The AI Capabilities Overhang

Today on the AI Daily Brief, the AI capabilities overhang and what to do about it. Before that in the headlines, why cloud code is officially breaking into the mainstream? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors Zencoder, Landfall IP, robots and pencils and super intelligent. To get an ad-free version of the show, go to patreon.com/aidelybrief or you can subscribe it up on podcasts. Ad free is just $3 a month. I really tried to price it in a way where anyone who really doesn't want to hear those ads can get it for less than the price of a cup of coffee at this point. But if, on the other end of the spectrum, you are interested in sponsoring the show, send us a note at [email protected]. And lastly before we dive in another quick call to check out aidebintel.com, a little later this week, people who sign up are going to start finding out what this cool looking chart is. I promise you, you won't want to miss it, again that's aidebintel.com. With that out of the way, let's dive in. Welcome back to the AI Daily Brief headlines edition, all the daily AI news you need in around 5 minutes. The story of this January so far has been the slow but steady settling in of the notion of a shift in capabilities, centered upon opus 5, clawed code and now more recently clawed co-work. Turns out the change is not just among the highly enfranchised AI users on X, the Wall Street journal declared over the weekend that clawed is taking the AI world by storm and even non-nordes are blown away. The WSJ writes, "They call it getting clawed-pilled. It's the moment software engineers, executives and investors turn their work over to Anthropics Cloud AI and then witness a thinking machine of shocking capability, even in an age of wash and powerful AI tools." The article noted the huge wave of positivity on social media, with many non-technical people using clawed code to develop their first piece of software without knowing the first thing about coding. It also noted that clawed code is being deployed for a range of other use cases including health data analysis and expense report compiling. The Atlantic had a similar take, writing, "Move over chat GPT." The article says, "Though clawed code is technically an AI coding tool, hence its name, the bot can do all sorts of computer work, book theater tickets, process shopping returns, order door dash. People are using it to manage their personal finances and to grow plants." I don't know what it says about the Atlantic, that the first example they reached to is book theater tickets, but there you go. The author remarked that they used vibe coding tools for the first time in preparation for the article, and was astonished that they could create a new personal website in minutes without any coding. They went on to spin up a dozen additional projects over the next few days. They texted a friend to try it out and received the response. It just does stuff. Chat GPT is like if a mechanic just gave you advice about your car. Clawed code is like if the mechanic actually fixed it. To be honest, I don't really think that does it justice. I think it's more like clawed code, is like if when you dropped off your car, the mechanic you could request any other car, and all of a sudden a few minutes later it would just be there waiting for you. A user named Alex Lieberman was profiled for the piece, and claimed that in terms of implication this was even bigger than the Chat GPT moment. However he added Pandora's Box hasn't been open for the rest of the world yet. That might not be the case for long however with major publications now raving about Anthropics product lineup. Clawed code creator Boris Cherny remarked on the overnight success that was yours in the making, saying, "Glad to see Clawed code starting to break through. It's been a year of very hard work and we're just getting started." Ashni Mido writes, "The front page of the Wall Street Journal today is about every day people using a command line interface. If you are a business leader and not revisiting major operating assumptions about the world, you are doing yourself in the people who depend on you a massive disservice." One other Anthropic story the latest on their reported fundraising. That round that we've been hearing about that values Anthropic at 350 billion is apparently getting supersized up to potentially 25 billion. That includes about 15 from Microsoft and Nvidia and another 10 from VCs and other investors. Among those VCs is apparently Sequoia. Feels like we're probably within a few weeks of this closing so I'm sure we'll get more news soon. Now, despite the Wall Street Journal writing so glowingly about Clawed, US users clearly remain concerned about the technology overall. According to a new survey commissioned by Google and conducted by Ipsos, AI users are now in the majority. 