279 | Anthropic changing the world - again! Automate any knowledge work from your phone 🤯 leaked ‘Mythos’ model that spooks Wall Street, Musk’s 50× chip ‘Terra Fab’, OpenAI kills Sora and more AI news for the week ending on March 27, 2026

SECURE YOUR SPOT FOR THE AGENTIC AI COURSE: https://services.multiplai.ai/agentic-course

Are you already falling behind in the AI race—without even realizing it?

AI isn’t just evolving—it’s accelerating at a pace that’s rewriting how businesses operate, compete, and grow. The real shift isn’t coming someday… it’s already here.

In this episode, you’ll discover how today’s most advanced AI tools are enabling individuals to build full-scale business systems from their phone—and why the real limitation is no longer the technology, but how effectively you use it.

If you want to stay relevant, the solution is clear: learn how to think, build, and operate alongside AI—or risk being outpaced by those who do.

In this session, you’ll discover:

•  How Anthropic is shipping AI updates at an unprecedented pace—and why it matters 
•  What “AI psychosis” is and why even top experts feel they’re falling behind 
•  A real-world example of building a fully automated YouTube growth engine in days 
•  The rise of autonomous AI agents that can execute business tasks end-to-end 
•  Why humans—not AI—are now the biggest bottleneck 
•  The 3 critical skills you need to succeed in the AI-driven future 
•  What major shifts at OpenAI signal about where the market is heading 
•  The growing impact of AI on jobs, hiring, and business structure 
•  Why unconventional thinkers and skilled trades may dominate the future workforce 
•  The emerging risks, opportunities, and ethical questions shaping AI adoption

About Leveraging AI

• The Ultimate AI Course for Business People: https://multiplai.ai/ai-course/
• YouTube Full Episodes: https://www.youtube.com/@Multiplai_AI/
• Connect with Isar Meitis: https://www.linkedin.com/in/isarmeitis/ 
• Join our Live Sessions, AI Hangouts and newsletter: https://services.multiplai.ai/events

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Transcript

GMT20260328-131818_Recording_1970x1390 0:00

Hello, and welcome to a Weakened News episode of the Leveraging AI Podcast, a podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. This is Isar Metis, your host, and this episode is going to be a little different. I am going to start by sharing with you the world I live in right now, the crazy world of extreme AI capabilities, and it's going to connect to a lot of things that Anthropic shared in the past few weeks, including this past week, as well as some other people, specifically Andre Carpathy that I found feels very much like me. We're also going to talk about the two ways, alex Karp, the CEO of Palantir thinks you can stay employed in the AI future, which are unique, interesting. And once you think about it might be actually true. And then we have a very long list of interesting rapid fire items, including some major changes at OpenAI, new regulation guidelines from the White House, a new massive chip production factory that is planned in the US and many more other things. So we have lots to talk about. Let's get started. The chief information officer of one of my clients sent me a very interesting screenshot earlier this week. The screenshot shows the calendar of the last 52 days and what Anthropic shipped in each and every one of those days. For those of you who are watching this, either on YouTube and or on Spotify, you can see the screenshot right now, but it shows a calendar starting in the beginning of February and going all the way through today and showing that in most days through the last almost two months, anthropic has released at least one thing. On each of those days, other than Sundays and Mondays in February, almost every single day, they released a few things. So the pace they have been releasing stuff is an average of 1.4 releases per day. 74 total releases in 52 days, around four different product domains, developer tools, and cloud code with 28 releases, desktop and automation or cowork related stuff with 15 releases, API and infrastructure with 18 releases and models and core platforms with 13 releases. Each of the teams was working independently and releasing stuff as it was ready without waiting for what the other teams were doing. Now obviously the two biggest ones were the releases of the models with Opus 4.6 and Sonnet 4.6, and then their expansion to 1 million tokens. But they also released things such as multi-agent code review and loop for up to seven days of autonomous task iteration and dispatch, which we're gonna talk about a lot in a minute, but it allows you to run anything cloud code or cloud cowork from your phone. And then computer use, which we're also going to talk more about and schedule tasks, which I've been using a lot and it's absolutely fantastic. Cloud code added agent teams for parallel workers. It added debug skills. It added SSH support for remote machines, multi repo sessions, git work, trees, local plugins, security scanning, integrations with Slack, Figma, telegram, discord, CPS desktop app extended to Windows features parody with Mac Os and many, many more. I ran software companies most of my life. The biggest company I ran was nearly or even close to what Anthropic is, but it was over a hundred million dollars. Company. Shipping something new every single day is insane. It is something no other company that I've heard of or that I worked for or worked with or was running myself was even close to, we're talking about a company that has tens of millions of users that uses their product every single day. This is showing you that they have figured out something that no other company on the planet ever figured out. They're able to now do things that are beyond anything we know as far as the ability to ship working effective products on a daily basis. The guy that released this summary of everything they released is called Powell Hurin. And he was talking about the debate of which product to choose or which company to prefer at specific times. And he said the following, this isn't about picking the best AI tool, it's about recognizing when one platform's shipping velocity, makes the decision for you. And I agree 100%. If you've been following this show regularly, you know that I've been all in on Claude for a while, but specifically spending 95% of my time the cloud ecosystem since December and this week has been a complete game changer. Again, and it happened because Anthropic released dispatch. So what is dispatch? Dispatch allows you from your phone, from your cloud application on your phone to control everything in your cloud application on your computer. In other words, all the work that I'm doing in Claude Code and in Claude Cowork, I can start or continue from anywhere as long as I have my phone with me and as long as my computer is open and running at home, which it usually is, and now this sounds maybe interesting, but I wanna give you a real example of what happened this week to tell you how crazy the situation is. I have a YouTube channel. Some of you are following it. Some of you are consuming the content on YouTube, but the YouTube channel has always been an afterthought for me. I've never dove deep into the YouTube universe, but I have the podcast, I'm recording it on video, and I