191 | The craziest week in AI news history 🤯 Microsoft, Google, Anthropic and OpenAI major announcements, and more important AI news for the week ending on May 23rd, 2025

26:13So not just these devices, but presumably the software and a lot of other stuff around it, and the user experience is going to be managed by Johnny, which personally I find exciting. But now let's talk about what are the implications of this. So nobody knows what they're building. There are a lot of rumors and a lot of conversation on X and online and across multiple platforms. But the reality is nobody knows. But what we do know is that it's going to eliminate the need to pull out a hardware device like we're doing today. So in the conversation, which I'll put the link in the show notes and you can go and watch, it's a, I don't know, 15, 20 minute video that shows you the process and their conversation talking in a coffee shop in San Francisco.

26:51Very well produced, by the way, a very laid back, chill, down to earth way to explain what they're after and also the backend story. So in this conversation, Sam Altman says, well, if I now want to use ChatGPT, I need to stop what I'm doing with Johnny right now, lean down from my chair, grab my backpack, open my computer, start it up, go to ChatGPT and then start engaging with it, which doesn't make sense when you have tools like ChatGPT and similar AI capabilities. So the assumption is they're building some kind of a wearable interface that will allow you with voice and maybe video and probably both, especially if you're talking about a family of products, to engage with them in a most intuitive, human way, like talking to it and showing it different things.

27:32Is that going to be glasses? Is that going to be a bracelet? Is that going to be an earpiece with AI capabilities built into it? I don't know, but it's probably some combination of these things. Now, as exciting as it is, it raises huge amounts of very large questions in my head. The first one, which I talked about many times in this podcast, is privacy. What we are walking into, and we're going to talk about new glasses in a second, what we're going into is a universe where everybody will record everything all the time, and it's going to be processed by AI all the time. Now, I'm sure there's going to be some rules and regulations where you can and cannot wear these devices, but I think over time, this will become the norm and everybody will just be used to that being the case.

28:15But take it into school and education. Do we allow our kids to use these devices at school or at graduate school? Is that okay? Is that not okay? It's also going to dramatically increase the digital divide. So if today you have people who don't have access to computers or the internet, well, now you'll have people who have AI in their ear, in their eyes every single second, and the people who cannot afford it. This is a completely different level of force multiplier that we didn't have before. It's the first step, if you want, into a cyborg era where humans are collaborating with machine in a very seamless ways in everything that we are going to do.

28:52Now, you want to take it into a more foo-foo, crazy direction? Think about Elon Musk's brain chips. That's a technology that is still not there for what we're talking about, but that's Elon's goal, and that's where the technology is going, meaning sometime in the future this may be five years this may be 10 this may be 20 sometime in that time frame you'll be able to connect to ai tools straight into your brain you can think about something and get ai results instead of having to show it things and talk to it now the question is as crazy as it sounds will you allow your kids to go through that operation and install that in their heads i will let that sink for a minute the ability to know everything that ai knows in your head without anything other than thinking about it.

29:35That's the ultimate cyborg and knowledge straight in your head. Now, I know what you're thinking. You're saying, you're crazy. There's no freaking way I will ever do something like this to my kids. But what if three quarters or nine out of 10 kids in their class or their higher education has that? What happens then? What happens then is that the social pressure will be such that you will be forced to do it. and you'll basically have to choose, do you want to stay a part of the advancing humanity of the 21st century, or you want to become the Amish of the 21st century? I have a feeling that there's going to be a huge amount of quote-unquote Amish people, at least in the beginning, that are saying we're not doing this, this is beyond what we think is human, but I think over time, again, this will become the norm, and because you will want to stay competitive in the world as it is, you will upgrade and you will do these things and everybody or a lot of people will have the functionality, which will increase the divide even further.

