No Priors 第144期｜Sarah 与 Elad 展望2026年AI

How can we even begin to wrap this year up? The AI field has grown, breaking out into the mainstream and taking center stage with policymakers. ChatGPT shipped massive numbers and asked for massive dollars. Gemini and Google roared back strong.

On the application front, AI coding has shifted to agents and is eating up all of our inference capacity. Doctors are adopting clinical decision support en masse, as are law and customer support. Enterprise adoption is accelerating.

On the research front, the race has multiple live players, with open source closing the gap, too. A handful of Neolabs, new research labs got funded this year, and the narrative is changing. Ilya is calling it the age of research. People are trying different ideas around diffusion, self-improvement, data efficiency, EQ, large-scale Asian collaboration, continual learning, and energy transformers. It’s more open than it’s ever been.

Finally, we had a lot of attempts to make AI reach into the real world, with renewed optimism around robotics. Next year, those companies are going to start making contact with reality. From a prediction standpoint, personally, I think we’re going to see somebody make a lot of money—hundreds of millions of dollars—trading markets with LLMs next year. It’s inevitable.

So we’re in the second or third inning. Markets are running a little hot and a little volatile. It’s hot in the hot tub. Elad, it’s been a year.

I know. How’s it going? 2026, baby.

Are you feeling the AGI? Are you feeling the AI winter in a good way?

I think I’m actually just feeling microplastics. I think I’m now 80% microplastics and just increasing my microplastic consumption. A friend of mine actually launched a new water brand that has no microplastics, by the way. It’s called Loop, and they have glass bottles; also, the cap doesn’t have plastic.

Does it come with continual testing?

Yeah, that’s continual testing for you. They did actually try to take out all the microplastics, and I guess water in actual glass bottles has more microplastics than plastic bottles because of the cap.

Okay, we’ll check back in with you in ’27 to see if you feel—

Yeah, but I’m just completely horrified by plastic. I’m actually really worried about microplastics. What about all the little glass particles? Aren’t you worried about that? People talk about microplastics, but not microglass. I’m much more concerned about that.

I don’t think those particles end up embedded in you permanently.

Silicon. You’re not worried about silicons? When I go to the beach, I’m like, “Oh no, microplastics everywhere.” I’m actually very willing to insert silicon in my [laughter]

Wow, that was—yeah, I’m not going to say anything. We can keep going. What’s happening in AI, Elad? Where are we, and what are you most excited about?

Yeah, I guess for ’26 there’s a bunch of stuff I think will be interesting that’s coming. I think there are probably 4 or 5 things. One is that people will proclaim yet again that AI isn’t doing much and that it’s overhyped, like that MIT report people are quoting that I thought really didn’t matter. The reality is that the technology takes, like, 10 years to propagate, and people are getting enormous value out of AI already and are going to get way more out of it in the future.

Undoubtedly, next year there’ll be these overstated bubble claims, as well as “AI actually isn’t working that well” kind of claims. That happens every technology cycle, and we’ll just hear it again next year. There’ll be pundits and discussions and just a bunch of waste of time on it.

I think another prediction for ’26 is that the next set of verticals will hit massive scale. This year we saw the consolidation of coding into a handful of players, medical scribing into a handful of players, and legal into a handful of players, like Harvey and others. I think we’ll see that next set of consolidated verticals happening. That’ll be interesting. I can keep going—I have a bunch of these. Do you want to go next? I just did 2. Why don’t you do 2?

Maybe I’ll react.

Or react.

I’ll react, and then I’ll give you 2 predictions. I have to think of my predictions while I’m reacting, so I’m glad I have at least 2 threads.

I think that the overall sentiment on AI in the investing landscape is a lot of people getting stressed about the amount of capital they have at work, and then just a level of uncertainty around the adoption cycle and technical bets that people are making that they don’t have full first-principles confidence on coming to roost. I think any number of exogenous factors, plus noise about the speed of adoption—which, by the way, seems blinding overall, and we can talk about what the constraints are—

Yeah, it’s so fast I don’t even know what people are talking about. I just saw a report from this group called Offcall that talked about adoption of AI by doctors. There is just amazing adoption across several different categories, like documentation and clinical decision support, with things like Abridge and OpenEvidence, and obviously the general models.

There’s massive enthusiasm from most of the physician profession here. Of all the domains that were professionally considered more conservative, the fact that there is this desire to have things that make work better seems like it will obviously continue in the other professions. I think this is super underdiscussed: The people who tended to be the slowest adopters of technology love AI. That’s physicians, that’s lawyers, and that’s certain accounting types. It’s actually kind of fascinating. It’s compliance—it’s all the people who never adopt technology who are now adopting this stuff fast. I do think that’s really notable and very underdiscussed.

It will keep happening. There are actually lots of professions where being able to reason and interact with unstructured data is very useful. I expect that there’s going to be some negative market current. If Nvidia doesn’t overperform by some massive amount one quarter, everybody’s going to freak out. But I think that has very little to do with the fundamental secular change.

Yeah, it has to do with microplastics and Nvidia. It’s my 2 cents.

It has to do with microplastics, as you said.

Yeah, it’s true. Actually, the silicon there is in the air. I bet they have microplastics all over the place. It’s messed up, Sarah.

It’s part of the trade. If you make $20 million as an average Nvidia employee, you also have to have microplastics in your blood.

Don’t listen to this, Jensen. Jensen’s our next guest. You can’t hear that.

1% microplastics in the blood.

I think a third area is that the next set of foundation models is going to come. By that, I don’t mean the new labs and the next-generation LLMs, which of course will happen. I mean physics, materials, science progress by models, and math progress.