just clicked deploy, and every time I had a new episode, I also released it on YouTube. Now the channel was growing, not a lot, but it gets a decent following with hundreds of people and sometimes thousands of people watching every episode, and I was happy about it because again, it was an aftermath. The core for me was always the podcast, but because it started growing, I started receiving emails from different companies or individuals that provide services to promote your channels on YouTube through the right levels of optimizations. And I got one of those emails this week. I actually saw the email on my phone as I was walking from my car to the gym, and the email had a screenshot of my channel and it said, your content is actually pretty good, but it lacks any kind of optimization for titles and keywords and SEO and tagging and social proof and a lot of other stuff that I actually don't really understand about. And obviously, hence my channel is suffering. This person suggested that he will provide me a free full audit and then I can decide if I wanna work with him. That triggered me to say, this is actually something I need to address. So as I got to the treadmill, I opened dispatch and I was chatting with dispatch through voice type and telling it what I want to do. I actually gave it access to the email itself. I gave it access to my YouTube channel, and I asked it to look at the really incredible LinkedIn content machine that I've built together with Cloud Code, cloud Cowork, and a lot of other infrastructure as a model of what we might be able to build for YouTube. By the time I got off the treadmill, the plan was already being executed. Now, when I got home, I had a million things to do and I didn't get a chance to even look at it. But that evening my son had soccer practice, so when he was practicing soccer, I was back on dispatch, continuing to develop and evolve the plan. The next morning I took my kids to school and at the drop off line to my son. As I was waiting in line, I went back to dispatch, got some additional feedback on what was the current status, gave it additional feedback from me and continued the plan. When I got home that day, I had version one ready to go. Now, version one sounds like something very, very basic, but here's what it does. It scrapes the videos that are shared by the top AI performers on YouTube. It looks at what they publish, when they publish, what titles the news, what titles they use, what descriptions they use, what keywords they optimize for, what thumbnails they're using and so on and so forth. Basically everything on how they do what they do, it then builds a strategy around it. Then it goes back to the content. I'm actually sharing the content that exists in the podcast and and it then creates the new titles, the new descriptions, and so on for my content. It also goes in, automatically cuts the content into shorts that will cover relevant topics based on what's successful on YouTube right now. And it adds the subtitles and everything to them all automatically and it does all of that while connecting to Clickup, which is like monday.com or Notion or Asana or Jira, whatever task management tool you're using. So I can see everything that's happening and I can approve it, and I can have my final say on everything that it's doing, and I'm communicating in Clickup as if it was a team of human individuals. I can chat with it and give it comments and get its thinking and so on. All in the chat inside of clickup, just like I would've done with a regular team. And then it goes and fixes and deploys everything based on my approval or my team's approval. The system also learns over time, so it looks at what it suggested and then how that video performed and it'll continue to improve on its own without me having to give it additional feedback and to make it even more fun, I added the capability for it to go back and actually fix all the videos that exist there right now based on what it is learning. So my older videos are also gonna get additional love on YouTube. This process is something that agencies offer for a lot of money. This process will be better in the way it's running right now because it's not a cookie cutter thing that an agency offers. It actually optimizes all the time based on what's happening in real time about AI on YouTube and will use my content in order to align it with that. It also makes suggestions to me on what other pieces of content I should be recording based on what it's learning from the other creators. I created this entire process while being on the treadmill, watching my son playing soccer and being in line by dropping off my kids. This is where we are right now. Now to make this even crazier, another thing that Anthropic release this week is the ability for this tool to control your computer. Anything on your computer, not just a browser, but open any application you can open and operate it with mouse, keyboard and clicks just like you would now, I did not try it yet. I have some serious security concerns and I don't want to give it access to my main computer, but I am planning to try this out on a home computer that I can wipe out and just test it out to see how it is going to work. But what does that mean? So think about any scenario you can imagine. Let's say you're getting an email from a client that is asking you for an urgent proposal and to invoice them right away for something they want executed in the next couple of days. But you are on the road. You are on a business trip, or you are with your kids somewhere or whatever the case may be. You can open dispatch and literally tell it what to do. I would like you to open my email, whether it's Outlook or whatever thing on the browser and look for the email from so-and-so. I would like you to go and check the recording on Fathom from my latest conversation with them to understand more details. Then I would like you to write a proposal. already have a full proposal writing process that is running through Anthropic anyway, so it is going to write the proposal. Then I can ask it once it has the proposal to go and open my accounting software, find that client, create an invoice, and mail the proposal and the invoice to them, and it will follow all of these instructions. Now, this is obviously really incredible, really bad and really scary all at the same time. Everything I've been feeling about AI just gets amplified more and more and more definitely every single week. And with the pace Anthropic is shipping right now, it is just becoming absolutely insane. Now, while this is all Anthropic base, there's obviously other companies out there like OpenAI and Google who are also shipping, maybe not at the same pace, but on weekly basis, new capabilities that allow us as business people, as organizations, as society, to do more and more things. Now I'm feeling literally every single moment that I can build new capabilities, test new stuff, push the frontier, and deploy production ready solutions for my business anytime. And this is, in other words, there is no limit to what I can build with AI agents right now. Every minute that several of my agents are not progressing at the same time, I feel like I am doing something wrong, like I am missing out because I'm not maximizing my ability to generate more value with the resources that I currently have. So the good news, I can build