66% of respondents said they had used AI in the past 12 months compared to 48% in the 2024 poll and 28% in 2023. This was the third year of the longitudinal survey which was conducted in late September last year so relatively up to date. The survey pulled around 1,000 adults from each of 21 countries. Respondents were evenly split when it comes to AI job disruption, with 50% saying AI in the workplace will create jobs and 50% saying it will eliminate jobs. Still the majority of survey participants were in favor of fostering advancements using AI at 58%, compared to 41% who wanted to protect industries that might be disrupted by AI. Not surprisingly, AI optimism was closely tied to AI use. 70% who said they've used AI are optimistic about its benefits. And of those who use AI a lot, 86% were excited. There's also a strong correlation between countries with high levels of AI use and high levels of optimism. The US however ranked low on both use and optimism. Just 40% of US survey participants said they have used AI in the past year, which was the only country without a majority of AI users. As a point of comparison, the UK was at 56%, Mexico was at 66%, while the UAE, Nigeria and India were all north of 80%, only 33% of US respondents said they were mostly excited about the technology, the worst national result in the survey, and vastly beneath the overall result of 57%. Now, I actually think this is a major national issue. The implications are not just the user numbers for open AI and anthropic, it's about the seriousness with which people are taking the potential disruption for this technology and preparing themselves for it. I believe there continues to be a strand of people who are hoping to just wait it out and return to the world at once was, and obviously I do not think that that's going to happen. Moving over to an issue that has become part of the political cannon fodder around AI, which is of course data centers, XAI's Colossus 2 has now reached one gigawatt of capacity, becoming the first training cluster to cross that threshold. The data center is now drawing more power than the city of San Francisco, but comparison the first Colossus cluster has a total capacity of 300 megawatts, while open AI recently disclosed that they have 1.9 gigawatts across their entire training and inference fleet. Construction began in March of last year, so this milestone was 9 months in the making. The only other cluster that's closest anthropic in Amazon's new Carlisle data center, which is expected to hit one gigawatt sometime in the first quarter of this year. Open AI's Stargate Abilene is expected to come online over the summer. For now, XAI is the only company with access to this much compute, which is exactly what we discussed as the big potential opportunity that could translate into differentiation for GROC in the year to come. Colossus 2 is also using Blackwell GPUs, making it one of the first training clusters to run Nvidia's latest hardware and the only one at this scale. The cluster reportedly contains 550,000 GPUs as currently configured. As Amateo Kaplan put it, "Gigawatt GROC" has arrived. Now staying in Musk world for a moment, Elon Musk is seeking up to 134 billion in damages from open AI in Microsoft as his lawsuit heads to a trial. A trial date has been set for late April, and during a hearing on Friday, Musk's lawyers quantify the damages. Their argument is that Elon is entitled to a portion of open AI's current $500 billion evaluation due to the $38 million in seed funding he donated to the nonprofit in 2015. Musk's lawyer wrote in court filings, "Just as an early investor in a startup company may realize gains many orders of magnitude greater than the investor's initial investment, the wrongful gains that open AI in Microsoft have earned, and which Mr. Musk is now entitled to discourage are much larger than Mr. Musk's initial contributions." Which is a very legal ease way of saying, "If that $38 million had been an investment into a four-profit startup, it would have been a heck of a lot more than $38 million by now." The filing also says that Musk plans to seek punitive damages as well as an unspecified injunction. Open AI's lawyers rejected the approach, stating that his "methodology is made up, his results unverifiable, his approach admittedly unprecedented and his proposed outcome, the transfer of billions of dollars from a nonprofit corporation to a donor-turned-competitor, implausible on its face." Open AI for their part continue to deny the premise of the lawsuit outside of the courtroom. In a statement they said, "Mr. Musk's lawsuit continues to be baseless and a part of his ongoing pattern of harassment, and we look forward to demonstrating this at trial." This latest, unserious demand is aimed solely at furthering this harassment campaign. Now on X, the discussion centered around pages from Greg Brockman's private notes that were revealed in the new filing. One especially frequently shared passage from 2017 red, "This is the only chance we have to get out from Elon." You see, the glorious leader that I would pick, we truly have a chance to make this happen. Financially, what will take me to $1 billion? TV DOS at Menlo Venture said, "Deep down, it really is about the money." Now several other quotes from the filing paint open AI in a very poor light. The quotes. Altman said, "Eelon is cherry-picking things to make Greg look bad, but the full story is that Elon was pushing for a new structure, and Greg and Ilya spent a lot of time trying to figure out if they could meet his demands." Altman continued, "I remembered a lot of this, but here is a part I had forgotten." Elon said he wanted to accumulate $80 billion for a self-sustaining city on Mars, and that he needed and deserved majority equity. He said that he needed full control since he'd been burned by not having it in the past, and when we discuss succession, he surprised us by talking about his children controlling AGI. Altman continues after that quote, "I appreciate people saying what they want, and thinking it enables people to resolve things or not, but Elon saying he wants the above is important context for Greg trying to figure out what he wants." With the trial less than three months away, the story is unfortunately going to be a big overhang for open AI as they try to execute on a pivotal year. Despite signal, this is going to make a lot of