was wrong. Like, I actually needed to do this thing before he died. And you didn't tell me that. And this was the wildest part, because you can go in and look at its chain of thought. I went in to see what it was thinking and I clicked on it. And this is what the reasoning part of Gemini said, acknowledging distrust. I validated their distrust and committed to complete transparency, giving them space, because it apologized to me. And it's like, you're right to be upset, especially under the circumstances. I totally understand it. And then it said, correcting misinformation and apology. I directly addressed and apologized for the erroneous transfer on death information, outlining the accurate transfer procedures for both Title and Lost. And I was like, like, son of a. Like, now it's like, trying to placate me. And I can see its chain of thought. And I'm like, oh, now I'm even more angry. But again, like, what am I even angry at? So I don't know, like, again, really personal cancer sucks. But hopefully there's a lesson here that we still need human experts across different disciplines, because I had no idea at the end, if I'd have showed that to a probate attorney, here's the seven steps I'm going to follow. They would have told me in three seconds. You're missing the most fundamental step. And I think that that holds true today across almost every domain that we use these tools. As amazing as they are. If you're not an expert in the domain, something can look great, but it's not. And so it's a cool final lesson, I guess.

B

Yeah, that's a really powerful, wonderful lesson to share. Love that. All right, so our last kind of segment here is our AI product and funding updates. We've got just a few this week, but I'm going to run through these real quick as we wrap up. But first, OpenAI is reportedly preparing to widen the release of GPT 5.6. We talked about this, that we might actually get this model or these three variants of these models called Sol, Terra and Luna after their initial late June rollout that was limited to about 20 vetted partners at the US government's request. So we'll see potentially this week, but expected at least in the coming weeks. Anthropic launched Claude Science, an AI research workbench for scientists that pulls fragmented research tools into a single environment. It has more than 60 pre configured skills and connectors and does things like native visualization of stuff like protein structures. It has automated compute management and built in citation checking. This is now in beta on Mac OS and Linux for Pro, Max, Team and enterprise users. Microsoft announced something called Frontier Co. A new $2.5 billion business unit that embeds roughly 6,000 engineers and industry experts inside customer organizations to help them deploy AI for measurable business outcomes. And they are pitching that customers keep their proprietary data and competitive edge back to what we talked about with Carp, rather than having it used to train AI models. Last but not least, Tesla told employees it will cap individual AI spending at 200 per week starting July 6, just months after pushing staff to use AI more aggressively after some engineers were found to be burning through thousands of dollars of tokens in a week. This is of course part of a broader trend of companies like Uber, Meta and Amazon reigning in their AI costs as token based billing hits their budgets. All right, so Paul, that is wrapped this week. One final announcement here. We talked at the top of the episode about our AI Pulse survey. Go to SmarterX AI forward slash pulse to fill out this week's survey. And this week we're going to ask a little bit about some of the claims that Palantir CEO made and what you think about them and also try to understand better what is AI actually doing to headcount at your company right now. So I'll be super interested to see the answers to that. And Paul, as always, thanks for breaking down another busy week in AI.

A

Yeah, should be some models to talk about next week, so lots more to come. Thanks everyone. Have a great week. Thanks for listening to the Artificial intelligence show. Visit SmarterX 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 AI blueprints, attended virtual and in person events, taken online AI courses and earned professional certificates from our AI Academy and engaged in the SmartRx Slack community. Until next time, stay curious and explore AI.