anything, literally infinite capabilities. The bad news is I always feel stressed to go back and do more stuff. And while I couldn't do this on the road, now I can, which again is good and bad at the same time. But then I listened to an interview with Andre Cari. I learned he feels more or less like me. Now, we talked about Andre many times in the show. He was one of the co-founders in OpenAI. He was director of AI at Tesla. So he knows one or two things about AI and progress, and he said some profound things on what he's doing and how he's feeling about AI right now. So he said, when we're talking about one of the people that is the most advanced in the world, that every time he shares something. Tens of thousands of people follow what he's doing. We shared about some of the stuff he released in the past few weeks, including the loop of auto learning AI that trains additional AI capabilities. Again, one of the most advanced people in AI in the world, maybe set the following. I want to be at the forefront of it, and I'm very antsy that I'm not at the forefront of it. I see lots of people on Twitter doing all kinds of things, and they all sound like really good ideas and I need to be at the forefront, or I feel extremely nervous. He describes his situation as a state of AI psychosis. So the AI psychosis that he's talking about, he refers to is characterized by an intense, obsessive focus on utilizing AI tools driven by the belief that their rapidly expending capabilities make almost everything or anything possible, and that all he needs to do is figure out how to do this. He talks about the fact that he created just for the fun of it, An AI agent called Adobe after the Adobe House Elf built on open cloud that controls all the smart system in his house. He's saying he built it in a few minutes with a single shot prompt that then now allows him with simple voice commands through WhatsApp to control his sound system, lighting, security functions, shades, HVAC, pool, spa, and even his security cameras that detect things outside and provide him text messages to WhatsApp when a new package arrives on UPS. He did this in a single shot prompt. Now he feels very much like me that the limiting factor right now is not the capabilities of the ai, but rather the human in the loop And just like me, he believes that the goal of all of us needs to be. How to safely and effectively minimize our part in the chain of processes that needs to happen to go from an idea to a full production system that can actually do things effectively. He's stating that he completely shifted his approach from having AI write 20% of the code when he's doing 80% of the work manually in December two, as he assumes having not writing a single line of code since December. So the three main points that he is saying, and I agree with all of them 100%, one, is that everything is infinite. You can do anything literally with AI right now. Second is that you are the bottleneck. We, the humans are the bottleneck and we need to remove ourselves from the process as much as possible. And the third is that everything is a skill problem. Meaning the technology is good enough and it's just about figuring out how to use it in an effective way. And if you could not achieve something, the problem is you and not the technology. So where are we in all of this? We are in a situation, and we're gonna talk in a minute about the new model from Anthropic that is going to be even better. And the new model from OpenAI, both of them are rumored but probably imminent, and they're above and beyond the crazy good models that we have right now. And the only thing that stands between you and your company, 10 x-ing everything that you do, a hundred x-ing everything that you do is knowing how to do it effectively and pushing the limits and the boundaries of what's possible with these tools right now. Now, as you probably know, if you've been following this podcast in the past three and a half years, I've been on a quest to get as many people, companies, and organizations to learn how to use AI effectively. This has literally been my main focus on the day to day on everything I do, both from a tactical perspective and from a strategic perspective. I think this is critical for the future of our planet, of our society, of my kids, of your kids, to know how to use this technology effectively. Now, if you've been following the show in the past few weeks, I've shared that I'm launching a new course. The course teaches how to build these incredibly magical, multi-agent orchestration solutions that I have been building every single day in the past three months. Now, I shared in the past few weeks that if you're interested, you should drop me a message on LinkedIn and or email so I know how many people are interested in this course. My plan was to get 10 to 15 people to say that they're interested and launch an early bird session that will be deeply discounted, that will be in a small group, that will allow me and the people who are participating learn together and make progress. And it will allow me to start on a smaller group, and it will allow this group to really get a bigger portion of my attention. And then everybody wins. The problem is I got significantly higher demand than I expected. I already have 25 to 30 people who said that they're interested, regardless of when it's going to be, what's gonna be the price they want to join the course. Now, since I do not want 30 people in the first course, I decided to make the commitment to this first early bird batch. A little more formal in the show notes of this episode, there is a link for you to sign up and save your seat for the early bird session. The early bird session will have a, at least $500 savings. those of you gonna go to the landing page will understand what's the other potential savings. But if you also join the membership, it will save you even more money. But to focus on the course, it will save you $500 on the course itself. To secure your spot. There is a $100 down payment commitment towards the course. So I know how many people are really serious. Now I have my Friday AI Hangouts community meetup that happens every single Friday at 1:00 PM Eastern on Zoom. And I have been doing this for more than two years, and there's currently around 30 people on these calls. It is growing more or less every single week. But because these people have been in my community for a while, I shared that with them first, and I already have six people signed up for the early bird session, meaning I have nine seats left. Now, I'm not saying this to drive urgency. I'm just telling you, this is going to be the most transformative course you ever took until this point in your life. Now, AI may change a lot, and this, there might be another crazy thing that happens in the next six months or 12 months or whatever. But to this point, this new capability is literally life changing, business changing, society changing. And if you wanna be at the forefront and know everything that I know, what I'm building, how I'm building it, what are the processes, what are the mindsets, what are the technologies? What is the underlying architecture, infrastructure, all of that I'm going to teach in this course. If you take this course, the only thing that's gonna limit what you can build is going to be how good your ideas are, how good your judgment is to