2025 AI : 10 Things Coming In 2025 (A.I In 2025 Major Predictions)

In this video, I've scoured through everything that I could personally find that I know is going to be there in 2025. So let's not waste any time and jump into exactly what is going to happen in AI. So one of the first things that is going to happen is, of course, agents. Now, with agents, one of the biggest companies in the world in terms of AI is OpenAI, and they are going to launch an AI agent in 2025 called an operator agent. And this is going to be an agent that can use a computer to take actions on a person's behalf, such as writing code, booking, travel. According to people familiar with the matter, this is something that was leaked via the information, which is a very reputable source in AI. So it's quite likely that we will get this in 2025. So now that we know OpenAI is going to produce this agent, we do know that this is going to be something that is probably of a high quality. Many other companies have already rushed out their AI agents, but OpenAI is likely waiting on something that will blow us away. Now, if you're wondering on a timeline for this AI agent, I did make a video when this information was released, but the cliffnotes version of that is that in a staff meeting on Wednesday when it happened, OpenAI basically announced that this is going to happen in January. Now, with the AI industry, timelines do differ due to a variety of different factors, but essentially we could be experiencing this in just a month. So I won't be surprised if next month we do get some kind of research preview that OpenAI demos to us and shows us what an agent is capable of doing. But I do know as well that it could be February. So since AI agents have been a bit tricky and we haven't got them to work completely reliably. Now, if you're wondering what other kinds of agents OpenAI are gonna be releasing in 2025, we also have the fact that they are going to be releasing several related agent research projects, and another one is not just one that works on your computer, but it's gonna be one that execute tasks in your web browser. So this is gonna be something that you have in your browser and it's gonna be able to do things for you. And I've already seen examples of web browser agents that I'm gonna cover in a moment. And Sam Altman has actually spoken about agents. He believes they will change everything.

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. 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 that. If that works as well as we hope it does, that can, that can really transform things.

Now he actually also gives us a very interesting demo. Not really demo, but more of the sort of example that we're going to be getting with AI agents in this extended talk on his channel.

Maybe I can give the following example. When people talk about an AI agent acting on their behalf, the main example they seem to give fairly consistent is oh, you can like, you know, you can like ask the agent to go book you a restaurant reservation and either it can like use OpenTable or it can like call the restaurant or whatever. And you know, it's like, okay, sure, that's, that's like a mildly annoying thing to have to do and it maybe like saves you some work. One of the things that I think is interesting is a world where you can just do things that you wouldn't or couldn't do as a human. So what if, what if instead of calling one restaurant to make a reservation, my agent would call me like 300 and figure out which one had the best food for me or some special thing available or whatever. And then you would say, well that's like really annoying if your agent is calling 300 restaurants. But if it's an agent answering each of those 300, 300 places, then no problem. And it can be this massively parallel thing that a human can't do. So that's like a trivial example. But there are these limitations to human bandwidth that maybe these agents won't have. The category I think though is more interesting is not the one that people normally talk about where you have this thing calling restaurants for you, but something that's more like a really smart senior coworker where you can like collaborate on a project with and the agent can go do like a two day task or two week task really well and you know, ping you when it has questions but come back to you with like a great work product.

Now if you want to know about that voice agent that he was actually talking about, they actually did a very short demo. This is something that you probably didn't see because number one, like I didn't even cover this and no other channel has really covered this because this was a private demo that OpenAI actually did, but it actually showcases what OpenAI's agents are going to be doing in 2025 right now where they're going to be able to call different stores and do things on your behalf. So you can see right here you can say, hey, can you call these stores? And this assistant is going to be able to do many different things for you. Okay, calling now.

Hello, is this the tech store in LA Defense? My name is Ada and I'm calling on behalf of a customer.

Yes, that's the tech store in LA Defense. How can I help you?

Great. I'm calling to check if you have a one plus 12 and flowy emerald in stock. Our customers, Katya is interested in picking it up today.

Oh, actually we do have one plus 12, but we don't have that color.

I'm sorry.

Ah, thank you. And what time do you close today? Will likely come by later to pick up the phone if she's happy with the color.

Yeah, I mean, we don't have emerald.