people look greedy and ugly. Hopefully, we won't have to spend too much time on this. I'll probably start to err on only sharing the really big highlights, where it becomes a major, inescapable point of conversation as it has been for the last couple of days. This show, however, will not become a play-by-play court drama, as interesting and salacious as it might be. For now, that is going to do it for today's headlines. Next up, the main episode. If you're using AI to code, ask yourself, are you building software or are you just playing prompt roulette? We know that unstructured prompting works at first, but eventually it leads to AI-slop and technical debt. Enter ZenFlow. ZenFlow takes you from vibe coding to AI first engineering. It's the first AI orchestration layer that brings discipline to the chaos. It transforms freeform prompting into spec-driven workflows and multi-agent verification, where agents actually cross-check each other to prevent drift. You can even command a fleet of parallel agents to implement features in FixBug simultaneously. We've seen teams accelerate delivery to X to X. Stop gambling with prompts. Start orchestrating your AI. Turn raw speed into reliable production-grade output at ZenFlow.free. If you're listening to this, you already know how fast AI is writing the rules for innovation, disruption, and value creation. This new era demands a new kind of patent law firm. Landfall IP was built from the ground up to operate differently, orchestrating how human expertise and AI work together for better patents at founder speed. Created by world-class patent attorneys who saw a better way, Landfall IP lets AI execute the repeatable while attorneys elevate to create the exceptional. Landfall isn't adapting to AI, they were built for it. Have a new idea? Try the Discovery agent for free. It's a confidential tool that helps innovators synthesize their inventions and instantly see patentable insight. Visit landfallip.com to learn more, that's landfallip.com. Today's episode is brought to you by robots and pencils, a company that is growing fast. Their work as a high-growth AWS and Databricks partner means that they're looking for elite talent ready to create real impact at velocity. Their teams are made up of AI native engineers, strategists, and designers who love solving hard problems and pushing how AI shows up in real products. They move quickly using RoboWorks, their agentic acceleration platform, so teams can deliver meaningful outcomes in weeks, not months. They don't build big teams, they build high-impact nimble ones. The people there are Wicked Smart with patents, published research, and work that's helped shape entire categories. They work in velocity pods and studios that stay focused and move with intent. If you're ready for career defining work with peers who challenge you and have your back, robots and pencils is the place. Explore open roles at robotsandpensals.com/careers. That's robotsandpensals.com/careers. Today's episode is brought to you by Super Intelligent. Super Intelligent is a platform that very simply put is all about helping your company figure out how to use AI better. We deploy voice agents to interview people across your company, combine that with proprietary intelligence about what's working for other companies, and give you a set of recommendations around use cases, change management initiatives, that add up to an AI roadmap that can help you get value out of AI for your company. Now we want to empower the folks inside your team who are responsible for that transformation with an even more direct platform. Our forthcoming AI strategy Compass tool is ready to start to be tested. This is a power tool for anyone who is responsible for AI adoption or AI transformation inside their companies. It's going to allow you to do a lot of the things that we do at Super Intelligent, but in a much more automated, self-managed way, and with a totally different cost structure. If you're interested in checking it out, go to AIdailybrief.ai/compass, fill out the platform, and we will be in touch soon. Welcome back to the AI Daily Brief. Today we are talking about something called the AI capabilities overhang. Now this is something I think about a lot, but the specific context for it was an article that came out as part of the broader set of assets around OpenAI's announcement that ads are coming to chat GPT, with them basically saying that part of the issue is access and ads are going to help them with that access issue. Now in that blog post called AI for self-empowerment, OpenAI defines the capability overhang as the gap between what AI systems can do now, and the value most people businesses and countries are actually capturing from them at scale. In other words, the delta between AI's current capabilities and society's current usage of them. And what's important about this concept is this is not about some future state. This is not, in other words, a debate about AGI or Super Intelligence or anything like that, it is instead a discussion of the current state of play and how far behind different types of people and groups are in taking advantage of it. So what I want to do today is talk about the AI capabilities overhang across six different groups, individuals, communities, municipalities, educators, businesses, and sovereigns. For each of those groups I want to talk a little bit about what the capabilities overhang looks like at the moment, what some of the answers to that overhang might be, and how we, and this is the Royal We, I could mean society, I could mean the listeners of this podcast, but how we could support tackling