269 | The world as we know it is over- AI can do any knowledge work. New models: Sonnet 4.6, Gemini 3.1, Grok 4.2, AI leaders sound the alarm, but the US is pushing forward, and more important AI news for the week ending on February 20, 2026

26:29And he's stating that this kind of surveillance is fairly important for the company, meaning anthropic, and for democracy to put a red line of what is acceptable and what is not on both these topics. He was also talking about economic disruption, and he mentioned, and I'm quoting, we believe that AI will greatly grow the economic pie, but we need to work together to better manage that time of disruption and bring better prosperity smoothly to all. And again, do I think AI can generate huge business value? A hundred percent. Can it grow the overall economy? For sure. Will it be distributed in a way that will be helpful to the vast, to most of humanity?

27:08I sadly think that in the current setup, the answer is a hundred percent no. And this could lead to a complete chaotic situation. And this is what he's talking about, that maybe in the long-term future, AI will provide prosperity to all and a future of abundance, as he said many times, but the short to midterm is going to be really problematic and really scary. And everybody, governments, academia, industry, and cross-national partnerships have to happen in order for this to evolve in a positive direction versus a negative one. Or as he summarized, AI has the potential to cure diseases that have been incurable for thousands of years to radically improve human health and to lift billions out of poverty.

27:50But because it is happening so fast, it may lead to a time of disruption. And I'm certain in the second half, I hope for the first half. Sam Altman also talked about the urgency of global regulation on this. And I'm quoting, centralization of this technology in one company or country could lead to ruin. This is not to suggest that we won't need any regulation or safeguards. We obviously do, like we have for other powerful technologies. Brad Smith, the vice chair of Microsoft, also had similar warnings. And he talked about the fact that the digital divide or the technology divide may grow a lot bigger unless we take specific steps.

28:31And he said, and I'm quoting, closing this economic divide, exurbating it, and making it wider. And he stated that the key to enabling people to use this technology is skilling for people and that we need to focus on, and I'm quoting again, what people will use the technology to do for people. By the way, this point of skilling and training is something that was mentioned in a very clear way by Matt Schumer that actually had a few, more than a few paragraphs on what you should actually do, including start using it seriously, not as a search engine, including using the paid version instead of the free version, because there's a huge difference between the two, including going deep instead of wide on topics that you're using AI for, and including learning new skills that are AI-infused in order to increase your chances of actually having a job in the near to medium future.

29:23Now, we also heard some opposing aspects to this. First of all, Arthur Mensch, who is the co-founder of Mistral, which is a French AI company, said, these are mostly distraction tactics. In reality, the real risk of artificial intelligence in the near future is of massive influence on how people think and how they vote. And you heard me say that on this podcast many times, and I agree. The nearest term is from the collapse of what is real and what is not, and how can that impact how society works as a whole. I mentioned that multiple times in the past few weeks, but I'll mention it again. If you want to hear my detailed opinion, go back to, I believe it's episode 10 of the podcast.

30:03It was called The Truth is Dead. Just look for that episode and you can hear everything I think about this. But I agree with him that the nearest immediate risk is that. That doesn't take away from the much bigger risks moving forward. Michael Kratosis, who is the White House technology advisor, said that he does not agree that we need global governance. And he stated, and I'm quoting, we totally reject global governance of AI. AI adoption cannot lead to brighter future if it is subject to bureaucracies and centralized control. I strongly disagree with what he's saying. I know that this has been the position of the current administration to push forward as fast as possible, regardless of anything.

30:42I think this is a huge mistake, and I hope we're not going to pay a very dear price for that mistake. By the way, Emmanuel Macron, the president of France, was actually pushing back a little bit on the US position that always looks at Europe and says, hey, what's too much regulation is actually doing? And he declared, and I'm quoting, Europe is not blindly focused on regulation. Europe is a space for innovation and investment, but in a safe space. I think the truth is somewhere between them. I have clients in Europe and I have people who are in my community from Europe and they definitely think Europe is over-regulated right now.