But we do have silver and, and black. So yeah, I guess she can come up. We close at 7pm Perfect.

Thank you. I'll let Katya know that you have silver and black available and she can come by before 7pm thanks for your help and have a good day.

Thank you. You too.

Now, they aren't the only people that are doing this. You can see that Google have actually showcased their customer agent demo and I do suspect that this will likely be rolled out in the later part of 2025.

Just launched a customer agent and it leverages Gemini and Vector Search to deliver a seamless shopping experience. What can we help you find? Well, I'd like that shirt, but I guess I have a few other specifications as well. So find me a checkered shirt like the keyboard player is wearing. I'd like to see prices, where to buy it and how soon can I be wearing it. And include the video. Now, the customer agent is using Gemini's multimodal reasoning to analyze the text and video to identify exactly what I'm looking for. Then Gemini turns it into a searchable format. Okay, how cool is this? It found the checkered shirt I'm looking for, right? And some other great options in no time. And that's because these results harness Google's trusted search technologies, which ensures customers like me get the right results in record time. The suggested products are grounded in Simple Fashion's inventory and historical performance data to make sure customers leave happy and with that purchase in hand. Okay, the first one is perfect. It looks like I can have it delivered in four days or pick it up nearby today. Like I said, I want to be wearing it tomorrow night. So I'm going to go with the local store to be safe. Of course, it never fails. They only have three left in my size. I don't want to miss out on wearing this shirt. So I'm going to go ahead, give the store a call and ask them to set it aside for me. But first, let me tell you what's happening behind the scenes. Symbol Fashion's customer agent is using Google Cloud's full suite of AI capabilities to offer customized support interactions. You know, facilitate transactions like purchases and returns and ensure that I'm receiving the most up to date information in real time. I'm so close to having this shirt for the concert. Let's give the store a call.

Hi there, this is the Symbol Fashion customer agent at South Las Vegas Boulevard. Am I speaking with Amanda?

Yes, this is Amanda.

Great. Thanks for reaching out. Amanda, I see you had a session on another device. I've sent you an SMS message with a link to our live chat companion. If you would like to switch to chat, please click the link. How can I help you today?

I'd like to purchase the shirt in my cart with the cart I have on file.

Absolutely. I see you're also a Symbol Fashion Rewards member. Looks like you have a 20% off voucher available to use. Would you like to apply it to this purchase?

Yes, please. That would be great.

The shirt you're purchasing goes well with these items also available for pickup in your preferred size. Would any of these be interesting to you?

Absolutely. Please add the white shirt and the boots to my cart.

Great. Your total is $203.73. Okay to proceed with the card on file? Yes, your purchase is confirmed.

You.

Do you need anything else today?

No, I'm all set, thank you. Incredible.

Thank you. So when we actually take a look at browser agents, Google actually did release this thing called the Search in Research. And this is just something that's absolutely insane. People are still sleeping on this incredible AI tool. This is called Gemini Deep Research. Essentially, it allows an AI agent to browse millions of different websites for you and create a full entire research report. This is something that I've been using every single day. And it is something that is incredibly underrated. But I do suspect that in the future, research is going to be done by AI agents that are able to browse millions of different websites and then you're going to come back to your computer and you get a ton of information that you can easily sift through or just simply sort via whatever it is that you do want. So this is something that I am currently using. Now, if you do want to know exactly the kinds of AI agents that are going to be there and the other kinds of AI agents, Google released something called Project Mariner. Now, Project Mariner is essentially a prototype experiment and in this video you're going to be able to see exactly what that does because it's just absolutely insane. Basically you have a one click tool, it says Project Mariner and you can basically do a variety of different tasks in your browser. So I'll let Google take it away for the next two minutes and then we can get onto the next point.