that capabilities overhang and improving the way that people are taking advantage of what's possible right now. So let's talk first about individuals. Now, this is admittedly a wildly all-encompassing category with a huge range of different levels of this particular overhang. While there are very, very few people who could claim that they don't experience that overhang at all, in fact, even as someone who spends basically all of my time on this, I think that there are entire categories of what's possible that I don't take nearly full enough advantage of. Most people fall somewhere on the spectrum from barely taking advantage to only just starting to take advantage. In fact, I think part of the reason that you're seeing so much excitement around Claude code and see it moving into the mainstream in the Wall Street Journal and things like that is that for people who are picking it up, it is radically and directly undercutting that capabilities overhang, by massively accelerating what people can do. But the implications of the capabilities overhang is dramatic. Skills that took years to develop can now be augmented or replicated in hours. Now this sort of commoditization of knowledge work creates displacement risk, of course, but it also creates incredible opportunity in terms of the massive leverage that it can give people. One of the implications of the capabilities overhang, however, when it comes to that individual focus is that personal economic modes are eroding faster than people realize. In other words, the gap between I should learn this AI stuff and I needed it yesterday is closing. To what are some of the challenges? Information is one, and by that I don't just mean information about what's possible with AI, although that's part of it, but also I think we have a real issue in the way that we discuss AI. Every survey that comes out shows that to be a little bit reductionist, but honestly not all that much, Eastern and lower income countries are extremely enthusiastic about AI, while people in Western and higher income countries are less so. There are all sorts of reasons for this, but what it means is that in addition to just a general information gap, you also have a massive enthusiasm gap, which means that people who don't like AI or wish it didn't exist are getting farther and farther behind, kind of hoping that it just goes away. Go on any social media platform and you will be able to find myriad posts about people enthusiastically "waiting for the end of the bubble so things can go back to normal." Despite the incontrovertible fact, known by everyone who has listened to this particular show, that there is no such thing as going back. So improving information availability about what you can do with AI, but also about to some extent the inescapability of some changes because of it is a key part of overcoming this overhang. Another part is of course access. People who can pay more right now have better access to AI. However, the gap isn't necessarily as big as it seems. Although chat GBT data shows that the typical power user of their system uses seven times more compute than a typical user, there's still incredible capacity available to anyone, even in the free versions of these tools. One of the things that's been super interesting to me watching people interact with the New Years AI resolution, which is the 10 week self-education program that came out of my New Years episode, is that a lot of folks, despite being on the very high end of enfranchised users, are seeing how much they can get out of the free versions of these tools. I actually think that that's incredibly valuable, and more instructive in many ways to the average user, than some insane person like me who's going to pay for the ultra or max subscription to every single tool that comes along. I will say that I think even with this, a quality of access is going to continue to be an issue, and probably one that gets worse. As much as we might not like the experience of ads in something like chat GBT, I do believe that it does extend access and keeps access democratized in a way that a non-ad supported model just couldn't. However, ads from the platforms are certainly not the only way to ensure access. There might be a role for government here, and frankly, it's one of the reasons that some of the ideas around things like stopping data center construction are so wrong headed, because they're likely to have the exact opposite impact where they actually further restrict access to only the people who can pay for it. So how can we support individuals overcoming the capability overhang? One is a different conversation about AI and an acknowledgement that it's coming. Two is continuing to look for ways to democratize access, whether that's from the platforms themselves through ads or other models, or through public-private partnerships or some other larger type of initiative, anything the last piece, and something that certainly we will be trying to do a lot of this year with this show is self-education opportunities. So many people are engaging so deeply with this new year's AI resolution that I am 100% sure that we will release other similar time-bound, but ultimately self-directed types of programs. Next up, let's talk about communities. Communities hold many of the assets that AI can't replicate. Trust networks, local context, physical gatherings, shared identity, accountability structures, and the overhang actually increases the value of these assets. As digital interactions become AI-mediated, and in many cases for people harder to trust, in-person community becomes a premium good. In fact, as