31:17I think the US approach is wrong as well. And there's probably a happy median between them. And I really hope that the entire world, including China, including India, including Russia, including Europe, including everybody, will find a way to find that happy median sooner rather than later. Now we'll switch to our rapid fire item. The first two items are directly related to everything we talked about in the deep dive on how AI is impacting the real world. And we're already seeing some very significant signs of how it's impacting real life in real companies and real aspects of the economy and glimpse of where this may lead to.

31:50So Accenture has announced this week that it is conditioning the promotion of the leadership roles on regular use of AI tools. Associate directors and senior managers across most of companies' 780 ,000 workers must demonstrate regular adoption of AI in order to get a promotion. Tool usage is becoming something that they're tracking. So how many times you're going into AI tools, how frequently you're using them, and for what aspect is becoming, and I'm quoting, a visible input to talent discussions. Their CEO, Julie Sweet, stated that all employees would be expected to, and I'm quoting, retain and retool at scale.

32:32And she's also acknowledging that the company would exit employees without, and I'm quoting again, a viable path for skilling to acquire needed capabilities. This is the first major company that I hear speaking very clearly out loud saying, you're either going to use AI or you are out. And it's starting at the highest level. So they're starting with the managers. And this will obviously then trickle down to lower tiers of the company as well. And based on everything that I've been seeing in the past few weeks in things that I'm doing and that leading companies that work with me in my consulting hat are doing, they're 100 % correct in that approach.

33:09They may not fire these people that quickly, but definitely in the long run, people who will not know how to use AI effectively to at least a decent level will not be able to be productive to an organization. Another interesting topic that is related to AI adoption and how quickly it can spread in different areas is Eleven Lab has just secured an AI UC1 certification, and they're becoming the first company to offer insurance to cover for AI voice agents. So let me take a step back and explain what the hell is AIUC1 certification. AIUC stands for Artificial Intelligence Underwriting Company. It is a new insurance company that ensures the usage of AI in different companies.

33:51And they have created a certificate program that is called AIUC-1. What that program does is it tests different AI solutions for data privacy, security, safety, reliability, accountability, and society. and you have to pass rigor testing, including 5 ,835 technical tests across 14 risk categories in order to earn the certification. But what this means, it means that right now, if you are an enterprise or a smaller company and you want to be able to be insured against claims that come because you are using AI as your sales team or customer service and so on, you can do that. Meaning one of the biggest barriers that companies had for AI voice agent adoption is now removed.

34:37For those of you who don't know, Eleven Labs is one of the leading, if not the leader in the voice AI space, and its 11 agents solution is now being deployed at 75 % of Fortune 500 companies, including companies like Cisco and Square and other giants. And their ability to now provide insurance is obviously going to accelerate the adoption to more aspects in these organizations and many more organizations. The way Eleven Labs were able to do this is by building a three-stage protection into its platform. The first part is pre-production where they have safeguards and red teaming and simulations.

35:08The second one is in conversation enforcement via system prompts and guardrails and real-time moderation that can automatically end unsafe conversations and ongoing monitoring with automated evaluation criteria that is applied across all calls. So with these three tiers, they were able to now get the certification, which means anybody who's using 11Lab voice agents can now be insured against the outputs of what these agents do. From that to a few news on shifting capabilities, and we're going to start with OpenAI. Sam Altman just told CNBC, first of all, that Chinese tech companies are progressing, and I'm quoting, remarkably fast.

35:41The initial part of this conversation was in which Sam Altman said, and I'm quoting, the progress of Chinese tech companies across the entire stack is remarkable. When he's talking about the entire stack, he's talking about the fact it's not just the models, but also the underlying infrastructure, as well as the hardware that is now Chinese built and how they're closing the gap across the board. He also said, we are growing at an extremely fast rate right now. I think as long as we can have reasonable unit economics, we should focus on continuing to grow faster and faster, and we'll get profitable when we think it makes sense.