Today I want to tell you about Project Mariner. It's a research prototype exploring the future of human agent interaction. And it's built on Gemini 2.0. Like with all new technology, it's important for us to build this responsibly, which is why we're starting small. We'll be getting feedback from a group of trusted testers and using their experiences to really shape how Project Mariner evolves. Let me show you how it works. So Project Mariner works in the browser as an experimental Chrome extension. I'm going to start by entering a prompt here. I have a list of outdoor companies listed in Google sheets and I want to find their contact information. So I'll ask the agent to take this list of companies, then find their websites and look up a contact email I can use to reach them. This is a simplified example of a tedious multi step task that someone could encounter at work. Now the agent has read the Google sheet and knows the company names. It then starts by searching Google for benchmark climbing. And now it's going to click into the website. You can see how this research prototype only works in your active tab. It doesn't work in the background. Once it finds the email address, it remembers it and moves on to the next company. At any point in this process, you can stop the agent or hit pause. What's cool is that you can actually see the agent's reasoning in the user interface so that you can better understand what it is doing. And it will do the same thing for the next two companies, navigating your browser, clicking links, scrolling and recording information as it goes. You're seeing an early stage research prototype, so we sped this up for demo purposes. We're working with trusted testers to make it faster and smoother and it's so important to keep a human in the loop. After the fourth website, the agent has completed its task listing out the email addresses for me to use. And there you have it. We're really just scratching the surface of what's possible when you bring agentic AI to computers. And we're really excited to see where this goes next.

