Transcription
So, opening, I have explained why 2025 is probably going to surprise the average person. Now, probably won't be surprised if you're watching this video and you're in the AI space. But the recent blog post from Sam Altman is quite telling, and it actually shows us a few things that I didn't even know at the moment. But that's what this video is for: a deep dive into all things AI. So let's not waste any more time.
So, one of the crazy things from Sam Altman's blog was actually this right here. Now, there's like three statements on this page that I do want to break down in more detail, but honestly, the one highlighted is the one I want to focus on. And what he says here is that "we believe that in 2025 we may see the first AI agents join the workforce," and "join the workforce" is in quotation marks. I believe that's because when you say "join the workforce," it just means that they're going to be, you know, working alongside you and, you know, materially change the output of companies.
Now, this is something that is a remarkable statement because this is, I guess you could say, one of the first times we're getting true, you know, digital employees that are going to be able to do a lot of tasks and a lot of the, you know, tedious tasks that are involved in many various businesses. And, you know, once you free up your time in business, you're going to be able to use that time to be remarkably more effective. So this is going to be a really interesting thing because I've actually paid attention to Sam Altman's blog post for the last three years, and he's actually quite right about a lot of things. And I do believe that's become pretty obvious considering he's a CEO of a leading AI company, and it is their job to sort of, you know, predict the future, see where things are going. So seeing the trajectory of AI the past few years, it's going to be really interesting to see how the end of 2025 is compared to the kind of software that we do have now.
Now, the crazy thing about this as well, and I will come back to this page, is the fact that like most people also don't understand that there are certain AI agents that have already joined the workforce. Like, trust me, guys, there is a silent revolution going on at a large majority of companies that you might not be aware of. "Joining the workforce," yes, that is a statement, but you might want to pay attention to the companies that have already deployed AI agents in their businesses. This is the CEO of Klana who actually spoke about how they managed to deploy customer support agents that actually fill the role of 700 employees. A lot of business leaders are asking themselves, like, "Where is the actual implication on my business?" And this was the first time we at least felt like, "Look, here's a practical application that isn't like just a small—it's real life." Consumers prefer it because it's actually higher customer satisfaction than the human agent; it's faster to use; it's higher quality; and it's having a profound impact. But we also believe that society and politicians need to recognize that this is not something that's going to change in 10 years; the implications for society are going to come in the coming years, and it's time to think a little bit proactively about what measures could be taken as a consequence. And I think his statement is quite true: most people don't realize that, you know, agents are currently being deployed. There's a variety of different agents that you can build. One of the agents that, of course, he talks about is customer support agents, and I think that kind of agent is really important because not only is it something that is quite tedious—customer support is, you know, one of those things where usually the customer has a complaint, so they're already on edge, they're already, you know, angry and charged up—and I've heard the conversations sometimes that people have, and it can be pretty awful to actually be a customer support staff. So I think you guys need to actually understand that this is something that is, you know, on a smaller scale, slowly happening across a variety of different companies.
And I'm going to show you guys quickly just a few seconds of ElevenLabs' new AI conversational agent, which is actually really cool. Building conversational AI agents used to be immensely complex and demand huge resources, but not anymore. "Start building your agent by creating a voice or exploring our vast high-quality library to discover one that's perfect for your application." "Hello and welcome to DHK online. How can I help you today?" "With a wide range of the best LLMs to choose from, you can then upload your knowledge base and define the goal and personality of your agent, or alternatively you can integrate your own server to take full control." "Hi, this is Eric from Bound Events. How can I help?" "Hey, I accidentally bought a ticket twice. Could I get a refund, please?" "Sure, please read out your reference number and I can begin to process that right now. Our system will analyze and evaluate your transcripts, providing you with valuable insights, conversation playback, and the ability to test and define success criteria, and it's all possible in 32 different languages." "Welcome to DC Medical. How can I help?" "Hello."
The actual ability for these AI agents to have fluent conversations in a variety of different languages—imagine you're in another country and you call up their customer support hotline and you can't speak the language—but considering it's an AI agent, it's able to instantly translate fluently to whatever language you might be speaking. This is something that I think is a remarkable achievement and is something that is actually a really good feature of AI agents.