pick the right ideas and evolve them correctly, and your ability to learn how to maximize your tokens every single day. It'll enable you to basically enhance and or automate every aspect of your business that is done in front of a computer, and obviously the same thing for your personal life. So if you're interested, go to the show notes right now and reserve your spot to the early bird session. Now, by the way, if you are a business leader and you have a company and you want to train this for your team, I'm also doing this training for companies either in person as a two day workshop or on Zoom. I already have three organizations who booked me to teach them this transformative knowledge and technology, and I'm going to deliver it to these three companies in the next two months. So if you do wanna teach this to your team, your organization, please reach out to me either on LinkedIn or through my email, Now, let's continue with some additional crazy news from Anthropic. So. Based on some leaked documents that Fortune got their hands on. Anthropic has a new model that has the code name Mythos or Kabar, and the new model is described as a step change in performance is in currently in early access testing with customers. Now, apparently this is a step above its current Opus models, which again, are absolutely incredible and enable me to do more or less anything I can imagine. So it's very hard for me to understand what the next level up can be. But based on these documents, the new model is demonstrating dramatically higher scores on software coding, academic reasoning, also in introducing a completely different level of cybersecurity risks. Now, the details became available along with almost 3000 other unpublished documents due to a human error in Anthropic content management system, and the company confirmed that that's actually happened. Anthropic acknowledged that this model is, and I'm quoting currently far ahead of any other AI model in cyber capabilities. Opposing unprecedented cybersecurity risks, which is currently the main reason they're slowing the release and they're increasing the testing before they are making this model available. Now, the news about this send the cybersecurity stocks like Palo Alto Networks and Zscaler and CrowdStrike and Okta to plunge six to 10% on Friday, which tells you that in addition to the capabilities themselves, this can send additional shock waves into the stock market and the economy as a whole. And based on these documents, the new model is noted to be very compute intensive. And that's a quote, which means it's going to be very expensive to operate, but if it really provides these incredible capabilities, it will probably provide positive ROI. Staying on anthropic, a update on their evolving situation with the Pentagon. So a federal judge has temporarily blocked the Pentagon's designation of Anthropic as a supply chain risk, making an early legal victory for the firm, and that's according to Axxis. US District Judge Rita Linn criticized the Trump administration move as potentially unlawful and retaliatory stating It appears to be designed to punish Anthropic for disagreeing with government position on AI use. If you want the exact quote from the judge, it was nothing in the government statute regarding the supply chain risk designation supports the orian notion that an American company may be branded a potential adversary and saboteur of the US for expressing disagreement with the government. Anthropic were obviously very happy. Their spokesperson said, and I'm quoting, we're grateful for the court for moving swiftly, and please they agree with anthropic is likely to succeed on the merits. While this case was necessary to protect Enro, our customers and our partners, our focus remains on working productively with the government to ensure all Americans benefit from safe, reliable ai. Those of you who don't know what this is about, we covered this thoroughly in the past few episodes, but Anthropic has the only system right now that is currently running on highly secured government and military systems, and the government requested anthropic to remove the limitations that they put on the system, which they put in their contract that does not allow it to spy against American or to be used for complete autonomous weapons. Anthropic did not cave in, and as a result, the government has labeled them a supply chain risk, which is something that has never been assigned for any US company and only for adversarial companies such as Chinese manufacturers and so on. The statement from Secretary Hegseth also said that any company that is doing business with the government should stop doing business with anthropic immediately, or face the risk of being removed from government supply chain as well. This was also stated by the judge as unlawful, stating that that's not how this designation actually works. The bottom line is Anthropic already suffered some cancellations from different companies. Based on this designation, they're now able to, for now, stop the designation knowing this particular administration, this is not the final word of this conversation, but we will see how that evolves. but this is the current latest update on the situation. Now, I told you I wanna mention one more thing that relates to how the world is evolving with ai, and I'm worried about this, and I shared my worries with you multiple times. The ability for anyone to create any automation and any process right now puts many, many jobs at risk. And it means that the future and potentially the mid future. So not the immediate future because there's friction in the way things gets deployed. The stuff that I can do as small businesses can do is not doable by very large organizations. I have many clients who are in hundreds of millions or billions of dollars in revenue, and they still are confined to using Microsoft copilot because of data issues and security issues and so on. So the world is gonna take a while to completely benefit from the capabilities that are available right now, but it will go beyond that, which means we have a few years to figure it out. Alex Karp, who is a very interesting person, but definitely a very smart and successful entrepreneur. He's the CEO of Palantir, one of the companies that has grew like crazy in the past few years by supplying AI based and other solutions, mostly in the defense world to the US government and other governments. He believes that in the AI future, there are two. Ways to be successful in a workplace. It's either possessing vocational training, in other words, carpentry, plumbing, electricians, et cetera, or being neurodivergence, which basically means having conditions such as autism, A DHD, dyslexia and so on. And he's basically stating, and I'm quoting everybody with normal shaped scales are dyslexics, meaning the things they can do that used to be valuable is not so valuable. The things that they need to learn to do is be more of an artist, look at things from a different direction, be able to build something unique. So what he's basically saying that, if you're not a plumber or an electrician or something like that, your only way to survive in the AI future is to be able to approach things with a unique, non-standard approach. Not to put the money where his mouth is, and it actually the other way around because they put the money there before