Transcription
Good morning. I'm Gene Alvarez. And I'm Tori Paulman. And I think you'll agree, Gene, it's been a hell of a year. Oh, how many of you feel that 2025 has been a year filled with chaos, right? We've had geopolitical disruptions. We've had economic disruptions. You've had talent challenges running around trying to find these prompt engineers, people who can work with AI. We've had more emerging technology come at us this year than we ever have before. You've also had to deal with the changing cyber security environment and all the challenges it has brought. This has been a year filled with chaos.
Now to do that you needed to be a superhero. And in 2026 your superhero journey continues. And if you think about it, you're probably thinking, "Oh, Superman or other superheroes like that." Well, my superhero is Frodo. Frodo in Lord of the Rings has to go to Mount Doom. But Frodo doesn't do it alone. Frodo has others with them. And you're going to find others who help you drive responsible innovation, operational excellence, and digital trust. And you're going to meet three new companions. The first one is the architect. The architect is going to help you with AI platforms and infrastructure. Then you're going to meet the a the synthesist who's going to help you with AI applications and architecture. And finally, you'll meet the vanguard who's going to help you with security, trust, and governance.
Now, I'm going to answer the most popular question we get about this research every year. So this way everyone gets the answer all at once and that answer is how'd you come up with the list right now. Gartner tracks over 650 emerging markets. We follow where the VC funding is going. We do an analysis of all the patents submitted and then we have over 130 hype cycles and 2,000 innovation profiles that we go through. We also do an open call for submissions to our 2,500 plus analysts and ask them to submit so we can find out the ones in their desk drawers that they haven't even started to write about. And then finally, we pick the list. Well, our desk drawers are full, aren't they? Yes.
Now, we're going to tell you not just about the trends, but really how long we think it'll take for them to mature. In the along the way we're going to like highlight some real world stories but you need to know that these are early days for these trends and there aren't hundreds of examples of other organizations getting it right but that gives you an enormous opportunity to be the first to tackle these trends. Now the time horizon for this year's trends include now like right now in the next 1 to 3 years and then we have near which is trends emerging and maturing in the next 3 to 5 years and then we have the far out future trends that won't mature for probably 5 years or more. Now each of these trends are represented by one of our superheroes. Now you may have noticed something. There are no trends in the far out future category. Well, this is because investors have noticed that there isn't a single day that happens where we're not talking about AI and the invisible hand of economics is out there driving these markets at full speed. In fact, the pace is so intense that Gene had to update the trends from last year just because of geopolitical shifts.
Now, what does this mean for you? The most important thing for you to know is that yes, market conditions may speed these trends up, but they may also slow these trends down. Now, as Gene and I walk you through the trends, we're going to give you three key pieces of information. First, what is it? Mhm. Second, why is it important to you now and also in the future? And third, what should you watch for and do? Now, this is more important than ever before because time is changing. The pace of innovation is changing. And in fact, if you think about the next wave of of innovation, it's not going to happen next year. It's going to happen here this week.
So, let's get to know our first superhero. So our first superhero is the architect and in the architect we're going to see AI development platforms and AI supercomputing platforms. So let's take a look at our first trend AI native development platforms where people and AI team up. Now here is where we're going to see that AI now will join programmers. Each of your programmers or developers is going to have a Jarvis to work with them. And what this is going to do is it's going to give you a platform creating software where AI is a member of your team. Now these agents will help build alongside your teammates. So now think of this. You have 10 developers working on one project. Imagine now if we can have five teams of two partnered with AI and these tiny teams now can deliver five projects instead of one.
Now why is this important? Well, this is important because we've already seen this change the startup culture. One of the reasons why we had so many new technologies this year to go through was that we were seeing startups with very small teams using AI to create these new offerings for you. We're also seeing that this is going to be the solution to your developer productivity problem. Think about it. How many of you in this room do not have an application backlog? Right? No one. Everybody does. So this is going to be solution for that productivity problem. It's also going to help in terms of bringing non- tech developers or business users into these teams for development creating applications specific for your organization.
Now, what are some of the things to watch for and do here? Well, when it comes to those things that you need to watch for and do, right, first off, we have platform teams now. So you may have those two developers working with their Jarvis who work on a specific software platform and they develop all the applications on that platforms. So you'll see them organized in this way. But now because you're producing more and more applications, you're going to have to make sure that security guard worlds are also built into this development process. And then now when you look at your application backlog, many of us think, well, I'm going to get an application from the cloud. You may start to think first, can I have a tiny team build this application for us so that it can differentiate our organization in our industry?