Now, me personally, when it comes to AI agents, I personally like workflow agents. So this is Make.com, which allows you to collect things like ChatGPT agents to ElevenLabs agents and connect this all up to create videos. This is something that I actually created for my community. I also created this lead generation agent, so that if you're trying to get leads for your business, you have one that basically uses these workflow agents, and then when that data all gets processed, Claude, Haiku, quickly just analyzes the data and then puts all the business details in a Google table or a Microsoft Excel file that's like this. And if you ever want access to any of the agents that I show you guys, you can just import the blueprints. These AI agents are actually available in my community, just in the AI agent template section. And every Wednesday, I actually show a very unique way to use these AI agents to make money. Every Friday, I actually break down in detail how you guys can actually use new AI tools that nobody is actually using. And every Sunday, I actually do do a case study in my Academy to show you guys exactly what's working within the AI industry. If that's something that you are interested in, don't forget to check out my AI Academy Community, which I recently launched.
Now, another thing that they also state is that they continue to believe that "iteratively putting great tools—tools—in the hands of people leads to broadly distributed outcomes," and this is rather important because they actually hint at how they're going to be deploying these tools in 2025. So whilst yes, 2025 might be a blockbuster year for AI, the word "iteratively putting these tools in their hands" actually means that things are going to be released in an incremental basis. And I think if you guys can reference the video I made two or three days ago where, you know, they talk about how the AI is going to take off, one of the things that we wonder is, is it going to be a discontinuous takeoff where we have rapid jumps and capabilities? But right now, it seems like the way that these companies are deploying things, it's going to be a very smooth, slow, continuous takeoff for everyone.
Now, additionally, as I was talking about agents, I couldn't help but notice something from Logan Kilpatrick, and this is the guy that is the lead product for Google AI Studio and is working on the Gemini API and AGI at Google. Now we can see here that he says 2025 is actually the year AI vision capabilities are going mainstream, but that 2026 will be the year of agents. So the reason that this is interesting is because some people have said 2025, and some people are saying 2026, and this is why I've titled this "Are Their Heads," because when we actually take a look at things, we can see that OpenAI is setting 2025 for the year of agents, but other companies like Google are saying 2026 will be agents. What I think they actually mean by this is that 2026 is going to be when we have fully autonomous, reliable agents, but things like browser agents are still going to be something that occurs. Now, he actually expands on this a little bit, and it's actually really important you pay attention here because he said, "There's a 12-month capabilities-to-wild-scale production gap. Most vision use-case work now but aren't widely deployed," and I think that's really true. But agents still need a little more work for billion-user-level skill. So essentially here, he's actually referring to the reliability of these agents because, of course, a billion-user-level skill is a large amount of people—a ridiculously large amount of people—and you can't have an AI agent that fails 10% of the time because that is over 100 million people that are going to be experiencing some kind of sloppy service.
Now, the reason I say that this is rather interesting is that it does potentially mean that maybe OpenAI are a little bit ahead because if Google are saying that, look, AI agents—fully autonomous AI agents—are going to be in 2026, and then also their competitors, Anthropic—the CEO of the company recently said this in an interview—and pay attention because he actually talks about agents in a little bit more detail: "If you want an agent to act in the world, um, usually that acting requires you to, you know, engage in a series of actions, right? You talk to a chatbot; it only answers, and maybe there's a little follow-up, but with agents you might need to take a bunch of actions, see what happens in the world or with a human, and then take more actions. And so you need to do a long sequence of things, and for that long sequence of things to actually work, the error rate on each of the individual things has to be pretty low, right? If I'm a robot and I'm like, you know, okay, I'm going to pick up this thing and walk over there and I'm going to pick up that, you know, I'm building the house or something, there's probably thousands of actions that go into that. And so all of this is to say the models need to get more reliable because the individual steps need to have very low error rates, and I think part of that will come from scale, um, like we need another generation or two of scale before the agents will really work." And so this is actually really true because when we actually look at the benchmarks from Claude's recent release—and currently Claude is the frontier model in terms of—I don't want to say raw intelligence because that would be 01 or 03—but when we actually look at, you know, these large language models that are able to use a computer, this is the TOW benchmark that tries to, you know, put these AI systems in realistic scenarios and have them, you know, do things agentically, we can see that currently these models aren't performing that well, and this reliability is something that they need to iron out. We can see that the past metrics here, like 46% for the first try, and as they try more and more times, these results just get worse and worse. Now, of course, there are a variety of different AI agents out there, you know, other companies that are going to be introducing new frameworks, and of course, maybe by the end of the year this is going to be something that is close to 90%, but recently with these current models that are actually really good when it comes to agents and doing things agentically, they just don't really go that well.
