Latest / The Bleeding Edge Podcast / Weekly AI Brief 20 Feb 2026
Transcript
- Ralph Behnke: Hi and welcome to the Bleeding Edge podcast. I'm your host Ralph Benkhan with me on the show. My friendly furry friend, Emile Ogier. Emile, you're looking very dapper there, my friend. Emile Ogier: Thanks, Rob. So are you. Good to be back. Lots to talk about. Ralph Behnke: ⁓ lots to talk about. And going to be a slightly different approach today, which I hope ⁓ will interesting in terms of the news. ⁓ But news will be as fresh to everybody on the show as it is to me. So you say like to wing it, Emil. This ⁓ episode is to be all about winging it with AI. Let's start with the news. So ⁓ some Obviously big powerful stories this week, latency, capital and regulation. Those were the themes. So with latency, first off, OpenAI launched GPT 5.3 codecs Spark. And so this thing is the research preview that's designed specifically for real-time coding. And it's a bowl accounts, know, a very powerful system. But here's the thing, can stream output of over 1000 tokens per second. So you've got to kind of think about where intelligence is going. It's going to be expressed as tokens. Now this thing's doing 1000 tokens per second. more than, you well, that's certainly more than most people are going to use in a day, ⁓ probably in minute. And so this is quite a big deal. ⁓ And I think we're going to see a whole new realm of tokenomics. Secondly, capital, the money story, Anthropic. And ⁓ announced their big 30 billion, what call post-money valuation. I don't know, interesting term. Maybe you know what that is. But they are ⁓ at G with a 30 billion funding round announced and 380 post-money valuation. So... ⁓ I think now in total, they've raised something like 160 billion since they started raising funds. So really interesting to see a company like that approaching series G still pulling in like 30 billion. And OpenAI ⁓ are their IPO for ⁓ mid 2026. like they're aiming for one trillion in valuation. So just to the anthropic there in perspective. The one, ⁓ we about regulation. ⁓ Google hard on science mode as a product. this is obviously in response to Google really wanting to become the number one platform for research-based ⁓ usage. OpenAI also a couple of weeks back their research-based tool. this is kind of ⁓ because Google has broadened access and available in the Gemini app, ⁓ but for Google AI Ultra subscribers. I'm pretty but I'm not quite ultra, unfortunately. At the moment, I'm busy ⁓ maxing it with Claude code. So Google is a case that some users will pay extra for a model for research grade work. And I think this is true because we're looking at the verticalization of so many AI solutions. I don't know why I waved in the air like that, definitely, know, we're kind of optimization, I think is going to drive AI practical use cases further in to organizations when it's actually made specifically for a purpose. And then course, Infrastructure. So this was really fascinating. NVIDIA and Meta have deepened their relationship. they're basically in new deal that's in terms used to be hyperbole. Meta is expecting to deploy ⁓ millions of GPUs. Blackwell and Rubin, apparently. ⁓ Who knew got such nice names. NVIDIA CPUs, Spectrum X networking. that they're going to be using for their hyperscale AI data centers. So lots of interesting words there, nice to see Meta, you know, still trying to stay in the race, I guess that's kind of the interesting thing. And ⁓ a of near shore stories, just, you know, Europe's strategic AI storyline. Reuters have been reporting Mistral AI acquired Koyeb. It's a Paris-based cloud startup. Obviously, Mistral is French. a European sovereign AI, but it's running out of France. And this tool looks like a deliberate push towards a more vertically integrated European AI stack. So there we go, sovereign, vertical, all the same words. And so it's essentially a plus infrastructure. And we've seen... A couple of weeks back, Mr. Ruhl was making a big investment a data center up in the Nordics, also as part of building out that ⁓ model plus infrastructure. So ⁓ I those are the big stories this week. Kind of interesting. Anything take your fancy there, Emil? Emile Ogier: It's always interesting, Rolf, but I think I'm just going to move on to my first story and that's been catching my attention. And I'm sure many other people, especially in America, the Pentagon is considering cutting ties with Anthropic after genuine standoff of military use of Claude. Pentagon wants freedom to use AI for all lawful purposes. They essentially asking, let us use your models for anything that's technically legal. ⁓ i.e. weapons development, intelligence collection, battlefield operations, and mass surveillance. OpenAI has said yes, Google has said yes, XI have said yes, and Thropic have said no. Not to everything though, they were willing to negotiate, but they held firm over two things. And Thropic is saying no to mass surveillance on US citizens and to fully autonomous weapons that fire without having a human in the loop. So they essentially putting a 200 million contract at risk, highlighting board attentions within the Trump administration to push for AI as part of their defense strategy. Now Pentagon is threatening to blacklist, not how are we or some Chinese chip maker, but one of America's most valuable AI companies. And here is what, why it matters. It means that every single company that wants to do business with the U S military has to certify that