Palantir launched a neuro dive, a neurodivergent fellowship program offering salaries between 110,000 to 200,000 annually to attract individuals with unique strengths and unconventional thinking and creativity. Now, carp is not the only person who thinks this way. BlackRock, CEO, Larry Fink has also highlighted a growing demand for skilled trades due to AI infrastructure development. Also a Gartner study projected that by 20 27, 1 in five sales organization within the Fortune 500 companies will actively recruit neurodivergent talent to enhance performance. And I wanna connect this back to what I told you I believe are the three most important things to be successful in this era. Great ideas, which again, people who think differently have different ideas, which will diverge from the common, which means they can do stuff that other people cannot. Solid judgment, meaning prioritizing the right ideas and then knowing how to move them in the right direction correctly. And then the third is being able to define your ideas into detailed requirements that the AI can execute in an effective way. So my thoughts are that I agree that neurodivergent people, people with unique thinking, unique approach to life that has a different way of thinking will be extremely valuable. But they will need to work with other people who can have solid business judgment, and that can have the solid way to turn ideas into detailed requirements, or at least learn how to use AI in order to do that, which is one of the things I'm teaching in my course. But either way, we're looking into a very different feature than the current present. We will need to reimagine more or less everything in the workforce, in our society in probably the next decade or so. And now two rapid fire items. We're going to start with OpenAI. OpenAI made some major changes in the past two weeks. We shared with you last week. That OpenAI announced that they're going to focus on business valuable use cases and focus on enterprise clients, and that they're going to minimize side quests. While this was a statement last week, it became very practical this week. So this week OpenAI has canceled soa. So SOA the app and SOA the capability, meaning creating videos with OpenAI tools will cease to exist. This is apparently due to the fact that the video generation is a huge consumer of compute that is not generating nearly close the revenue of the compute that it is consuming. As a result of that, Disney has canceled their agreement to invest a billion dollars in OpenAI. That agreement was including in licensing the main Disney characters into the Sora and OpenAI universe. No money has changed hands so far, and that deal has been confirmed to be canceled by both OpenAI and Disney. Now OpenAI is also rolling out a new shopping experience within ChatGPT that is shifting the focus from the direct in chat purchases via what they called instant checkout to a more robust product discovery with a comparison engine that will allow people to find and compare the products that they want. But the checkout will happen either on third party apps inside of ChatGPT, or on the third party platforms. So this is another initiative that OpenAI focused on a lot last year that is getting abandoned, and that's the ability to purchase things straight on the ChatGPT environment. Now the new shopping experience enables users to find and compare products by uploading either images or describing what they're looking for and their budget and preferences and so on. And ChatGPT will offer visual and contextual information about potential opportunities that they might want to purchase. OpenAI is focusing is improving speed, relevance, and product coverage versus the actual checkout process. Merchant on the other side can now share product feeds and promotions to ensure that their products are fully represented within the ChatGPT environment. Several big companies, including Sephora, Nordstrom, and Shopify and Walmart already support this new architecture from OpenAI. We're gonna talk more about Shopify later on, but they are making a lot of moves to be ready for the Agentic era. So we already have two immediate casualties of the removing side quest. The big one is Sora, the smaller one is the immediate instant checkout inside of ChatGPT. I'm sure we're going to see more of those in the next few weeks. OpenAI also announced some changes in the leadership and the focus of it. Sam Altman is going to focus a lot more on raising funds and preparing for the IPO and their compute infrastructure while Fiji cmo, the CEO of application is more or less going to run the day-to-day operation and the development of the OpenAI solutions, and since she clearly stated that they're going to move away from side Quest and that the company is in, I'm quoting, orienting aggressively towards high productivity use cases tells you exactly the direction that open AI is going. They don't really have much of a choice. Connecting you back to how we started this episode, the. Pace in which Anthropic has been shipping really useful things and has been growing its consumer base and has been growing its revenue. They are catching up to open AI with the projections right now saying they might surpass open AI's revenue at the current growth rate before the end of this year, while being a significantly smaller company who raised significantly less money and has significantly less overhead. So OpenAI has to pivot. In order to stay competitive with Anthropic, they have to ditch their consumer facing, or at least reduce their consuming, facing dependency. Right now it is 60 40 towards consumers on open AI with a goal to make it 50 50 in the next few months. It is completely the opposite with Anthropic that has most of the revenue coming from B2B versus B2C. That is also something that makes a very, very big difference in the IPO because it's much more predictable long-term revenue that investors like, so we will see OpenAI making that shift and we'll see less investment and less releases that are consumer related and more and more focus on the enterprise aspect of ai. Now, ChatGPT also made some product announcements. First of all, they have changed how the model picker looks like. While previously you were able to pick the model, that's no longer the case. You can now only pick instant or thinking these are the two options that you have. And behind the scenes, ChatGPT automatically routes request two different underlying models based on the prompt complexity and what it is trying to achieve while minimizing the amount of tokens that it is going to use in order to achieve the best results per token, that is possible. Now, the reality is it didn't totally go away. You can still have access to it. You just need to know that it's possible. So under the configure menu, inside your settings, which most people will never actually access, you can switch back and enable it to show you specific models. So if you do wanna go back, instead of selecting instant or thinking, you can click on Configure and then you can get access to one of the previous models all the way back to oh three, which is the oldest model that is still available through that menu. Now the fact that the AI decides on its own, which models and which level of thinking to use is helpful on the big sense that it will try to maximize the value per token. But the disadvantages is