we think about the individual overhang, local institutions are sitting on distribution and trust infrastructure that could be leveraged to help members navigate the AI transition, but of course most aren't thinking this way yet. Indeed, the challenge for communities is that community institutions tend to be the most strapped for hard resources, and so the people who are involved in leading communities have to trade time for everything. Basically, when there isn't money, you require people to volunteer time and service. That means less time for keeping up with all of these opportunities. But if we can position community institutions as the human layer in an increasingly AI-mediated world, there is a ton of potential power in these institutions taking on renewed importance in this new age we're moving into. To support that, I think we have to start by supporting their leaders. We need dedicated resources and leadership support and training for these particular types of institutions that are not necessarily just about the same things that are going on with individuals, but are really about how to become a node for disseminating and supporting transition among constituencies. Closely related to communities is the capabilities overhang for municipalities. Municipalities are, of course, the public and governmental complement too many of those community institutions. Like communities they're strapped for resources, they have old patterns of doing things that can be very difficult to change. Are these groups potentially some of the biggest beneficiaries of the efficiency gains that come with AI? One study found that 30-50% of municipal staff time is spent on tasks that are already automatable or dramatically acceleratable right now. And it takes about five seconds to think about some of the examples, changing review time for permitting in land use, moving from hold times, phone trees and manual routing within constituent services to instant intake automated routing and proactive follow-up. There are potential implications for public works for social services, for records, for courts, for revenue and finance, for public health, you name it. AI efficiencies are so full of opportunity among municipalities. And for that reason, I think we should be spending way more time energy and resources on trying to fix this particular category of capabilities overhang. What does that look like? Well, of course I don't know, but I think that some opportunities include different types of public-private partnerships. I gotta say, right now the model labs in many AI startups in general don't necessarily have the best brand and reputation. The AI industry as a whole is suffering from a broader sense among many people, especially in the West, that techno longer exists to serve people to the extent it ever did, but instead serves only to enrich the people who create that technology. Seems like a pretty good time to try to engage in some public-private partnerships that actually bring the benefits of AI to a wider audience. Frankly, this also strikes me as an opportunity for perhaps a different class of business that has different incentives. I think there are very clear, profit-motivated business opportunities in AIifying how municipalities work. But I think that trying to find a new generation of entrepreneurs, maybe some of whom have backgrounds in that type of municipal service, who are designing leaner, more capital-efficient providers, who are going to be able to offer municipalities, contracts and services that provide that transitional support without gouging them. It just strikes me that this could be a great moment for a new class of civic-minded entrepreneur to really do some damage in the best possible way. When it comes to educators and education, goodness gracious, where to even begin. This is something we've talked about a lot in the show, although not for a little while, believe it or not. But by and large, education is stuck being concerned that students can now cheat on the test, when the real problem is that in the future that we're moving into, the test doesn't matter. We need nothing short of a radical re-evaluation of everything that we teach. To be wildly oversimplified and reductive in a way that the educators among you are going to be cringing your faces off by apologize in advance, let's start by separating everything into three buckets, the skills that are definitely still relevant, which, by the way, there are many. Critical thinking, ethical judgment, creative problem solving, human interaction and empathy, relevant, relevant, relevant, relevant, in fact, more so. Also interestingly enough, a set of skills which we've often pejoratively called soft skills that we haven't had nearly enough emphasis on in our education system for a very long time. So we've got that one bucket of definitely still relevant. Then we have the things that are definitely changing in relevance, subjects that are absolutely and undeniably being transformed by AI tools, writing in composition, research and information synthesis, programming. We don't have to fully throw out the baby with the bath water to recognize that we're talking about a lot more than going from adding with an advocate to adding with a calculator when it comes to how dramatically this set of skills is changing in terms of how humans are going to interact with them. And then of course there's perhaps the biggest category, which we will generously call who the hell knows there is going to be so much in this category where we simply do not know how AI is going to impact it and having the humility to understand that some stuff