Now, there was also this very interesting demo that there wasn't too much revealed about, but this was the runner H agent. So this is the kind of agent that can scan the web and do a variety of different tasks for you. This is a really, really interesting AI agent because it is currently state of the art in terms of the things that it can do. So I will be very intrigued to see how the other labs manage to outperform this current AI agent because they did, you know, just come out of stealth and just drop a bomb. And you can currently sign up for the wait list, which I currently have. So I'm going to be pretty excited to see if this manages to compete with any of the competing labs. I do want to see if it manages to use some of the proprietary models it currently, according to my research, has used, I think a 2B model and a 3B model. So it's going to be super, super fascinating to see how they manage to execute on this. And they can scan the website and just do a billion different things. So Internet users are going to get really, really powerful. For example, another thing that I think is going to be doing really well is this kind of workflow agent. So these kind of workflow agents are essentially where you connect a different number of apps together. And this one is one that I built to be able to actually get business details. You can see, get business details. It runs through a few APIs and actually uses Claude to be able to store all those business details in a nice organized table. And this is basically something that I'm building for my community. Someone wanted to have an automated way to actually get leads, such as websites. And of course they wanted all the contact info to be able to be automated, just an automated way to grow your business and find leads. And I feel like stuff with that in 2025 is really going to becoming even more commonplace. Of course, if you want something like that, don't forget to check out the agent workflow section of my community. This is where you can see I've got ready to use AI Agent workflows automate your income streams and of course scale your business. Then of course what we do have is physical AI. Now physical AI will be one of the biggest themes for 2025. And, and this is because many different companies are going to be really, really pushing the boundaries of what it comes to when they are going to be making these robots super human like and realistic. The reason being there are a variety of different advances that now make it easier to develop humanoid robots in the sense that you can make them a lot more realistic, there's a lot more environments that you can test things in, and we're likely going to see a real increase in the number of humanoids around, and these things are going to be doing a lot of human work. Now, wondering what I'm talking about? If you aren't familiar with the Boston Dynamics demo, this was something that I don't want to say it broke the Internet, but I don't think people really, truly realized what this meant for society. I've already seen several companies and countries having, you know, their humanoid robots, but this was one of the first ones where we actually got to see a fully humanoid robot that was basically doing something autonomously. And not only was it doing things autonomously, it was even sometimes making its own mistakes and, and then correcting its own mistakes and being able to continue. Now, this is, of course, what many companies are trying to do, companies like Tesla and Figure. These companies are really trying to ensure that they can, you know, expand the workforce by building automated workers. And these kind of robots are going to be the kind of employees that we're going to be seeing every year increase in terms of the sheer numbers of these individuals. So it's going to be really interesting in 2025 because it's quite likely that, you know, some other companies are going to come out of the woodworks and show us the kind of humanoids they've been working on. And it's quite likely that we're going to see more and more extended demos of these robots being able to perform certain tasks and of course, be able to do them really, really effectively. Like what we see here, the robot manages to make a mistake. It's like, wait a minute, what just happened? And then it manages to fix this. Now, this also is a completely electric robot. So this is something that is 100% electric, which means, of course, it's better for the environment. Of course, it's a lot lightweight as well. It's a little bit more lightweight. It's not like bulky and hydraulic like the previous version of Atlas, which is what we had before. But something like this is going to be super, super interesting because I think once we start to get a sheer scale of these robots that are able to do a lot of human tasks. It's going to really start to boost the economy. And I think certain things will change now. I think we won't probably get like robots that are going to be completely working in factories. I mean, there's definitely going to be some that are working in factories, but I don't think the full scale comes to the later years. But I will expect some really, really, really impressive demos. And one of the companies that managed to do something really incredible was this company right here. Of course, you've got 1x robotics. They've been able to make this robot that is remarkably effective at doing human like motions. It's really smooth, it's really effective. It's. This is something that is really, really cool that a lot of people have seen. We've seen them testing their robots and we do know that some major updates are coming in 2025. And this robot AI startup, Physical Intelligence, actually raised $400 million from Bezos and OpenAI. And that was because they built this first foundation model for robotics that is essentially called PI zero. And this is where they have a generalist policy that is able to do things autonomously. So these robots just literally take two arms and they're able to do a variety of different things completely autonomously, which is pretty crazy when we take a look at what it's able to do, it's able to fold clothes, it's able to, you know, pack a cardboard box. It's literally able to do a variety of different things that are really, really incredible. And honestly, looking at it, you wouldn't think that this is autonomous, you would think this is teleoperated. But no, this is just a generalist policy that is simply, literally automating this completely, completely autonomous. So previously when people were, you know, saying, ah, this demo is not autonomous, it's not autonomous, why do we even care? Well, this is actually autonomous. So this is something that, you know, you should probably be paying attention to because it's like this is where the real innovation is coming from and where things are really starting to change. So this is one of those things where I think most people, like I said before, aren't paying attention to, need to be paying attention to. And, and it's quite likely that future iterations of their generalist policy like pi2, pi3 are going to be rapid, rapid improvements. And when I watched this entire video, I made an entire video on it. I was quite surprised that this was the first time that we'd seen this and the video didn't even. I don't even think it got like, more than 20,000 views. But I think later on in the year, we start to see more robotics demos because their policies actually do generalize to other robots, which is what we've seen with other robots. So I might include some of those demos here, but it's quite likely that robots are going to get a huge boost. Now, that's also because of Genesis, which is that robot simulator thing, which essentially it manages to map reality, not actual reality, but I guess you could say it simulates reality. And because it can simulate reality really accurately, that means that it's able to literally be able to be trained in that simulated reality. And then when it's actually imported into the real world, that means that it's going to be able to perform a lot more effectively. So that is why I think this Genesis thing is going to speed up everything by tenfold. And that is why I think then in 2025, we're going to have tons of robotic demos that are just super, super surprising, that are probably going to blow some people's minds. We've already seen robots walk really effectively. We've already seen robots be able to be completely autonomous and do tasks that humans would already do. So it wouldn't be surprising if, of course, now we do get those demos that are going to be very, very effective. And here we have the head of Microsoft AI actually talking about how we are going to be getting infinite memory, which is essentially where these models that we do have are essentially now never going to forget anything. And this brings up the very interesting debate, if a model, if an LLM, if an AI system never forgets anything you tell it, why would you switch to another AI system? So I think companies are rushing to get out this, you know, unlimited memory kind of thing, and it's going to be really important. But he literally says here that this will come online in 2025.

Memory is the critical piece, because today, every time you go to your AI, you know, you have a new session and it has a little bit of memory for what you talked about last time or maybe the time before. But because it doesn't remember the session five times ago or 10 times ago, it's quite a frustrating experience to people because you don't go and invest deeply and really share a lot and really, you know, look to build on what you've talked about previously because you know it's going to forget. So you sort of tap out after a while and it turns you know, into a shallower experience. But we have prototypes that we've been working on that have near infinite memory. And so it just doesn't forget which is truly transformative. I mean, you talk about inflection points. Memory is clearly an inflation inflection point because it means that it's worth you investing the time because everything that you say to it, you're going to get back in a useful way in the future. You know, you will be supported, you will be advised, it will take care of, you know, in time, you know, planning your day and organizing how you sort of live your life. So it's that capability alone, which I expect to come online in 2025 is, is going to be truly transformative.