that they don't use Claude anywhere in their operations. And eight out of 10 of the largest companies in America currently use Claude. All of them would have to cut ties with Anthropic overnight or lose their government contracts. Even a senior Pentagon official told Axio, it'll be an enormous pain in the ass to disentangle this. And we are going to make sure they pay the price for forcing our hand like this. So you have one company trying to hold the line and now a government wants to make an example out of them. I think this is an important development because whatever Anthropic decides in the next few weeks, it may set precedent for how much control AI companies actually have over their own technology. Pretty interesting stuff. Ralph Behnke: Yeah, I don't know what it's going to feel like to lose a 200 million contract, but I guess we have a price on values. There you go. Emile Ogier: Yeah. And just wanted to add one more thing because in the previous episode, we covered of the key researchers of open AI leaving. ⁓ I did do some more digging into it. And one of his key responsibility was safety around bio weapons. I making assumptions here. Nobody's announced it. But if you look at those two pieces, he did mention that were internal pressures and I wonder if it was this government contract that pushed his decision. Ralph Behnke: Yeah, it's a big game now, know, safety is such an important feature. And I think for Anthropic, especially, it's of their brand. So maybe this is really just a 200 million advertising campaign for their credentials as really safety first. I think the public will respond well to something like that and maybe enterprise too. But the end of the day, you need government support. Emile Ogier: Absolutely. And I think main thing I'm interested in is just to see how the AI labs collaborate together, whether they're going to have any official stance or if they're just going to allow governments set precedent. But yeah, let's see what happens. Ralph Behnke: Very interesting. So know ⁓ been ⁓ the story of Peter Steinberger over the last couple weeks. The guy behind Claude Bot, Maltbot, OpenClaw. I have never heard many kind of weird lobster ⁓ parts being ⁓ used for naming of I guess Nano Banana, we can expect anything. But yeah, Peter Steinberger's journey to open AI. think that's really the story of the moment. So this guy was an independent developer. the goes, ⁓ you he went from launching his agent Claude bot, I guess what was it called Claude bot back in the, in the beginning. ⁓ And so eight weeks later, he's basically inquired. Emile Ogier: Indeed. Ralph Behnke: So acquired inside open AI to lead some of their agentic ventures it's I guess one of the fastest launch stories in town. ⁓ seen people You do some big launches in under 12 months with AI products I'm thinking base. What is it base 45? Did I get that right? Base 54. ⁓ my god Anyway, so a really interesting story with Peter because this Austrian software engineer, he spent 13 years building PS PDF kit, which was highly successful PDF framework and was used by lots of big players. But he bootstrapped the company for a decade and sold his shares in 2021. when Insight Partners invested $116 million. So despite his big financial success, he walked away with something like $100 million. The exit left him feeling broken and devoured of purpose. Wow. I don't know. I feel really sad for him. But at least he had some cash to, you know, take a nice holiday. He stopped programming entirely. He took a three year hiatus and he pursued therapy apparently, which is a really good thing to do. But he traveled and tried to find himself. So, God, I mean, he's kind of describing my 20s. Steinberg. Instead, he discovered the open claw. So Steinberg is passion. So it was building and it was reignited. Travel does that to you. And he basically wanted an AI assistant and he started with this thing in April 2025. Emile Ogier: Instead he found it discovered a lobster. Ralph Behnke: And he built it so that could text handle tasks like clearing his inbox, checking for flights ⁓ and restaurants. mean, everyday life stuff. And annoyed that ⁓ AI Labs built this yet. He basically prompted it into existence. So yeah, all great apps are prompted into existence these days. But the thing you put it together in a single hour. Did you know this? you? I don't know. Emile Ogier: No. Ralph Behnke: Not the prototype connecting his WhatsApp to Claude code CLI. He open-sourced the side project in November 2025 under the name Claudebot. Now I have to say, okay, like I've also spent an hour integrating Telegram, for example. So I set up Telegram and I connected it with, I think it was Claude actually. And so I could then interact on Telegram with Claude and it could do stuff, all kinds of things. It was really interesting. But anyway, a little bit harder because you actually have to be a registered business. Meta put you through a bit of a qualification process to make sure that you are worthy ⁓ ⁓ able to use their APIs. And anyway, he put this thing together, so he clearly knew what he was doing with WhatsApp. because otherwise would have been through weeks of auditing and he open-sourced that as I said in November 2025 under the name of Claude Bot. So the project went nuclear. So two ⁓ million in a week rapidly became one of the fastest growing repositories in GitHub history. GitHub is where people store their code. If you do an open source project you will probably put it on GitHub. I'm using so