30:36Sorry if I went a little foo-foo on you, but this is where my head is going when I'm seeing these things. And I know it sounds crazy, but I think from a technological perspective, this is where it's all converging to. Now, to bring it back down to earth,$6.5 billion is a huge amount of money for a product that doesn't exist. So why invest such a huge amount of money? Well, the annual smartphone revenue in 2024 selling smartphones to people was over$800 billion. That is supposed to grow to almost$850 billion by the end of the decade, unless something else comes in and shifts our spending from cell phones to a different way to engage with the world.

31:20So we're talking about a number that is getting close to a trillion dollars in market. And if you're in there and you control the hardware as well, meaning if OpenAI stops being dependent on delivering their tools and their capabilities through Apple or Google or Microsoft, they are now a major player in that universe. And again, this is just the smartphone world. What if some of the work we're doing on computers right now can be replaced by the devices that they're building. So that gives you an idea why a$6.5 billion investment that sounds absolutely insane is a very solid investment if you can build the next iPhone first before everybody else, better than everybody else, because you have the guy who built the original iPhone and the team of the most capable people in the world in creating devices, hardware, and designs that are perfect for their need.

32:17So since we started talking about devices, there are two pieces of news that came out this week. One, in the Google I.O. announcement, Google introduced a new version of Google Glass. Remember Google Glass from X number of years ago? The only geeks in Silicon Valley were and that were not very productive and died very quickly. Well, Google now introduced new glasses that are running Google Android XR, which is something they developed in partnership with Samsung that features cameras, microphone speakers and optical lenses to display text and other information. And very different than the geeky, weird looking glasses of last time, they're doing it very similar to what Meta is doing and they're partnering with glasses manufacturers and designers, in this case, Warby Parker and Gentle Monster.

33:02So they will design the glasses. Google will provide everything else. Now, the way this is going to work is it's going to stream back and forth to your phone, allowing the glasses to be significantly smaller and lighter because it needs less component in the actual glasses because most of the compute will be done on the phone that we're carrying in our pocket as well. I'm sure that step one and that step two will actually change that into an independent device, but this is going to be the first real competition to Meta's highly successful partnership with Ray-Ban. They've sold over 1 million units in 2024 and they doubled that amount in the first half of 2025.

33:35So they are the only company in the market right now and that's going to be a burning market moving forward. The other company that made a similar announcement is Apple. Apple just announced that they will release an AI-powered smart glasses by the end of 2026, with a large-scale prototype production starting at the end of 2025. Very similar idea, cameras, microphones, and speakers enabling photo and video capture, real-time translation, turn-by-turn direction, music playback, etc., etc., etc., and obviously working in collaboration with a new version of Siri that is yet to be introduced. But you see where this is going.

34:11As I mentioned earlier, everybody will wear devices that are connected to the internet that has AI powering them that will allow us to engage with the world around us in a very different way than we're doing right now. And if you look at the news from Project Astra that I shared earlier, it will also be proactive in providing us information. Combine that with display on the screen and you will walk around feeling like the Terminator, where you can see data about everything that you see pop up, about people's heads, about things you want to shop, about buildings and everything else straight into your eyes.

34:44So now, after these really crazy four announcements in a single week, let's go to some rapid fire items. The first one is actually from Satya Nadella. We talked about Microsoft earlier. Satya had a very interesting interview to Vanity Fair and another one to Bloomberg. And he's basically saying that he's building an AI that will replace him. The exact quote is, I'm trying to make myself obsolete. And he was talking about how he's using specific co-pilots and agents in his day-to-day life as a CEO of one of the largest companies in the world. He basically is talking about how he prepares podcasts to listen in his commute, something that I'm doing all the time with Notebook LM.

35:22And he can, quote unquote, talk back to the radio and ask follow-up questions. He shared that he's using at least 10 tailored large language model bots and agents in his work to summarize messages, prepare for meetings, and conduct different pieces of research, both internal and external. So think about the power of a CEO that is completely connected to everything that's happening in his company and in his industry and can query about that and get information about this almost immediately. It's nothing like we've ever had before, because before he had to spend hours and sometimes days in collection of data across multiple departments and people to then be brought to him.