36:14So you're probably asking yourself, what do you mean when you think it makes sense? You just decide to be profitable? And the answer is yes. There was a very interesting interview of Dario Amadei with Rakesh just a couple of weeks ago, by the way, Rakesh had Elon Musk and Dario back to back in two separate episodes, both definitely worth a listen. In that interview with Dario, he states that he still thinks that we're going to get a country of geniuses in a data center within the next two years. This basically means that you're going to have Nobel Prize level in any topic in abundance for anyone.

36:43Now, he's saying that this can unlock trillions of dollars in potential revenue, but what is going to slow it down is the rate of diffusion through the economy. So as I mentioned earlier, it's going to take time for countries, companies, and individuals to learn how to leverage the technology and how to go through red tape in means of data security, integrations, and so on in order to actually see the benefits. And this is the reason why he's hedging a little bit his investment in future compute. He also said that they've been growing 10x every single year since the launch of the company. They reached 10 billion this past year.

37:21He assumes at the current pace, and he thinks they're going to make it, reach 100 billion in revenue in 2026, which means they will make a trillion dollars in revenue in 2027 if the curve continues. He doesn't think it is going to continue, but he's saying that based on that, he should be able to invest$5 trillion in compute commitment for the next five years, $1 trillion per year, and he should be able to pay it. But he's saying that even if he's wrong by 20 % and he's not going to make a trillion dollars in two years, he's going to make$800 billion in two years, which is still insane amount of money.

37:56And it is an even more insane amount of money for the company that is three years old. He will still go bankrupt. So if he commits to the one trillion per year and he only makes 800 billion per year, nobody will give him $200 billion to pay for the rest of it in order for the company to survive. What he's saying in general terms is that the way all these companies are operating right now is they are trying to project how much demand they're going to get. Based on that demand, they're going to decide how much compute they need in order to serve the demand versus how much compute they need in order to continue doing research.

38:30And he's saying that if they miscalculate their projection, two things can happen. If they have more demand, that means they need to do less research and they may stay behind. And if they have a little less demand, they can do more research to build better models, to be more competitive, to get more demand. But if they get significantly less demand, they will go bankrupt because the commitments are so crazy and the gaps between being successful and not successful are so high. But what it also means, connecting it back to the point from Sam Altman, it means that they can technically decide when they become profitable.

39:01And he's saying that point will come with the point of diminishing returns on the, not on the technical scales of the models, but on the business value of the models. We'll get to the point that the models can grow from a technical perspective, maybe at the same scale they are right now, but it will make very little difference to the economical value that they provide. At that point, you don't need to do as much research at the same pace, which means you need less compute for research, which means the extra money that you have doesn't go into buying more compute for research. It actually becomes profit.

39:33By the way, since we talked about Dario in relation to OpenAI, and we already covered the India AI Summit, Sam Altman and Dario Amadei were the only people that in a very visible way declined to hold hands in the final image, in the group photo of all the leaders that were on stage at the same time. So if you look it up, it's actually pretty hilarious. Like kids in the kindergarten, everybody's holding hands and raising their hands together other than two people that somehow somebody decided to put one next to the other on stage, which is Dario and Sam, and they are avoiding holding hands while everybody else is.

40:05Now back to OpenAI. OpenAI just acquired OpenClaw. So we talked about OpenClaw several times in the past few weeks. OpenClaw is a viral open source AI agent project that was created by Peter Steinberg has took the geeky AI universe by a storm. It is basically an AI agent that can work 24-7, 365, and complete whatever tasks you give it access to. It can connect to every system in your tech stack. It can run locally on your computer and really run every single thing that your computer has access to. And the acquisition was announced on February 17th. They are presumably going to keep this as a completely separate unit, meaning it is going to run OpenClaw as a separate department unrelated to the main OpenAI universe.

40:49Now, this acquisition is interesting from several different reasons. One, it was very obvious somebody's going to pick it up because of the amount of buzz it created. The technology is not that unique. I mean, once it was obvious how this tool works, anybody, literally you yourself, can replicate how this is working. But the buzz and the community that built around it is worth owning because all these people are at the far edge, the tip of the spear of AI experimentation. And the people who are experimenting with it are willing to take risks that enterprises won't. So if you own this technology and you have the ability to put it under a microscope and see exactly how people are using it, what use cases, where is it breaking, where is it safe, where is it unsafe, and so on, you can then bake these capabilities into the safer, bigger OpenAI technology instead of just keeping it in a box somewhere.