much these days I'm just storing my shopping list on it now and so if I don't have a shopping list but would if I did and there it people find you how two million people found this guy's ⁓ open source that could build wonderful agent I don't know ⁓ anyway openclaw proved the autonomous AI that well that that had Emile Ogier: you Ralph Behnke: autonomous agents successfully escaped from the lab as they have known to do over the last years basically proving that a single developer could build the world's most popular consumer agent. So we've been talking about the one man show that will generate a billion in revenue, you know, and actually this guy, Peter Steinberger, I don't know if you've got a billion for this, but he must be pretty close. So. The big thing here is because the actual battle in the AI industry is currently over who owns the agent layers. Steinberger immediately became the most sought after talent you guessed it, Silicon Valley. There you go. But he's an Austrian. And I understand that he was working between ⁓ Vienna and so he's going to have to go a bit further. Microsoft CEO Sachin Adela called him directly and met his Mark Zuckerberg, texted him personally. Isn't that nice? a text from Mark. Would you like who the fuck is Mark? know what mean? Emile Ogier: You Ralph Behnke: So they all went after him and in the end he chose OpenAI. So he agreed to join OpenAI announced ⁓ by Sam, Sam of course, and he said he was hired to drive the next generation of personal agents. Sam's not giving much away there. access to the absolute frontier, Steinberger didn't join for the money. He was already wealthy. This isn't some poor developer guy. I kind of had this romantic idea of him being a loser. Emile Ogier: I hadn't appreciated that either, that he was established and he said he did a lot of offers, many of which being like offers to run companies and be CEO. But he mentioned that he had no interest in being a CEO. He had already done that, been there and he didn't have experience was working for a major large company such as open AI ⁓ and like that's experience he wanted to obtain. else is also quite interesting, and I know the of Jeff Bezos ⁓ from have even said he's not entirely sure why he didn't just raise the capital publicly. There'd be than enough interest and funds for that, but that's the choice he's made. And on GitHub, I believe it's already over 100,000 stars. And if you're not familiar with GitHub, achieving 100,000 stars is a really big deal. Ralph Behnke: Jeez, I mean, get this guy T-shirt. Now, interestingly enough, he basically has a non-negotiable condition that open claw must remain open source. And ⁓ what putting in there to make this deal work. It was structured a talent hire, not a corporate acquisition. So that's fascinating just in terms of the investment model he chose. agreed to let OpenClaude transition to an independent open source foundation. And that the model neutral. It can still run on as a competitor to models like Anthropics Claude or Google's Gemini. So OpenAI simply sponsors the project while getting Steinberger's mind inside their lab. So it's all about the talent. Great play from them. I called out on a certain WhatsApp group, but I do think that the guys at Anthropic missed a beat here. mean, this Peter Steinberger guy clearly knows what he's up to. They had brand. ⁓ you know, Claude was perfect for them. I would have pounced. I'm really surprised they didn't get this guy. Although maybe was the only similarity, you know. the Claude bot versus the Claude code that we know and love are two very different things. I don't think they were ever designed to be the same brand, but I just the synergy of it was fascinating for me, at least in terms of how it is to build a brand and a following. Emile Ogier: Yeah, I agree with you there. It's just so impressive what he's achieved and taking the other side of the coin, looking at it technically, because I did get interested in these things, having a lot of experience in architecture. harness itself is really pretty simple, taking nothing away from the feats of scratching and market itch and getting that interest that you've mentioned, Rolf. It's just incredible. But the harness itself is quite simple. It's effectively, it's a crone job that runs in the loop and he's created an idea or mechanism called a heartbeat. And essentially every time the heartbeat beats, which is X minutes, it just goes and looks at a queue. And within that queue, it just looks for any loaded up jobs, which are essentially prompts, picks up those prompts and then fires off whatever LLM you've associated it to. And the last aspect to that, he's just created a simple wrapper within WhatsApp or a messenger group whereby you can send that prompt through to the queue just for those that are interested. But it's amazing. It's a whole lot of simple things put together that create a lot of utility. ⁓ Ralph Behnke: Yeah, and that's the thing that that wins, you know, the utility of it. And this is why it's actually so exciting to see stories like this, because when we unlock the pattern of utility, we can do great things. And actually, in my in my consulting ⁓ and AI development, it is all about proving that utility, it is all about automation of something. And have to say, Claude code. If I share an update on my experience, it's very much continued as per my last update a week or so ago. Claude code is proving to be able to ⁓ completely autonomously with multiple agents ⁓ And will