Okay, so we have supercharged tiny teams, but they still need to find AI resources in order to develop and innovate. That's where AI supercomputing platforms come in. Now, where I'm from, when I grew up, this is what directions sound like. Okay, you're going to want to turn left where the old country store used to be. Right, this is a map drawn from memories and legacy knowledge. Now, AI supercomputing platforms are a little bit like GPS, right? They combine accelerators, orchestration, and high-speed infrastructure to help your developers develop in real time. So these things make realtime decisions routing your um your AI development to the right platform at the right time. This then hides all that AI complexity from the developer.
So why is this important? Well, we all want our AI running in the most optimal uh you know compute environment. One where we can think about cost and um and efficacy, right? But I like to think about trying to find a parking spot in a crowded city. It's often very frustrating and very time consuming. These platforms boost that speed and efficiency. And the examples that we're seeing uh Gene are things like biotech companies using AI supercomputing platforms to model vaccines and therapies in weeks instead of years. We're seeing financial services companies use these to model risk portfolios so that they can de-risk their portfolio management process. And we're seeing energy companies use these models to to map extreme weather so that they can optimize their grids.
So what could you watch for and do? We'll start by pinpointing where your high AI compute traffic jam is and focus on those areas because that will enable your developers to develop faster. You should explore how hybrid architectures and modular infrastructure will support your most important use cases. But remember, this isn't autopilot. And I think we all we all have good, you know, GPS experiences today, but you all remember the people who drove into the lake, right? By the way, that happened three times. So, you need to keep yourself out of the lake by building new skills for securing and governing composable platforms.
So, our next superhero that you'll meet on your journey is the synthesis. And here we have multiagent systems, domain specific language models, and physical AI. So let's take a look at multiagent systems. Now, I know this one may not be a surprise to you. Some of you may already have one agent in your organization or two. If you have two, you're into the world of multi-agent systems. But one of the things about multiagent systems is that they're modular and that they handle specific tests t tests. So, for example, an F1 team, the pit crew, everyone handles a specific task when that car comes in. I handle the left front tire, Tori handles the right front tire. We specialize in it and we are very good at it which means we can reduce hallucinations but yet support a complex workflow and we're orchestrated in a dynamic model because this car is going through different things in the race and when it comes in there are different things we may have to do to improve its performance and that orchestration is what happens with multiagent systems.
Now we're going to see this on single platforms. So you may have a single platform provider that has multiagent capabilities, but then we're going to go to cross-platform. So we're going to have model context protocols to look at agent protocols to look at and then we may even see the dawn of an internet of agents. Now what to watch for and do here? Well, one of the things that you'll need to do is start building multiagent systems now. But build your agents small, specific. Think of that F1 team and how specific they are in their test. Don't build these large monolithic agents. They become too hard to manage and they bring in the ch the challenge of having hallucinations or other problems occur. You want to also make sure that you don't think of these multiagent systems as a human. They augment a human. They work alongside with them. They will be dynamic and be able to call each other as needed to handle your organization that F1 formula car. I'm sure glad you're human.
Now, you heard about context in the keynote this morning, and we know that most large language models have gobbled up everything. They know everything about Taylor Swift, and they know everything about finance. They're a little bit like the Library of Congress in my mind. They're packed with every single book that's ever been written. But domain specific large language models are a little bit more like the the New York University law library, right? They're specifically focused on the task that you need to do. Now, think about the vast amount of clinical trial data out there for those of you in healthcare. Just to put this in perspective, last year alone, there were more than 350,000 clinical studies produced. Wow. And I did a little math. Uh, sorry, there were 50,000. And the math is going to make more sense now. So, did a little math on this, right? If there were 50,000 clinical studies produced last year, how many full-time employees would it take to read those and inform your scientists? Well, that is that number is 350.
Now when we think about what the value is to organizations, well it's obvious they spend less time searching. They get better results. So what do you you know what do you need to think about in terms of its importance to you? Even if you're not in healthcare, let's think about those of you in government. When do builders find out that they're out of code compliance? At the inspection, right? Or we could say just way too late. CIOS are sitting on a value gold mine. You have the ability to build domain specific large language models as a digital service. Now imagine a building code compliance model that both inspectors, employees, and consumers, builders can use. That would be very valuable. And in turn, that's going to improve the trust, security, and adoption that you're after. Mhm.