And for those of you who think that that just might be one clip. He actually said this twice in another interview. He said, like memory is going to be done, like we will nail memory in 2025.

What's interesting about these, I'm really confident 2025 memory is done.

Permanent memory.

I mean, if you think about it, we already have memory on the web. We retrieve from the web, you know, all the time.

Quite what's interestingly as well, we also have another employee, an OpenAI employee or member of technical staff. I'm not entirely sure about this employee. They are quite shrouded in secrecy. But we can see right here that they've posted soon because someone actually posted say $200 a month and no increase in user memory is whack. What's the outlook on that? And you can see here, he says soon we're going to have infinite memory with this infinite icon right here. So it's quite likely that we're going to get infinite memory very, very soon. It's just a matter of when they decide to roll that out. Take a look as well. There are going to be more thinking models. Thinking models are models that are a lot more smarter. Now, for most people, you won't actually realize this at all. You won't really know what's going on. And I say this because currently the O1 models, from what I've seen, they're only really applicable to those who are in some form of really extreme academia like mathematics, physics, science, coding. Anyone that's at that real level is going to be getting value from it. So for those of you who aren't at the area where you're going to be using these thinking models, then of course this is going to be something that isn't of value to you. But it is something that is essentially the new paradigm in 2025 simply because of course the fact that the GPT series is a little bit slowing down. But the only thing with that is that because they're slowing down, they're of course finding new ways to scale these models to make them smarter. And of course they are doing that with the O1 series and thinking model. You can see right here that Google is of course adding their weight when it comes to the thinking models they are deciding to introduce Gemini 2.0 flash thinking. An experimental model explicitly shows its thoughts. And this is something that is really cool because it's something that is a little bit smarter than the other models because essentially rather than just responding instantly, it thinks through its problems, then searches through them and then comes up with a response. Now I actually did test this myself and I decided to ask the model how to solve the riemann hypothesis in 2025. Give a hypothetical. Of course I don't think it's going to be able to do that in 2025, but you can see if you open it up, you're going to be able to see exactly what the model is thinking and how it got to its response. Now I personally think that in 2025 this is going to be valuable because sometimes when you input a prompt and you see that the model makes a mistake, you can kind of look back at the steps and trace where it made a mistake with your response and the output, which is of course really useful. But I think something like that is going to be super, super valuable. Now, when we actually also do take a look at how the faster iteration cycles are going to come, we also do need to look at the fact that there are going to be an incredible level of progress. So you can see right here that this person at OpenAI talks about how O3 is very performant and the progress from 01 to O3 was only three months, meaning that from O1 to the O2, which is essentially their second iteration of the model, they only had three months of time, which shows you how fast the progress is going to be. And the new paradigm of chain of thought to scale, inference, compute, which is way faster than pre training paradigm of a new model every one to two years. So previously they would have to do that thing where they spent so much time pre training, so much time gathering data, so much time annotating the data of course, then of course fine tuning it to specific to plastic domain and now all of that just takes three months because they are no longer doing that anymore. And this is a simply faster way to scale and this is all explained.

Here, actually see models improve faster in the next six months to a year than we saw them improve in the last year. Because there's this new axis of synthetic data generation. And the amount of compute we can throw at it is. We're still right here in the scaling law, right? We're not here. We haven't pushed it to billions of dollars spent on synthetic data generation, functional verification, reasoning, training. We've only spent millions, tens of millions of dollars.

Right.

So what happens when we scale that up? So there is a new axes of spending money and then there's of course test time, compute as well, that is spending time at inference to get better and better. So it's possible, and in fact, many people at these labs believe that the next year of gains or the next six months of gains will be faster because they've unlocked this new axis through a new methodology. Right. And it's still scale, right? Because this requires stupendous amounts of compute. You're generating so much more data than exist on the web, and then you're throwing away most of it, but you're generating so much data that you have to, you have to run the model constantly.