41:35And I think that is the reason why OpenAI decided to buy OpenClaw. Obviously, huge, crazy, insane success for Steinberger, who is now, I don't know because they didn't disclose the numbers, but either a multimillionaire or potentially a billionaire. I'm not sure exactly how much money he got, but I'm sure he got a crazy amount of money for a project he was working on for a very short amount of time. I'm also curious what was happening behind the scenes and how many other offers he got before he decided to go to OpenAI. If you remember a long time ago, when we talked about this two weeks ago, when the craze started, I told you that the original name was Claudebot, spelled not like Claude the AI, but sounds definitely like it.

42:12So it sounded like it was written like Claw, C-L-A-W. And yet Anthropic sent him a season deceased letter to change the name, which had to change twice. But I'm curious if a minute after the season deceased letter, they send him a potential acquisition letter. I don't know if we're ever going to learn that, but I assume there were other companies who were trying to buy OpenClaw other than OpenAI and even companies outside of the big labs. Like I see people as Salesforce being maybe interested in something like this or these kinds of companies as well. Another company that came to my mind is Meta, but they recently bought Manus, which is their path to potentially increasing their agentic capabilities.

42:48They're not exactly the same thing, but the general direction is the same direction. What Steinberger said, he said, what I want is to change the world, not to build a large company and teaming up with OpenAI is the fastest way to bring this to everyone. Now, by the way, one more thought from me. While the technology is really easy to build, OpenAI or any of the other big labs cannot release something like this. They are focusing on enterprise-grade solutions, and this is way, way too risky. And hence, despite the fact that they could build something like this, it wouldn't make sense on the main path.

43:20So having a parallel universe in which they can experiment and see what's happening is very, very powerful and going to be very helpful for OpenAI. Staying on the OpenAI topic, it seems that the ink is going to dry on their$100 billion round at an$850 billion valuation in the next few days if it hasn't happened yet. This is one of the largest, if not the largest, single funding round for a non-public company in history. The biggest players in this investment are Amazon, which are likely to put in$50 billion soft bank that are looking at$30 billion in NVIDIA that are reportedly nearing$30 billion in investment.

43:59If you do a quick math, that's$110 billion instead of$100 billion, which means they might raise more than$100 billion when this is all done. Now, OpenAI already committed to reinvesting a lot of that money into NVIDIA hardware running on Amazon AWS. So if this is not a circular kind of investment scheme, I don't know what is, but this is where we are right now. By the way, despite these soaring valuations, and they're talking about how fast they're growing, they are bleeding crazy amounts of money. And the numbers that have been thrown around is that they're spending more than$700 million a month for compute infrastructure alone.

44:37That's without the crazy salaries that they're paying and running the business and overhead and legal and stuff like that. To make it even more interesting, and we discussed this in the past, they have$1.4 trillion dollars of commitment over the next eight years for new compute, which means this hundred billion is a, maybe not a drop in the ocean, but a glass of water in a bucket to the amount of money they actually need in order to keep running their company without going bankrupt. A few interesting new things in the robotics space, Google Gemini 3 Flash, which is their fast and extremely effective model that they released just a few weeks ago, is now being used in order to train robots in an effective way.

45:16So because it knows how to analyze video very well, they're now using it to translate video input into JSON inputs for robots. So the AI is watching a video of a specific process, translate of a specific process. He watches it multiple times, about 50 to get it perfect. And then once he watched it 50 times, 50 humans doing the process, it will then create a JSON set of instructions where the robot can use to execute whatever the process was. What this means is that they don't need to write code or develop specific training environments for robotics. They can just show them videos of people doing the tasks and the robots can learn how to do it in almost a seamless way using a really cheap and effective model like Gemini 3 Flash.