Now, I do wonder if that is going to be because, you know, OpenAI is shifting their strategy as the rate of GPT AI improvement is slowing, and that is something that, you know, they've alluded to. And when I say alluded to, I'm talking about the fact that, you know, the GPT series is slowing down, just meaning that, you know, any gains they make now will simply be incremental. That doesn't mean 01, 02, 03, whatever series of models they call the 01 series, the reasoning series isn't going to be going on a steep, steep incline in terms of the amount of intelligence that model can produce, but I'm talking about the GPT series, like Claude 3.5 and, of course, GPT-4. Those kind of models won't likely have jumps in capabilities, but they'll need to be really good if we're going to use them as the basis of these agents.
Now, interestingly enough, the corporate vice president, business and industry, co-pilot, Microsoft, said that "by this time next year you'll have a full team of AI agents working for you," and this could look like anything from "an IT agent fixing glitches before you even noticed them or a supply chain agent preventing disruptions while you sleep, a sales agent breaking down the silos between business systems, the chase leads, and finance agents closing the books faster." And I actually think that that is really true because if you've been paying attention to any talks recently, you'll—Sam Altman, you'll hear Jen-Hsun Huang actually talk about how agents are going to be a really big thing. Agents are the thing everyone is talking about, I think for good reason. You know, this idea that you can give an AI system a pretty complicated task—like a kind of task you'd give to a very smart human that takes a while to go off and do and use a bunch of tools and create something of value—um, that's the kind of thing I'd expect next year, and that's like a huge deal. We talk about that like, oh, you know, this thing is going to happen, but that's like—if that works as well as we hope it does, that can—that can really transform things. And of course, if we take recently what Jen-Hsun Huang spoke about in terms of agents at CES 2025, he actually gave us an insight to how these AI agents are going to all be working together to, you know, I guess make a business more efficient and streamline operations.
Now, I got to be honest with you guys, the craziest thing from this document was, of course, the fact that they said—even though, you know, they speak about AI, and Altman has spoken about AGI a variety of different times—he says that "we are beginning to turn our aim beyond that to superintelligence in the true sense of the word," and I think that is one of the most profound statements because it's becoming more and more commonplace for OpenAI to say this, and the more and more they say it, the more and more I believe them because I've read the research papers, and it kind of does make sense. And they talk about how, you know, with superintelligence, "we can do literally anything else. Super intelligent tools could massively accelerate scientific discovery and innovation well beyond what we are capable of doing on our own and, in turn, massively increase abundance and prosperity." And you can see right here that they do acknowledge that this claim of, you know, superintelligence or whatever sounds like science fiction and even sounds, you know, somewhat crazy to talk about, but they've been there before, and they're okay with being there again, but they're pretty confident that in the next few years everyone will see what they see, and that "the need to act with great care while still maximizing the broad benefit and empowerment is so important."
So this is really interesting because, of course, they aren't just stating that, look, this is a wild claim; they're basically saying that, look, we know it; the only reason you guys don't know it is because you don't work at the company. So we're pretty sure in a few years you'll eventually catch on to what we already see. And it's going to be super interesting as well to see how OpenAI's products manage to change and integrate with their levels—SL stages of AI. Currently, we're at reasoners and human-level problem-solving, but level three is, of course, the big one where these systems are going to be able to take actions reliably. I mean, level four and level five, that does seem like crazy talk, but at the end of the day, I've seen crazier things in AI that I never thought possible. And of course, as you know, OpenAI are actually launching an AI agent very, very shortly, so it's going to be super interesting to see when this agent arrives, how it manages to perform, and the other things it's able to do.