And you can see the previous state of the art when we look at the benchmark that is frontier math, you can see epoch AI on frontier math is at 2%. And then we can see that the second iteration of the model, and I know it says 01 to 03, and that might confuse some people thinking that 01 is the first model, O3 is the third model. It's just a naming thing. O3 is actually the second model, which of course is confusing, but just trust me, you can see right here that this is where we actually get that result from. And I think that this is something that is really, really interesting because this model was able to get 25% on a test that it hadn't really seen before, which is one of the first times that we have seen that. And this is kind of, you know, the first inclinations that we're really, really on a new level when it comes to AI. So this is something that I think is going to be really, really interesting. And I'm going to be, you know, quite excited to see how this actually changes everything now, of course, as well, I do think that this is going to happen. AI movies previously, we've seen some incredible, incredible progress in terms of the rate of production, in terms of the quality of video clips from AI models. This clip right here, this is from Jason Sahada. This was absolutely incredible. Every shot was done via text to video with Google VO2. If you haven't seen VO2, it's honestly incredible. This guy basically made an AI short film. He generated a bunch of different clips, stitched them together. Of course he choreographed this short film. But I want to talk about how good this is in terms of comparing it to previous AI models. And we do know that this is just the best that you can get right now. And as the old saying goes in AI, this is the worst it will ever be. So if this is the worst AI video will ever be, we can imagine what it's going to look like at the end of 2025 when there are going to be vast amounts of improvements. It's quite likely that we're going to get 1080p 4k resolution, a lot of more consistency, things not morphing at all. I mean, right now it just looks absolutely incredible. And the quality probably doesn't even look that good because I've downloaded this clip. So, I mean, when you watch the original, which is linked in the description, you're going to see just how good this thing really, really is. So that is something that you need to check out as well. Now, of course, as well, there is going to be AI hardware. So in late 2025, I do think we're going to get an explosion in AI hardware. That's of course, because if agents manage to work, then AI hardware will manage to work. And essentially with that, what we do have is a situation where, you know, we've already seen that Joni, I have, has confirmed that he's working on a new device with OpenAI. You can see that the confirmation comes in major New York Times profile about what I've been up to since leaving Apple. So one of the key designers of Apple has actually left and has gone to make something himself. But of course, now you can see that with that, that actually now means that he's working at OpenAI and of course he's building some kind of AI product. So I will be intrigued to see exactly what kind of thing he is doing. And of course, since he's working with OpenAI, it's going to be some kind of physical AI product, whatever it is, whatever product it is that they are working on. So this is going to be something that essentially takes off when we have the agents out. I'm not entirely sure how long it's going to take to get agents really running smoothly. There are a few kinks that some of them did say could take until 2026. But I do know that it is quite likely that we will get AI hardware products and that is not just because of OpenAI, but of course as well, we've seen that Amazon's new Alexa voice assistant will use Claude AI. Now, currently we haven't had any major updates just yet, but when this thing does come out, it is quite likely that we will get a situation on our hands where we will have a model that is rather effective. Okay. And Amazon Alexa has taken kind of the backseat in AI because that was one of the first consumer AI products that people had around in their homes and it was kind of like the advanced voice mode of the decade. But now that is going to be using Claude AI and it's going to be able to be powerhouse. And they're literally calling it remarkable Alexa in here. And it did expect to launch in October and require a subscription fee, but we haven't actually had anything from that just yet. So I will be intrigued for more voice assistants. I do think that that will become commonplace. I'm not sure why just yet, but I do expect that in 2025 AI hardware might be a trend that some of these companies take advantage of. And you know, I actually recently looked at this website called friend.com and I think that AI is going to become a little bit more human. So one of the things that I think as well is that right now, whilst AI is, I guess you could say it's a bit weird in the sense that like, even when you talk to advanced voice mode, sometimes it's not human enough. It just. And I can't believe I'm saying that because it sounds remarkably human, but like, sometimes you'll know that it's ro. But imagine an AI that calls you and messages you and has weird things and their own problems and stuff like that. I think that is going to be a theme moving into 2025, late 2025. And this is one of the devices, this is a friend.com device. This is something that basically does exactly that. And I've spoken to this AI a few times before and it was remarkable how human like and realistic it was is. So that's going to be something that's super interesting as well. So with that being said, let me know what you guys think about 2025. I think there's a variety of different things and I think if I missed anything, don't forget to leave a comment section down below.