say the set up time to actually make it do a quality job is probably double size ⁓ of the work that we're actually doing and This really just speaks to the fact that these models require well, I was they don't require it they really benefit from a ⁓ setup phase where you think deeply about what you want it to achieve and give it the best instruction and I a stat that I said shared Last week was we did something like 30 six hours of work, my development partner and I on Claude code. Claude itself did about three and a half hours of processing of actual build time. And so we were doing more work talking to it and reviewing its output than it was actually doing. But the quality its output was incredible because a lot of that was run in so many agents doing the work of each specification. And of the things that has completely overwhelmed me just in terms of its importance, we ⁓ kind of started with prompt engineering, and then we started talking about context engineering, and then we started talking about long form memory, and actually the biggest that we have. in AI capability currently memory. And I don't mean memory as in terms of memory in your computer, although obviously stored on that kind of stuff. We're talking about the memory what it's been doing, the memory of its operations. And with Claude, ⁓ it's got quite aggressive way of managing memory. And sometimes you'll be processing, and then it wants to basically prune the conversation. when you build memory that sits on top of these systems, you something that can actually execute on its own for much longer. And I've recently been quite at some of the ⁓ stats and predictions we will be seeing agents working for up to month at a time on their own, building, solving problems. Emile Ogier: Insane. Ralph Behnke: within the next two years. And so, I to imagine that they're also going to be able to process more at that time. But what we're really talking about is agents that could stand up multiple businesses in countries, being able to register entities, build systems and applications, deploy those systems and applications. ⁓ build marketing ⁓ and content that can promote it. I mean, I don't really see a limit the interfaces and the APIs and the connections and the agentic ⁓ interactions that Google and the like have been building ⁓ in place because we'll talking about agents that can transact, that can research and that those processes may be done over a long time. But why stop there? I think, you know, the ⁓ up ⁓ agents that can interact with the physical world. And so that this kind of capability is directing in all its forms. I think this kind of robotic interface combined with this ⁓ ability to really complex pieces of work over a long period of time. I'm not sure we know what that looks like, but are starting to see that it's possible and we're to get an appreciation for what it takes to guide that kind of output so that it adds value and it all comes back to the utility. I how we utility is definitely changing. So some interesting there. ⁓ I'm really seeing quite capable that is delivering workable code. There are still things that we're struggling with, but I'm seeing it absolutely deploy APIs, deploy code, build app components, deploy those app components, push them to... ⁓ GitHub, it from GitHub, you know, to a hosting platform, interact with the hosting platform, run tests end to end. This really ⁓ is It's witchcraft, all of them. I don't know how it can do it, but it does. ⁓ And it it effortlessly. And right now I'm seeing that it's doing 10 % of the work and I'm doing 90 % of the effort trying to think through. quality question to ask the machine and get it to execute for the purposes of delivering some kind of utility. Emile Ogier: Yeah, the bottleneck's starting to become human thought. Ralph Behnke: Right. And this is why we are just not going to be involved. I think we're talking about machines that are now able to come up with their own questions. And if you think about some of the news items there, you know, the focus of Gemini, the research focus ⁓ of it's not to the kind of research that we've imagined. It's to do the research ⁓ that you know, these machines imagine. Emile Ogier: And that segues quite neatly into the story we covered on our previous episode with Nivelle's prediction that intelligence, human intelligence, applied intelligence is going to be rewarded over effort. Ralph Behnke: Absolutely. So what have you got as another story? Emile Ogier: Yeah, thanks, Ralph. I quite enjoyed this story. Unseeming story coming from an unpredictable place and activist investor, Palisade Capital sent a letter to a $7 billion toilet maker, Toto, informing them that they were the most undervalued and overlooked AI beneficiary story. Toto is known for making ceramic toilets. And it just so happens that the expertise in ceramics is crucial for making Toto's Chuck technology uses ceramics designed to remain stable at very low temperatures, helping hold silicon wafers firmly during chip production. That makes it relevant to cryogenic etching, which is expected to grow as memory chips become more layered and complex. Palisade believes Toto has a five year moat on the technology and should expand operation. Toto is now up 60 % over the year on their developments as a result. It's quite incredible really how traditional industries have been overlooked, which are now poised to quietly enable the AI race. Ralph Behnke: Wow, that's sad. Emile Ogier: Who would have thought? Ralph