46:01Another robotic-related announcement this week came from Unitary. Unitary has been one of the leaders in robotics in the world for a few years. They're a Chinese company who has several different models. They have manufactured an estimated of 5 ,500 units in 2025, and they're projecting to either double or quadruple that in 2026. So they announced that their goal is to ship between 10 ,000 and 20 ,000 robots this year. Now, their robots that are doing crazy things every single time they're being shown has captured a lot of attention. In the spring festival gala that just took place, they had an entire segment with multiple robots were performing perfect kung fu style action fully autonomously.

46:44They had some of them climbing walls and showing skills are way beyond walking and doing basic daily stuff. Now to make this even more interesting, Unitree has been profitable since 2020. Their revenue has exceeded$140 million this past year and if they keep on growing in this trajectory, they will be a highly successful player in the robotic space even moving forward. We've talked about them many times in the past when I was mentioning that their G1 robot, that is their little robot that they're creating, it's about four feet tall, costs between$6 ,000 to$16 ,000, depending on the variation that you have.

47:16And there's already companies in China that are renting them out for different applications. This is compared to tens of thousands of dollars or hundreds of thousands of dollars for full-scale humanoid robots from other robotics competitors. But I want to mention an interesting quote from their CMO as part of this announcement. if 2025 was about the race to mass produce, 2026 is about the race to deliver. The focus has moved from factory floor to the commercial channel where real measure of a company's health is no longer how many robots we can build, but how many it can successfully integrate into real world scenarios.

47:54Meaning, in addition to everything we talked about in the beginning of this episode about knowledge work and how AI can practically do basically any knowledge work right now, we are already seeing robotics company who are saying that their plan for this year is to focus on integrating robots beyond production floors and into day-to-day life. And the third interesting example, before I tell you what I think about the current status of robotics, comes from a company called Weave Robotics, who has launched Isaac Zero, which is a stationary laundry folding robot that is now available for purchase for$8 ,000 or$450 monthly rental.

48:33It knows how to fold most standard size clothing. It still does not know how to do blankets and bedsheets and something that is really large or inside out clothing because it doesn't know how to flip them around. It looks really weird. It's not a humanoid robot. It's like a laundry folding robot, but apparently folding laundry is not such an easy task. Figures humanoid robot, figure 02, achieved the first fully autonomous laundry folding back in August of 2025. Since then, one or two companies was able to replicate this, but this robot, for a really cheap amount of money, knows how to do this at scale.

49:04Now, the Isaac Zero robot operates in a hybrid model, meaning it completes all the stuff that it knows how to do on its own in an independent way. But when it needs assistance, a human specialist can take over for 5 to 10 seconds and intervene in order to get the job done and get the folding process done properly. Now connect this to what I said earlier related to robotics ability to learn from video and examples. And you understand that when the human intervenes and is helping the robot do whatever it is doing, it needs to do it only X number of times. And over time, it will learn how to do this on its own, which basically means that these humans that are helping remotely these robots to do their tasks.

49:44and they're not the only company who uses this hybrid approach, are on a basically doing, what it means is that the job of these people that are helping the robots remotely, and they're not the only company that is doing this, this hybrid approach, these jobs are basically in an hourglass. And the sand is falling down through the hourglass every time they're helping one of these robots do a specific task because after X number of times, they learn how to do it on their own. And then the humans that do the remote job are not necessary anymore. By the way, my other company that does AI agent-based invoice reconciliation works in a similar way where it can learn over time based on the corrections that humans are making to what the agent's output is.

50:22So this is not rocket science, and this is what we're going to see everywhere all around, agents learning what we do, learning how to replicate it, and then replacing us in doing that work. Opening parenthesis for that for a second, if you are in a company that does a lot of invoice vouching and it's taking you a lot of time and costing you money of people doing mostly that, reach out to me on LinkedIn. We're now accepting new clients to this really remarkable platform that does the work of a team of three people in minutes instead of a full team working on it for days. Another topic that I think will be relevant and applicable to many of you is that WordPress just launched a built-in AI assistant to create and edit websites.