36:00And then when he has a follow-up question, he has to wait another few weeks to get it. And now he can do it in minutes or maybe a day in the worst case scenario, which allows him to make better decisions faster, all based on actual information. This ties back very well to the topic I wanted to discuss in the beginning of today's episode, but I may just have to record a separate standalone episode about the AI company of the future. Now to stay on the agent-focused topic from Microsoft, Salesforce just signed an agreement to acquire Convergence AI, which is a London-based startup that is specializing in AI agent creation.

36:34They're one of the most used platforms today to create AI agents, and it's going to roll the technology and the team into AgentForce, which is their platform for agent creation. Marvin Putarab, the CEO and co-founder of Convergence said, our mission at Convergence is to help organizations stop viewing automation as just another tool and instead adopt it as the very way work gets done, unlocking new levels of innovation and efficiency. I cannot say it any better. This is where we're going, right? We're going from a point that these are going to be co-pilots, meaning things that assist us in doing the work, to the thing that's actually doing the work.

37:12Now, Salesforce itself, even without the acquisition of Convergence, is claiming that their own AI agents now solve 97 % of customer service queries, leaving only 3 % for human innovation. I don't know if that's accurate or not. I don't know if anybody tested that. But even if it's 50-50, it is very extreme in its impact on both the workforce as well as the efficiency in which organizations can run. Now, combine that with the fact that the 76 top American retailers are using Salesforce e-commerce platform. By the way, generating$136 billion in web sales alone in 2024. Just imagine what is the impact of all these agents on the economy and the efficiency of these companies from one side, and what does it mean to the workforce and then the people who have money to pay for these services on the other side.

38:00Staying on the same topic, let's talk a little bit about Klarna. We spoke about Klarna many times before on this podcast. there is a Swedish fintech giant that jumped all in into AI early in 2023, collaborating with OpenAI to develop multiple solutions for them, initially mostly for customer service. Well, they just announced that their employee count has went down from 5 ,000 to 3 ,000 since they started this initiative. The initiative didn't actually fire anybody, but it just went on a hiring freeze as they were developing AI capabilities, and through natural attrition, they lost 40 % of their workforce.

38:37Now, very early on in the deployment of their AI agents, the numbers they shared is that their new AI capabilities is replacing 700 human agents and cutting the response time from an average of 11 minutes to two minutes. That's$40 million annually straight to the bottom line. Now, additional information that was shared is that 87 % of Klarna's workforce currently uses generative AI every single day with non-technical teams like communication at 92.6%, marketing at 87.9 % and legal at 86.4%, leading the pack. On the flip side, it was very interesting to hear their CEO saying that the usage of AI-only customer service has actually reduced the quality of customer service.

39:23And they're now switching back. there looking to hire some human agents. And to quote their CEO, he said, from a brand perspective, I just think it's so critical that you are clear to your customer that there will always be a human if you want. So as expected, a blended solution of human support team plus AI to deal with most of the stuff is probably the way forward, at least in the foreseeable future. But the reality is that there is an entire industry of call centers and contact centers that support huge amounts of companies in the world and employing millions of people that will shrink from millions of people to probably tens of thousands of people because only a few cases will actually require human intervention.

40:06The output of the process that Klarna is going through is profound, but the efficiency output of Klarna's all-in approach to AI is incredible. Klarna's revenue per employee surged to nearly$1 million in Q1 of 2025, up 74 % from just a year ago. So every employee in Klarna on average, so you take the total revenue of Klarna divided by the number of employees, is now almost$1 million. The other interesting aspect of this is that their Q1 of 2025 revenue rose 13 % despite a significant reduction of headcount. By the way, that doesn't help them being profitable yet, they're still losing money and actually losing a lot of money right now.

40:49Probably a lot of it because of capital investment in AI. Either way, their IPO was pushed back due to uncertainties in the market right now. And I think if they figure out the AI stuff quicker, that's actually going to be a good way for them to get a higher valuation once they actually go public. A few pieces of news from OpenAI. Obviously, we cannot pass a week without some significant stuff happen on their front. So OpenAI is spearheading the development of a colossal 5 gigawatts data center in Abu Dhabi named Stargate UAE. The project is very interesting, first of all, because it's insane in size.