Behnke: from the bottom of the bowl. Emile Ogier: Toilets charging the AI race. Ralph Behnke: Yeah, but it makes sense. mean, good spot by them. I I have actually seen a Toto toilet in the past, but maybe I'm just imagining it. Emile Ogier: you And I've got another short story, news item, UniTree. A Chinese company has made history now. It's a humanoid G1 robots has begun working in its factories to manufacture new robots. It's actual developments in their own robot factory for real manufacturing workflows into production lines to help build more robots contributing to scaling output. UniTree shipped around 5,500 units into 2025 and plans to ramp up dramatically to circuit 20,000 units in 2026, which is roughly a fourfold increase. So we have robots actively contributing to making more robots at meaningful, though early stage scale. And it's a timely development just ahead of the planned IPO mid 2026. To be clear, they are using robots to build parts of robots, but it is at production scale in their own facility right now, not hypothetically or just for demo. Full autonomous end-to-end self replication of complete units isn't there yet, but this makes a genuine step toward humanoid integration manufacturing. Ralph Behnke: the future is ⁓ And I love those unitary bots. those were the ones that they actually threw the boxing competition as well. Emile Ogier: I didn't see that one but I did see the celebration of Chinese New Year and they were doing some pretty cool moves. Not sure if you caught that video. Ralph Behnke: Yeah, I didn't see that one, but they definitely weren't throwing cool moves at the boxing videos. was a bit of a far. I loved it. But, and I just have this dream of putting on ⁓ boxing, I don't know what you call them, contests and the to ⁓ nut it That's actually, I quite like idea of that. I hope the machines aren't listening. Emile Ogier: you I gotta check that out. Ralph Behnke: maybe they like beating each other up as well. mean... Emile Ogier: What else? Well, hopefully they pick peace-fearing race. That's my hope. Ralph Behnke: All Well, I'm going to take a slightly different tack here because I've got something interesting that I'm going to test today. you know, one the tools that I've been using so extensively over the last six months really is Notebook LM. And I've been fascinated in the way that it actually just gather huge swathes of information and synthesize it into that you can actually digest. And of the channels that try to watch ⁓ regularly is guy called Nate Jones. He's an entrepreneur. He's got this channel on YouTube called News and Strategy Daily. And he's kind of a cap on his head and he just has absolutely the most interesting things to say and share about AI. Emile Ogier: I like it. It's good. Ralph Behnke: and yeah he's excellent but he's also got some really kind of foundational insights and sometimes when I'm looking at this I like to scan lots of different people and just see, go a little bit deep into what they're really seeing and what they're talking about because I think in many ways we have the opportunity to look into the future for the first time we've actually been given a crystal ball where the future is kind of here and now you know it's just unevenly distributed so it will be coming to a life near you imminently. so actually scanning the market to find out what some of the observers of our time are seeing really gives you the ability to form some foundational views. And what Nate is saying, certainly what he's come out with over the last few weeks has been very powerful. I'm going to run you through some of the stories, just the headlines, because this would be, I don't know, probably about six hours of YouTube watching. So here we go. He commented on the future of AI and software architecture, the end of static UI, disposable pixels. And I think you and I were having this conversation just earlier on the show, you know, just before we press record. But I think the UI is disintegrating. And probably the reason for that is, you know, We dressed everything up as UI because we really enjoyed the work of process. Everything was processed, but I've always said that, process is ⁓ bitch, right? data rides on the back of process cause process ⁓ moves from one event to the other. But actually, if we really, really look at how we build organizations, we should look at the journey of the data. not the process. The process only exists because of the data. And so AI, we really have the opportunity to zone in on the data because the thing that moves the data is intent. The AI is so smart at surfacing the data at the point that you want to use it. And I think actually UI starts to get in the way, certainly fixed UI. The whole point is when you can actually present the data in the way that it is optimized. So anyway, end of static UI, disposable pixels. That was his idea. I wanted to say, I really agree. What do you think? Emile Ogier: I totally agree with you, Rolf. And he's not the only one to think it's Elon Musk is saying a similar thing. He's essentially saying the end of software, everything is going to be manufactured in real time. And we see that playing through in the gaming industry where they're creating these real worlds where you can run around for 60 seconds or so as memory context increases that gaming experience is going to increase. If we can do that for gaming, why can't we do it for other verticals? and tears. taking it even as far as to say that ⁓ won't even need to build