that you can have two users in your company submitting the same exact prompt at the same time and may receive answers from completely different underlying models based on whatever subtle differences or the exact time where they submitted, which means you may not get consistent results, which is obviously not a good thing. Now, I might be over exaggerating the situation, but I can tell you that my biggest problem with the current. Way that OpenAI models work is that they have broke some of my custom gpt. So I had several custom GPT that were running and working perfectly fine for months, more or less since the introduction of custom GPT and now all of them run significantly longer because it's now using reasoning model instead of old school step-by-step, just do what I'm telling you kind of model and in it fails at a much higher frequency than it did before, which is another reason why I switched the vast majority of what I do to the CLO ecosystem. On the good side. OpenAI introduced an interesting feature this week, which is your file library. So on the left side menu close to the top, there's now a library option that shows you all the files that you uploaded into ChatGPT. It is available to all paying clients, so basically plus pro and business users. And it may be rolled out to free users Later on, users can now quickly reference files in conversations and ask chat GPT questions about the uploaded content, even if it was uploaded in other conversations. This dramatically obviously enhances your ability to use existing data, creating more or less a builtin rag solution inside a regular chat two PT environment. So any file you uploaded in the past, you can go and see it right now in the chat two PT environment, and you can reference it in any conversation in the future without having to re-upload. The file limits is 512 megabytes for individual files, 2 million tokens for text and 50 megabytes for CSV and 20 megabytes for images. Now from a data privacy specifics, OpenAI can use the file to train the model if the improve the model checkbox in your settings is set to on, if it is set to on there, have been training on a lot of other stuff without you knowing it and it's something you can very easily turn off from your settings. Mine has been turned off from the moment that switch has existed, but you can do whatever you think is right for you. OpenAI is claiming that the file storage retention will say indefinitely unless you manually delete it. And the permanent deletion happens 30 days after your initial deletion. They're also stating that files from temporary chats and ChatGPT Health are excluded from this library. OpenAI also announced that they're acquiring Astro, which is creator of a widely used open source python development tools to accelerate codex, and the move is very, very clear to go beyond just code generation to full lifecycle of AI software development agents. This has been the push from everybody. There is a similar approach from all the different platforms, including the new rumored model from Anthropic. Astros open source tools that have been widely used around the world by Python developers is going to be integrated into Codex, but it is also going to stay open source and open AI is going to continue supporting the open source community and allowing these tools to continue grow independently from what they're going to do in Codex as well. While OpenAI is going to focus very aggressively on the enterprise, they are also going to double their workforce, or almost double from 4,500 employees right now to about 8,000 employees by the end of 2026, driven by an aggressive expansion into product development, engineering, research, and sales. In addition to that, they are on a hiring spree to what they're calling technical ambassadorship, which is people who are going to help OpenAI customers implement a OpenAI tools and technologies into processes inside companies. To me, this is showing another big difference between open AI and Anthropic. Anthropic is focusing on developing AI tools that will allow them to ship faster and be more competitive. Leveraging more ai while open AI that is trying to push other companies to be lean and use AI in order to grow is going to double their workforce. So that's a little counterintuitive for me, but when you raised $110 billion and actually this past week they announced they raised another $10 billion for a total of 120 billion just in the past few weeks, you should while you're racing to an IPO competing with two other gigantic IPOs, including anthropic and SpaceX, you should do whatever you need to do in order to come out ahead and with the right tools and the right focus and the right infrastructure in place. And what's another 4,000 employees when you just raised that kind of money? Now, speaking of anthropic, and now to a few rapid fire items about them. They are apparently planning an IPO for quarter four of this year that might be ahead of OpenAI and it will most certainly be after the SpaceX and X-I-I-P-O, which presumably the papers for that might be submitted in the next few days, potentially this coming week with an IPO date around June to July, so this summer, and it is planned to be the largest IPO in history. We're going to talk about a very interesting announcement from Elon Musk later on in this episode. But the anthropic IPO is moving forward probably before the open AI IPO, and most likely in Q4 of this year. And they couldn't do this in a better timing. Their revenue is surging. They are growing like crazy. Anthropic also claimed the second most used generative AI application daily overtaking deep seek grok and Gemini. And that's according to data from SimilarWeb. Claude reportedly achieved over 1 million daily signup recorded and 11 million daily users globally. In. Making it 180% growth in daily usage since the beginning of this year. So in less than one quarter, they more than doubled their daily usage. In late February, 2026, Claude downloads in the US actually surpassed the ones of ChatGPT. They have seen 140,000 daily downloads, more or less consistently exceeding those on ChatGPT, which saw around 124,000 daily downloads. So 25,000 downloads more than ChatGPT every single day. Everything you're looking at right now, it's showing anthropic winning. We shared with you last week that on the new AI spend in enterprise and companies in the us, 73% goes to Anthropic. To make this even more extreme, Claude has experienced an incredible 1487% surge in user sessions in March. To put things in perspective, that means an average user engages with Anthropic 38 sessions per week when OpenAI sees 18 sessions per week. I told you about myself. I spend every single minute that I have doing things in parallel on Anthropic, more or less any minute of the day. I'm not saying it's a good thing, it's marginal addiction at this point, but I'm spending hours and hours and hours running parallel sessions inside the Anthropic universe, and apparently I'm not the only one. And to tell you where this might be leading is actually a very interesting statement from Jensen Huang, the CEO of Nvidia, the most valuable company in the world right now. And in the All In podcast, he said the following, and I'm quoting, I believe Dario and Anthropic are going to be way better than that. When he's saying way better than that is. Dario was projecting that by 