All right. What should you watch for and do? Well, the most important thing you need to do here is be transparent about what your model knows and doesn't know. Like for example, who Travis Kelsey just proposed to. Now, you're not going to have to hire 350 clinical study readers. That's good news. But you will need new roles for context and machine learning. For example, you'll need a context engineer who is constantly feeding your model with the most appropriate sources and ensuring they're always up to date. And you'll need a machine learning specialist who monitors for what's called catastrophic forgetting. And this is where the machine forgets when they learn something new just the way that humans do.
So our next trend here is physical AI. And this is where digital rubber meets the road. Now, a popular question that I get about physical AI is, "How do I know if it's physical AI?" Well, here's the very complicated test you will do. If you can pick it up and throw it out the window, it's physical AI. It's that simple. Physical AI is AI designed to interact with the physical world. Your Roomba, it's designed to interact with the physical world and it senses what's around it. It can act in that environment. It works alongside you while you're living there with your Roomba. It shares that physical world with you.
Now, why is this important? Well, it's important because as we move into the physical world, there are challenges moving into the physical world, whether we're using robots, drones, or we're using devices. So, in the case of robots, I can have a robot that will do one of the things that my family hates to do, clean the table after dinner, right? So it can decide well what stays on the table like the salt and pepper shaker, what goes in the dishwasher, hopefully the dishes, and then things that go in the trash. It's going to have to work in that physical environment. But at the same time, I could be an energy company and I'm using drones to maintain clear lines. So what it's going to have to decide is what's the difference between a power line and a tree branch. And hopefully it only cuts the tree branch. And then lastly, my Roomba. My Roomba can tell whether if it's cleaning a tile floor, a rug, it can sense that there are cords, stairs, it might go down, and it can even discover the surprise package my new puppy has left me.
All right. So, as we move into the physical world, we're going to deal with unpredictability. And that means that these models are going to have to learn in what they do. Which means occasionally a salt paper pepper salt and pepper shaker might wind up in the trash. Hopefully we don't cut any power lines. But testing and learning what we have to do in the physical world requires that learning process. So in physical AI, remember if you can throw it out the window, it's physical AI. One thing you don't know is that Gene's wife has a very special salt and pepper shaker set. And it would be devastating if that got thrown away. Something you learn when you're working with your colleagues on a project. Yes. Yes. It's a pair of roosters that we got in Spain and I'm not allowed to touch them. Amazing.
All right. So, let's meet our last hero. It's we need a secret hero uh for the salt and pepper shaker. This one. The Vanguard's job is to keep you ahead of the curve through uh preemptive cyber security, digital provenance, and geopatriation. These are exactly the skills your organization is going to need to verify authenticity and safeguard your assets in, let's face it, a rapidly shifting world. Now, let's start with preemptive cyber security. Who remembers the Minority Report? Probably everyone. Now, in case you haven't seen it, and if you haven't seen it, I'm going to spoil it here, but uh in case you haven't seen it, Tom Cruise plays a police detective in what's called the pre-rime unit. He works with humans who are gifted with precognition, who predict violent crimes just moments before they occur. This giving the police the opportunity to arrest the criminal even before they do the crime. Now, I don't know what's more terrifying, what I have just described, the fact that this movie is 23 years old or the fact that Tom Cruise has not changed at all. Don't know. It's an avatar. Yes, exactly. He's an avatar.
Using preemptive cyber security means using AI powered sec ops. And what I'm talking about here is a world where prediction is protection. Now, why is this important? Well, I think we all know that most cyber security tools are reactive in nature. In fact, almost 80% of the market today is buying tools that help you react as quickly as possible. But that mean that means that you've only you'd only know you've been breached once the the damage has begun. Right now your threat actors are already using AI to attack you more, you know, with more intelligence and more precision. Preemptive cyber security is about using that same power against them to anticipate, deny, disrupt, and deceive. Now imagine this. You get an alert on your phone that there's a possibility of an attack on one of your servers. Automatically, your AI SecOps is deployed, creates a new honeypot server, lures that attacker in, and and keeps them safe. Now, they think that they were successful, and you have contained the threat.