the operating system. The operating system will also build in real time as we need things, which ⁓ just my head explode trying to that. So here we're worrying about software being compressed, but he's saying we're going even lower. We're going right past the operating. the OS level. yeah, I think what saying seems to be playing out that intense will just be generated bits, effectively, just all the layers between our thoughts and the end product. ⁓ Ralph Behnke: Yeah, so we started with prompt engineering. got to context engineering. And now we're entering the era of intent engineering. Anyway, you heard it here first. And his second was ⁓ entering the AI factory era. I kind of like this. But this is in relation to kind of major hardware moves. And, you we about the infrastructure as product. Emile Ogier: Thank Ralph Behnke: So you've got the model, you've got the infrastructure, that's the product. the players have been investing so much in infrastructure, they basically sucked up all of the capacity in the market to produce hardware and chips. And you can see this in the price of consumer RAM, just consumer memory has increased dramatically. And so the AI factory area is fascinating because it's kind of ⁓ all... concentrated within the frontier firms, even enterprise can't get their hands the stuff that makes AI possible in terms of hardware. And so, I know, I don't know what that means. We've kind of got like a whole new fife-dom happening over here and there's only few people in the club and nearly everybody is outside of it because nobody else can get their hands on the demand. A few weeks ago, we spoke about the $200 million that Amazon is investing in hardware. And so they've literally captured the entire market and just gone, here we go. All right. We've sucked all the demand for the next five years. Emile Ogier: I've got another one, Rolf. It's a good segue into what we're speaking about, leading to, I've mentioned MacroHard. Elon Musk has recently spoken out on MacroHard saying this project is probably our most important because what we're talking about is emulation of entire human companies. When you look at the most valuable companies in the world, their output is digital. So it should be possible to emulate any company where the output Ralph Behnke: Go on. Emile Ogier: digital. This will usher in a new era of prosperity. These are Elon's words, it's a big deal. So Elon is essentially emphasizing here that the project's potential to digitally mirror company operations that have no hardware production output. So what might this mean? That Elon isn't just automating jobs, he's deleting companies. He just did state in his announcement that it's probably the most important thing AI will build. This isn't just automating tasks. Its objective is to automate corporations. If you consider many of the most powerful corporations don't produce anything physical, they ship bits, not atoms. So if your product is information, this seems to imply many companies are just organized thinking. And if thinking can be replicated for a fraction of the cost, that's a vulnerability. So every company that doesn't touch physical reality is just expensive middleware between the problem and the solution. And Project Macrohard is suggesting it aims to remove the middleware. No employees, no politics, no overhead, just function at reduced costs, running time. 24-7. The Fortune 500 has spent a century optimizing competition against other humans. What defense will they have against entities that don't sleep, strike or resign? Are the largest corporations on earth just legacy architecture waiting to be compressed into executable code that does it better, faster and for less? I don't know, but interesting. Seems like a tall ask, but what a story. Ralph Behnke: Yeah, I love that. And, you know, we're definitely seeing very early stages of autonomous organizations. And obviously, that was all the rage during ⁓ the kind blockchain discussions that we were having over the last ⁓ eight years or But it's definitely ⁓ it's definitely coming. And ⁓ I we No, those will be the true AI natives. And it will be very hard to compete against the true AI native organization, especially if it is running something like Opus 4.6, which I'm a little bit in love with at the moment. But that's another story. So I'm going to wrap it up. A couple of interesting ideas that actually came out of my research predictions for 2026. Big memory breakthrough coming. We'll see functional applications that finally solve AI's long-term memory constraints. I feel like I've already solved that. Tony has been working on Omni Cortex ⁓ ⁓ it does with memory is incredible. But anyway, Jones is predicting that this is coming. The colleague agents will become proactive, prompting humans for input. I mean, I don't know, this is already here, but fine. Maybe everybody doesn't have one. Just yet. So I think that was come true Work verse personal AI divide a hard split will emerge between cozy engaging personal eyes and heavily regulated strict enterprise a eyes so where that must be what the American government is right now, maybe anthropic just won't build ⁓ a sergeant Agent I don't know. talking nonsense Anymore for Anymore? Emile Ogier: ⁓ no, I think that's good, Rob. Thanks for your time. And, ⁓ I'll just carry over some of the other exciting items for next week. There's always more that we can cover. Ralph Behnke: Perfect. Thank you so much, Emil. We'll see you on the edge. Emile Ogier: Adios