41:23The actual data center itself is going to spread over 10 square miles. Just imagine that. That's the size of a large neighborhood. Out of that, OpenAI will own only 1 gigawatt out of the 5 in cluster computing, but it puts a very clear statement of where OpenAI is going. And the quote from Sam is, by establishing the world's first target outside of the US in the UAE, we're transforming a bold vision into reality. Now, the other interesting aspect of it is the UAE government will be the first country in the world to enable nationwide chat GPT access. So the goal, going back to the growing digital divide from the UAE perspective, is that the government will finance chat GPT access to all the people who live in the UAE, at least in Abu Dhabi.

42:09This is an incredible move from the government. I really hope that additional governments will do that, but I really hope that the governments will also find ways to actually train people from young age all the way through professionals on how to actually use these tools and not just give them access. Now, the downside of that, which we talked about every time we talk about these mega project, is that this facility will consume the power equivalent to five medium-sized nuclear reactors. So the amount of power that this new facility will require is insane. And if it's not provided by clean energy, it has a very severe impact on the environment.

42:45But obviously, that's just the UAE project. OpenAI also announced an increased investment in their Texas data center, growing it from two buildings to eight buildings, bringing the total funding of that to$15 billion. It is going to be their largest OpenAI facility with 50 ,000 NVIDIA Blackwell chips. Now, all of this is actually allowing, obviously, OpenAI to reduce its dependency on compute from Microsoft as these companies are drifting further and further apart. And both these projects are a part of the Stargate project that was announced in January of 2025, together with President Trump and OpenAI Oracle and SoftBank to provide infrastructure for OpenAI's growth.

43:28And it is very clear that OpenAI is positioning themselves as the Microsoft and or Google of the AI generation with control over everything. They will control the models, they will control the agents, they will control the hardware, they will control the compute and so on and so forth. So a company that more or less didn't exist two years ago is gonna take a very, very significant role in world dominance from a technology perspective, which is absolutely incredible. A week back, OpenAI also unveiled Codex, which is an AI coding agent that integrates into ChatGPT and can handle multiple software engineering tasks simultaneously.

44:07Now, the new platform is powered by Codex1, which is a fine-tuned version of OpenAI 03. And Codex writes code, fixes bugs, run tests, etc., etc., more or less every aspect of coding. And it operates in a secure cloud-based virtual computer integrating with GitHub to get access to users' codebases, ensuring safety and defining and denying internet access during tasks in order to maintain your code yours and not giving it access to other things. The tasks that it can perform can run between one minute and 30 minutes. Again, a very big spread from the seven hours of Cloud Opus 4, but it's still allowing developers to delegate repetitive work to these agents as they are working.

44:48Combine that with OpenAI's recent acquisition that has not been finalized yet of windsurf which is one of the top code vibing tools out there and you understand that the ai coding market is on fire in a very bright glowing fire as well openai also announced that its chat gpt operator agent that is still in preview now leverages o3 as the model behind the scenes those of you who remember it's an agent that can run in your browser and operate your browser like you would on your own that they launched back in 2025, but only available to the people with a$200 a month pro subscription. And it had lots of issues in the beginning.

45:25So presumably O3, which is thinking capabilities, custom now tailored for this stuff, can really help users with filling up forums, grocery shopping, et cetera, et cetera. Basically everything we do on a browser. And this is obviously the next frontier of how we engage with the world. Going back to what I said multiple times, the days in which we browse the web the way we did so far are over and they're not over yet but the hourglass has been flipped and the sand is falling down and every day that passes we will start seeing less and less human traffic and more and more agentic traffic visiting websites which has as i mentioned multiple times profound implications including the way the web actually works by being financed by ads and by allowing people to create content that actually drives traffic to them well there's not going to be traffic at least not human traffic and everything else is going to change in the way we engage with data on the internet.