that we're going to teach could be irrelevant. But we just don't know and we have to hedge a little bit is I think a reasonable way to proceed. Then of course there is a fourth bucket new things that have become relevant. Some of that's just AI specific skills, but a lot of that's going to be in and around management and organization and basically the things that help people take advantage of the fact that each of them will in the future have access to talent that every corporation in the world would kill for today. And then from there we redesign the curriculum around a balance of these things we run big experiments we get OK with failure this area of disruption and change is going to be some of the easiest to talk about and hardest to actually do in practice. And the best way I think we can support is to create space for real change not incrementalism but true actual disruption. The second last category we'll talk about today is the capabilities overhang for businesses. And once again, just as the individuals are incredibly diverse so to is the capability overhang for companies as we saw with our AI ROI survey. There is a spectrum of capabilities overhang for companies at every different level. And while it may be the case that certain sizes of companies have different types of advantages over one another. I don't believe that on mass there is one category or type or size of business that is experiencing dramatically less overhang than some other companies are pretty much all dealing with going from no AI to figuring out how to use AI for efficiency. Or they are dealing with the challenge of moving from efficiency to actually leveraging AI for new opportunity in years of doing this and seeing thousands of thousands of executive interviews. I will say confidently that I have never seen a single company of any size including my own startup and incentivized enough to get this right that doesn't experience the AI capabilities overhang in some way. We just had a super intelligent off site where all of us sat down and basically tried to tear through everything we do and ask how we could AIify it even more. And the amount that we are not doing is immense. The big problems of the capabilities overhang with businesses involve really common patterns. The thing we hear about most at super intelligent is creating time to redesign. We have these whole new set of skills that people are expected to learn, but they're expected to learn them while also doing their normal jobs. Classic and quintessential conundrum of the moment is that we don't have time to learn the thing that could save us so much time. There's also the challenge of our normal disposition to wait for the future rather than to go invent it, which of course gets into the idea of new opportunities. We don't know what it's going to look like when every single member of every single company can use software to deliver on their KPIs and invent new ones, but we're going to start to find out. Certainly something that we could help with is that especially the farther away from the AI efficiency arrow we get, the worse and worse the resources to support people's education we find. There are still plenty of prompt engineering courses out there, but actual really strong resources on how to use coding tools for non-coders, how to build and manage agents, how to think about automations more systematically. These are much fewer and farther between. Now hopefully the market incentive for this changes that in short order because goodness gracious is there a lot of market opportunity there, but anything we can do to provide more resources for education or self-education the better. The last group we'll talk about are the sovereigns. And honestly this is the group that might be the most aware of their capabilities overhang of anyone. The overhang in this case is a national security issue. The delta between what's possible and what's deployed represents strategic vulnerability. It also represents a challenge in terms of who gets to define the future. And not only is there this strategic vulnerability, but sovereigns are also dealing with the fact of what it means to have everyone's understanding of the world mediated by LLMs that are reliant on a specific set of sources, which may not take into account the full complexity and cultural legacy of your particular nation. So there's so much in the AI capabilities overhang for sovereigns. Certainly in the medium term, first mover advantages in AI capability could create some seriously durable geopolitical asymmetries. Now like I said, this is the group that I think is most cognizant of this challenge and it's why you see so many nations treating AI infrastructure in the form of compute, talent and data as critical national assets. It's why you're seeing massive geopolitical realignment around this stuff and it's going to be fascinating to see how this continues to interact with the geopolitical conversation and the years to come. Anyways friends, that is our quick tour of the AI capabilities overhang the gap between what AI can do right now and how society in all of its various manifestations is actually taking advantage of it. I think we all in many ways have incentives to try to close the capabilities overhang among every type of group not only that we're participating in, but those around us. Hopefully identifying it as a challenge is a good starting point and that's what I've tried to do here and will continue to do on this show. For now that that's going to do it for today's AI Daily Brief, appreciate you listening or watching as always and until next time, peace. (upbeat music)