50:58So I told you last week that there's a new MCP access from Claude to be able to manipulate WordPress. But now there's a built-in WordPress AI assistant that can help edit websites and create new pages inside of WordPress just with simple prompts. The tool also knows how to edit and generate images with Nano Banana, translate and edit text, recognize site layouts, create new pages, and basically do everything a human can do inside a WordPress website. Why is this a big deal? Because WordPress hosts most of the world's websites. And so there's a decent likelihood that your website runs on WordPress as well.

51:33And so having this kind of capability makes perfect sense for WordPress. And by the way, for any other software company, adding AI assistant inside the software, even the software that I'm using right now, the application I told you before that helps me manage the news has an AI assistant built into it that can help me do more or less everything that I can do in the software as well. Every company on the planet will have something like this. The thing is that over time, this will replace the actual software because you won't need the software. What you need is the output of the software. And if you can do this with a few words or sentences, why do I need all the user interface that exists today?

52:06I don't know. And if that is the case, why do I need the software? I don't actually need it. I can do it with a third party AI that has access to the data. And this is, I think, where the world is going right now. We haven't talked about Apple for a while. And for many reasons. They haven't done anything interesting in the AI space. They are far behind and there hasn't been seems to be a cure on the horizon. But the news this week is that they are going all in on AI wearables. So we shared with you previously that they're working on a pendant that will be like an AirTag size pendant with two different cameras and a microphone and a speaker.

52:39Well, apparently they're working on a family of products similar to what we heard from OpenAI with their Johnny Ive team. So apparently they're working on that pendant together with smart card glasses, and AI-enhanced AirPods. Now, I told you before when we were talking about the pendant that I did not understand why the first step from Apple is not doing an AI-enhanced AirPods. It makes perfect sense to me. They already have a huge audience that absolutely love AirPods, that swear by them, and that puts them in their ear all the time, 24-7, unless they're charging them. And this would be the perfect audience to actually provide AI capabilities to.

53:12All you need is to add an AI layer on top of that. And even if you want to go crazier, add a small little camera in there that can see what's happening around you and you have a solution that already a huge audience is already using wearing doesn't have to get used to so that is a part of that and the third component is smart glasses it's not going to be augmented reality meaning it's not going to have a display it's just going to be able to see through cameras and be able to communicate with you through voice so you can speak and hear what it's saying back while it's being aware of the environment so kind of like the current generation of glasses from meta but not the next version of metaglasses.

53:48All these devices are going to be developed during 2026, and the release is expected either at the end of this year or the beginning of 2027. I said that multiple times before, I think AI wearables will become a big deal, and the more they are in a form factor that we're used to right now, so not pendants, but glasses and earbuds, I think they're going to see a huge success, and I think in the not-too-far future, most of these devices, definitely earbuds, will have AI infused into them. That's it for today. There are a lot more news that did not make it into the podcast episode, but they are available on our newsletter.

54:23So you can go and sign up and learn other things that happened this week. Also on the newsletter, we have other information. It's not just the news, but things that are happening that you can join, different courses that we're running, where you can learn different things, new apps that are coming out and so on. And you can sign up for the newsletter straight from the show notes. If you are enjoying this podcast, I would really appreciate it if you go to your favorite podcasting platform, whether it's Spotify or Apple podcast and rate this podcast, give us a review and also reach out to me on LinkedIn and let me know what you think and what else do you want to see in the show or what less do you want to see in the show?

54:55I love getting feedback from you and I'm getting several people every single week who are reaching out to me and I appreciate all of you. And also while you're opening your phone and you're looking at this app, please click the share button and share it with other people who can benefit from learning about AI, which is more or less everybody. And I'm mentioning just one thing that I'm looking for your feedback on the two things that I mentioned earlier, the infrastructure for agentic AI development for people who do not want to worry about the infrastructure, including all the mini agents and everything that I'm developing, as well as a detailed course for beginners to learn how to build really advanced multi-agent orchestration solutions for more or less anything in your company.

55:35That's it for now. Keep on experimenting AI, keep sharing what you learn, and I will be back and I will see you again on Tuesday. Abiful at ему17. Ev 'em Конкуная без сours. HeroinА с бросил колеб Тогда. Esrend 막ono.