2030 they might make a trillion dollar in revenue. So Jensen think that they will be doing much better than that. And the reason for that is that the one part that he hasn't considered, I believe every single enterprise software company will also be a value added reseller of Anthropic tokens. And they're going to get this logarithmic expansion. Their go-to market is going to expand tremendously this year. So what he's basically saying is, because so many tools are using anthropics technology, when I build applications, when I need a high-end model, I currently use Anthropics models. When I need lower end models, I'm actually using either Chinese models Grok 4.1 fast, which I find to be very effective from a value to money perspective. But what Genten is saying that every new application that is going to be built, that is going to be using the API of Anthropic in the backend, is going to be expanding the usage of tokens coming from Anthropic, which is going to dramatically increase their revenue from basically any new software that's gonna get built around the world. Now as part of all the announcements that they made, another cool thing that they released this week is cowork projects. So it's the same exact concept of projects they had on the web now exist on the desktop application as well. And those of you who don't know what projects are, they're basically mini universes of context and memory, meaning you can upload files to them and you can give it instructions, and then whatever happens in that conversation can benefit from the files in that memory, from the instructions, and it's learning from the memory of conversations happening in that particular project. The big disadvantage from my perspective is that these projects, while you can import the projects from your. Web environment, they do not sync, meaning it is not actually the same project. There are separate projects, one in your desktop and one on your web. Hopefully that will get changed in the future because I think that will be significantly more valuable. But the cool thing about it, it means that you can build a bubble of context for specific projects for specific clients, and then have your scales and plugins and the entire ecosystem that you're building on the Anthropic desktop application have access to those specific context information that is stored in the different projects. Now, we mentioned Elon Musk several times already. He made a huge announcement this week that he's calling the Terra Fab, which is going to be the largest chip manufacturing facility in the world that is going to supply chips for Tesla, SpaceX, and X ai. Or as Elon said it, and I'm quoting, we either build a terra fab or we don't have the chips. And we need the chips. So we build the Terra Fab. So what is the Terra Fab? It is going to be a fully vertically integrated manufacturing facility for the most advanced AI chips in the world. It's going to be built next to Tesla's Austin headquarters. With the goal of producing over one terawatt or a thousand gigawatts of AI computing power annually. To put things in perspective, this is more than the entire world chip production is right now by a big spread. So Musk is estimating that the entire world production of AI chips right now is 20 gigawatts per year, and he's planning to do a thousand that's 50 x what the entire world is generating in chips right now. Now this is planning to be the most advanced type of chips in the world with two nanometer process technology, and as I mentioned, a fully vertically integrated process that does not exist anywhere in the world. But the most interesting thing is that the facility will produce two types of chips. One is edge inference processors like AI five and AI six, which is used by Tesla's full cell, drives the cyber taxis and Optimus humanoid robots. But the other one is a higher power D three chip that is specifically designed for harsh space environments and orbital AI data centers. Now, to make this even more extreme, 80% of the production is supposed to go to space compute versus earth compute. And the reason is, as I shared with you in the past, Elon is a huge believer of data centers in space. Now, while I was very skeptic about that before, now that they're building this facility, it'll be very, very interesting to see how that actually works. But this connects to Elon's much bigger vision of allowing humanity to live outside of Earth and on other planets, and he sees this as a key component in that vision. Just like everything he's building in SpaceX now from a timeline perspective, he's planning to start production in 2026 and to grow it dramatically in 2027. That has been met with a lot of skepticism in industry for two reasons. One. Building a cutting edge semiconductor fabrication facility from scratch is not a simple task, and people say it will take a few years to build. That being said, it usually takes nine to 15 months to build a really big data center. And Xai with Elon's leadership was able to build it in three. So it has been happened before, but Elon has always been known for dramatically exaggerating the timelines of things that he's doing. That was very obvious with Tesla and SpaceX and other aspects. But the direction is clear. They're going to be building a facility that will be a major player in the manufacturing of chips, but they're going to use it for their own internal usage between SpaceX X, AI, and Tesla. And from news about companies to regulation, the White House has released its long and anticipated national artificial intelligence legislative framework on March 20th. The goal is to establish a unified federal approach to AI regulation and preventing states from creating their own laws, creating a patchwork of legislation that will make it hard to drive innovation through the US economy. This framework is supposed to enforce Trump's administration light, touch, regulatory philosophy on ai. The goal as they're stating it, is to balance rapid innovation with public trust across broad spectrum of AI concerns. So the framework outlines six objectives for the framework, including protecting children and empowering parents with digital management tools, safeguarding American communities and small businesses, respecting intellectual property rights, preventing government censorship, enabling innovation, and developing AI ready workforce. The administration is explicitly calling for Congress to preempt state laws regulation on AI development, arguing that. A patchwork of conflicting state regulation may undermine America's ability to innovate and stay competitive in the global AI race with China. Now the White House also advocate for sector specific regulation approach through existing bodies like you know, the legal aspects of things and the technical aspect of things and so on. Rather than creating a new single new federal AI rulemaking agency. I must admit this makes sense to me to an extent. I think the problem is the current federal channels have. Do not have the technical expertise or the understanding of where AI is and definitely know where AI is going in order to be able to address these things in an effective way. So while in general, a idea, I agree with the general concepts that its administration is putting in place, I think the reality is it is just going to leave. It may leave the door open for a wild, wild west of the AI