46:16Now, Jerry Twork, the VP of Research at OpenAI, said that that's just the beginning, and they're planning to make significant improvements to operator in the very near future to make it even more useful. Now, you start to connect all the dots on everything OpenAI is doing together with the latest release of Opus 4 and what it can do, and it starts giving you hints of what might GPT-5 look like, a model that can understand everything you need, that can connect to the internet, that can operate the browser, that can select which method to use in what steps, that has memory with all your historical data, which right now Claude doesn't have.

46:51And that's my biggest reason to maybe still go to ChatGPT, is that it already knows me and it can reference a lot of those things in the conversations. Combine that with MCP capability and the ability to connect to more and more data sources and tools, and you understand that GPT-5 and Opus 4 and all the new models that are going to come after that are going to change literally everything we know on how we work and even how we work with AI tools. Staying around the open AI topic, Sam Altman's world network, his digital orb that scans your eyes and make sure that you're actually a human, just raised$135 million in order to expand what they're doing.

47:28Fans will fuel their global network growth. They're targeting 180 million Americans for orb verified world IDs by the end of this year. I see that as a very ambitious and completely non-realistic target, but that's where they're targeting. For those of you who don't know what that is, it's a little orb that you place your eye against it, it scans your iris, and then it saves a digital ID on the blockchain so you can verify that you're actually a human and reduce fraud and verify that you are actually you. Overall, interesting idea. how that will actually work will be very interesting to see. And that's two very interesting pieces of news from government with its relation to AI.

48:09One highly controversial, the other one highly needed. So High Street Republican included a clause in their big, beautiful tax bill that was passed by the House Energy and Commerce Committee on May 14th, banning states and localities from regulating AI for the next 10 years, starting as soon as the bill is put into order. Their argument, which is the same argument that was announced by industry leaders in the Senate committee that we shared with you a couple of weeks ago, they're trying to prevent patchwork of state regulation that will make it very, very hard for these companies to maneuver through.

48:46This will immediately block over 20 California AI laws and 30 pending bills just in California and a lot of similar bills in many other states, including protection against deepfakes, and AI-driven healthcare denials, and many other laws that actually make perfect sense. A lot of people who object the law are saying that right now, the federal government did not put in place any rules and regulations to help Americans in dealing with the negative aspects of AI, and yet this law, if it passes the Senate, will be put into action. The flip side argument from Ted Cruz and others on the Republican Party is comparing it to the 1998 Internet Tax Freedom Act that fueled the e-commerce growth in the US.

49:29We talked about this when we talked about the AI Senate hearing. Now, from the state's perspective, 40 state attorney generals plus 140 organizations, including the Center for Democracy and Technology, arguing against the law, and they're saying that it's leaving consumers unprotected. As I mentioned, there's no clear federal law to protect any of us against the downsides of AI usage. The other legislation is not only about AI, but it's also about AI. President Trump just signed the Take It Down Act, which is a bipartisan law that is criminalizing the distribution of non-consensual intimate imagery, whether real or deepfakes.

50:07That law is effective immediately, and it was brought up by Ted Cruz, who is a Republican, and Amy Klobuchar, who is a Democrat. The bill passed the House with a 409 to 2 vote, and unanimously in the Senate. Definitely a great step forward. This law was actually promoted and championed by Melania Trump, and I'm very glad that it's actually passing. It's not protecting any deepfake, but at least it's protecting from harming individuals with intimate fake photos of themselves that can generate catastrophic harm for anybody, especially teenagers and younger people. The next piece of news is a resurrected attempt by the Y Combinator-backed company Firecrawl, who now allocated$1 million to hire three AI agents for content creation, consumer support, and junior development roles, paying each$5 ,000 a month.

50:57Now, they've done this publicity stunt a few months ago in February, and we shared with you about that. That didn't work very well, but this time they're saying that they already got 50 applicants in a single week since they posted these jobs, which is telling you how quickly the AI agent world is evolving with companies and individuals developing agents and want to sell their services to others, which is going to build a whole new layer to the economy. Now, the million-dollar budget also covers hiring the people who are developing these AI agents as full-time or as contractors to the company.