companies, which is not necessarily a good thing. Now we are in an election year. The midterm elections are about six months away, and we are going to see a lot of more and more political aspects of what AI impact might be. There's still not clear which of the sides is gonna pick what aspect of ai. I think there's gonna be a lot of attempts from research companies that are supporting both parties to understand what will resonate with the US population. Right now, it is not very clear, but I assume it'll play a role in these elections already staying on the topic of how AI actually impacts society, meta just laid off a few hundreds of employees on Wednesday affecting mostly people from reality labs, recruiting sales, Facebook and global operations. This follows an earlier round in January that eliminated 1500 jobs from the Reality Labs division alone. Now this is all done in order to reduce expenses in order to support the crazy investment that they are planning in ai. The company is planning to spend 115 to $135 billion in 2026 alone, more or less doubling their 2025 investment in AI infrastructure and talent. And it is very obvious that the focus has shifted 'cause Meta's Reality Labs Division, which is responsible for the Metaverse, which was the previous big focus of Meta, which if you remember back then was called Facebook and changed the name to Meta because of their focus on the Metaverse has accumulated over $73 billion in losses since 2020. With an operating loss of $19 billion in 2025 alone. So the shift is very obvious to reduce expenses on that aspect of the business and push as much as possible into the AI space. And meta is not alone. Where there has been over 45,000 tech jobs eliminated globally in just the first quarter of 2026. And AI has been cited as one of the factors in more or less, each and every one of them, regardless of sector and place in the world. That being said, there's a lot of discussions on whether AI is really the reason or it's just a good thing to say because if you just say you're laying people off because you're not being efficient and you over hire during the pandemic, then you're gonna get penalized by the stock market or investors. And if you're saying that you are building efficiencies with ai, then you're gonna get rewarded for doing the same exact action of letting people go. So it is not completely clear whether AI is the driver. From a very personal perspective, I don't see this evolving any other way. And I said that many times in the past. The things that I'm doing right now for my business and for my clients allows you to automate more or less any knowledge task that is done in front of a computer. With the current technology right now, that's before any new models get released, more infrastructure gets released, more tooling and scaffolding gets released with the current technology. And that means that many jobs are going to get affected. Now, can new jobs emerge maybe potentially. I don't think at the same scale as the jobs that are gonna get lost, definitely not in the short two medium term. Switching to another topic, Francois Cholet, who is the guy behind ARC A GI just released ARC AGI I three. So if you don't know what ARC agi I is Ark a GI was built as a benchmark to test the actual intelligence of ai. Instead of testing AI on benchmarks on existing things that AI was able to learn in the past, it is actually letting it deal with situations that are new. It's like little games and puzzles that it needs to solve. And it evolved from RKGL one to two to three trying to introduce new capabilities to the AI that is not aligned with anything it was trained on to evaluate the real quote, unquote, intelligence behind artificial intelligence. So humans were able to solve a hundred percent of the arc, a GI three environment upon first exposure without prior training or instructions. Current Frontier AI reasoning models achieve less than 1% efficiency on this new test. Now this highlights a crazy big gap that is obvious in AI, where it's extremely good at some things and has the intelligence of a fly on others. So the benchmark evaluates AI system based on their action efficiency, meaning how few yet effective steps an agent takes to achieve the goal in an unfamiliar environment trying to solve these game-like puzzles. The scoring known as relative human action efficiency, or RHAE compares the AI's actions to the human baseline that was tested on the same tests before. And for each level, the score is being squared to emphasize the. I really like the quote from Francois about what intelligence is, and that's really what he's trying to measure. And he said, intelligence is the efficiency with which you're going to make sense of new things, of new tasks that you've never seen before. So what he's basically claiming is that while artificial intelligence is right now incredible in doing things, it was trained on, and it is incredible, it is really bad at doing completely new things that do not align with things you was trained on before, where the human brain can actually solve these very easily in an intuitive way. And that gap still exists. It will be very interesting to see what are the next innovation in technology that are required to allow artificial intelligence to do similar things. And we'll finish with some new valuation numbers from some leading companies across different aspects. Replit. The coding startup companies that I really like has raised $400 million in a series defunding at a $9 billion valuation tripling its valuation from just six months ago. It brings their total funding to $878 million and positions them as the most valuable US startup with a female co-founder, which I really, really like. CEO Amjad Masad stated that PL is now being used by 85 of the Fortune 500 companies, positioning them as a critical tool for large businesses, at least in the US. Now, two great quotes related to this. One is, our mission has always been that every human with an idea and an internet connection should be able to build any app they want. And the other quote is, I usually say we didn't start from zero. We started below zero, like negative as immigrants. This quote comes from Haya ODed, who is the co-founder and VP of Design for Repli, and she's the wife of the CEO. And the reason I really like these two quotes is they basically tell you the story. The American Dream and beyond is now available to any person on the planet with a computer and internet connection. Anyone, anywhere that has access to the level of intelligence that these tools provide can now build anything they can think about that can provide value to others, and hence can sell and make money. This is a great note to end on. I will mention back again that if you're interested in our early bird course for building multi-agent orchestrations using the Claude ecosystem, you should use the link in the show notes right now to sign up because if you wait, it might be already sold out. Keep on experimenting with ai. Keep on sharing what you learn with others. Give us a review on your favorite podcasting platform. I would really appreciate it. Share this podcast with others if you think it is valuable. And we will be back again on Tuesday with an amazing episode teaching you exactly how to implement AI in your business and in your day to day. And until then, have an amazing rest of your weekend.