51:29And you're asking what these agents will do. So if you want to dive a little deeper, the content agents must autonomously produce SEO-friendly blogs, track engagement, and improve accordingly. The support agent handles tickets in under two minutes, and the developer agent codes in TypeScript and Go. And there's obviously much more detailed instructions in the job posting themselves. Now, there's an interesting catch-22 in all of this when you think about it, where companies are going to hire people, developers, who can create agents that will eventually will replace them as well because agents will know how to write code and create new agents.

52:02So that makes my head get stuck in a loop. But that's where we're going, and we're going there very, very fast. Speaking about startups, a Silicon Valley Bank report reveals that 40 % of US venture capital in 2024, went to AI-focused funds, 40%, doubling that share from just five years ago, leaving non-AI startups scrambling. This brought back a phrase coined by Mike Maples in 2016 called zombie corns. Zombie corns are startups valued at over$1 billion with stagnant revenue growth and deemed prospects for future funding or exits. To put things in perspective, in 2021, 138 enterprise software unicorns emerged in 2024, only nine, with not a single one in 2025 so far based on this report.

52:52That reflects a very significant funding drought for anything other than AI. Combine that with the fact that many of the startups that are actually doing well are getting killed by new AI capabilities features within ChatGPT, Cloud, etc. And you understand that the startup world is going through a very serious storm and it will come up on top and people will figure it out. But right now, it's a very serious issue for startups who want to raise money and grow stuff that is not AI related. And switching from software to hardware or specifically robots, which we didn't cover for a couple of weeks, Tesla released a new video of their Optimus robot showcasing some very impressive capabilities, performing tasks like throwing trash, vacuuming, steering food, and even moving Model X parts onto a dolly.

53:41The interesting thing about all of those is not the fact that the robot can do this, but that these robots learned how to do these skills by looking at first-person videos of people actually doing it. So they weren't programmed, they weren't trained in any way other than just watching these videos. And Tesla is planning to actually improve that capability of the robots to learn from third-person view, basically internet videos of every task that you can imagine that exist at abundance and allow the robots to learn through that, that is obviously going to dramatically shorten the time of training robots to do new tasks safely and effectively.

54:16If you remember, Musk thinks that humanoid robots are the biggest product ever, and he's estimating a $25 trillion market for autonomous robots in the future, doing more or less every blue-collar job, including house chores, as you've seen the robot doing. Tesla already begun limited production of the robot at the Fairmont factory, targeting 5 ,000 to 12 ,000 units in 2025, and then starting to sell them externally in 2026. We've heard a lot of predictions from Elon before that most cases, I would say almost all cases fell short. So these numbers might be optimistic, but the direction is very clear, especially that there's competition from many other companies who are building advanced humanoid robots.

54:57That's it for this week. There are a lot of other really interesting news that we just couldn't fit into this crazy episode. And you can find all of them in the newsletter, in short snippets, quick bullet points, and links to the actual articles where you can actually read more if you're interested in one topic or the other. You can sign up for the newsletter in the link in the show notes. If you enjoy this podcast, please open your phone right now, unless you're driving, and click on the share button and share it with a few people you know that can benefit from it as well. Having more people understand what is coming and how to use AI is critical for us as a society, figuring it out and enjoying the benefits while reducing the risks.

55:35And your ability to play a part in this is literally by clicking that share button and sharing it with more people, a very easy task that you can do in about five seconds. I would really appreciate if you do that. By the way, we just announced a new date for the AI Business Transformation course. The next cohort is going to open to the public in the beginning of August. So if you're interested in that, you can click on that link on your show notes and take a look at our courses. All those courses sell out. These courses have been transforming businesses for over two years. I'm currently running two of them in parallel, but most of these courses are private.

56:08And so if you're interested in a public course that you can just sign up to, go and sign up right now while we still have seats. That's it for today. On Tuesday, we'll be back with another How To Fascinating episode. And until then, have an amazing rest of your week.