I think what’ll happen is there’ll be 1 or 2 use cases where it works really well for something. They’ll invent some new material, or there’ll be some conjecture proved, or something. Then it’ll fall into this overstated hype cycle of, “It’s going to change everything about physical sciences,” or whatever. That one-off will be overstated, and in the long run, the trend will be understated and incredibly important.

That’s another prediction for next year: There’ll be a couple of anecdotal one-offs in science that will make people say, “Look, science is solved.” They’ll realize science has been solved, and then later science will be solved.

I have 3 quick predictions for you. One is that there’s going to be some collapse of sentiment around a set of robotics companies next year—not because the field actually isn’t going to progress, but because people are beginning to project timelines, and not everybody is going to deliver on those timelines.

What’s your timeline?

I think that we’ll see humanoid and semihumanoid robots get deployed at small scale in environments, whether consumer or industrial, next year. Not everything will work, and because there’s this hype cycle around humanoids overall, as soon as something doesn’t work perfectly—which it will not—people are going to freak out. Then there’s going to be some bifurcation about people investing—

Yeah, I mean, we’re in year 15, 17, whatever, of self-driving—something around there—and it’s really working now. But it seems like robotics should have maybe a faster curve, but a similar curve, right? It’s going to take some time to figure all this stuff out, and then once it’s figured out, it’s going to be really valuable.

The big question for me on robotics is interesting. If you look at self-driving, there are 2 dozen, 3 dozen, whatever, legitimate self-driving companies—really good teams and good approaches and all the rest. Then arguably the 2 biggest winners, at least now, are Waymo and Tesla, which were 2 incumbents, right? Waymo is Google; Tesla is Tesla.

So I wonder what will happen to robotics. It feels to me like Optimus, or some form of Tesla robot, will be 1 of the winners, most likely—high probability. And then the question is, does Waymo just adapt what it’s doing for cars to robots as well? Because there are some similar problems there.

Is it some other big industrial company? Is it startups? Who are the winners, and why? Structurally, when you have a lot of capital needs but also a lot of hardware and manufacturing needs, that's going to favor incumbents, as in self-driving, right? I guess, arguably, the other winners in self-driving are Chinese companies—Chinese car companies—which are banned from coming into the U.S. market, and those will probably also be winners in robotics, right? The most likely global winners in robotics will be some subset of China, plus Tesla, plus something else, right? Maybe one of the startups.

I think that's right, but in most industries, incumbents are more likely to win than startups if you're just looking at it as a numbers game.

I don't know. I don't think so. I think there are startup industries where startups should win and incumbent industries where incumbents should win, and they have different characteristics in terms of market structure, capital needs, certain types of expertise, and supply chain. I do think there are markets where incumbents should definitionally do better. They don't always, but they typically do. And then I think there are markets where startups will do better.

Sure, but I don't argue that some markets' moats are structurally deeper, right? One way that you might look at autonomous vehicles is that it's one very complex, single-use-case robot. It mostly does locomotion. It does lots of other necessary types of prediction, defensive driving, and whatever else, but it's a single-use-case robot.

Yeah, and we forget there are a lot of good ones like that. Dishwashers are great single-use robots. Vacuum cleaners are great. There are all these things that we actually have that are robots in the home that we pretend aren't. We forgot that they're robots. Elevators are robots.

No, seriously. Escalators are robots.

I'm going to use the language that, for a robot to be a robot, it has to be somewhat intelligent, right? A dishwasher doesn't count as a robot; it's an appliance. A self-driving car does count as a robot.

Where's the border of intelligence for you?

I think it's probably some level of generalization, right? It can work in different environments. It can work on different tasks. It can work on different objects. Otherwise, a self-driving car is okay?

Yeah, I don't know. I didn't have that complex of a definition. I just had it as something that will do certain preprogrammed types of labor for you. But maybe I have a better definition. Let me look up the definition of a robot: “A machine capable of carrying out a complex series of actions automatically, especially when programmable by a computer.” But all these things have chips in them now. Your dishwasher has a chip in it, right? Or a computer in it.

Okay, yes, but I would argue that robotics has not been an interesting area of innovation without intelligence. That's the relevant set for you and me and many people who are looking for something that changes quickly.

Yeah, that's cool. On the topic of robots, the biggest trend—perhaps one of the biggest trends—of 2026, 100%, will be that self-driving will really begin to matter. That'll be both in terms of your own car and in terms of Waymo and Tesla cabs. It's going to be one of the big things that's talked about next year. So, I think on the robotics theme, that's the big one.

I think if you look at all of the potential use cases for robots besides self-driving, self-driving is a single use case. The Optimus team actually proves this. If you take a model that is powering Tesla's self-driving and put it in Optimus, it can do locomotion, but it can't do many other things. And you still have to do the hardware, like manipulation. I think the advantages here are not as strong as you believe they are. Some set of startups—

The scariest competition is the Chinese, but I do think that there is opportunity here.

Oh, I totally think there's opportunity for startups. Don't misinterpret me. I just think it's not just the fact that you have a model or a base model. You have the expertise to build the model, but then you also have all the supply chain. I think that's really important because a lot of the same sensors that you need to use are already there, and you know how to think about actually procuring and scaling things. There's good overlap in terms of some of the other skill sets that are needed, which take a long time to build at a startup or are a little bit painful to build.

People do it. It's fine. I mean, I did it, and SpaceX did it, and all these companies have done it. It's extra stuff. So that makes sense. I do think some startups will succeed here. I'm just trying to think through, besides the startups, who's going to be big.

I also think there are 1 or 2 incumbent slots that will just default happen unless something very strange happens. One could have argued that should have happened in foundation models, where Google should have had a default slot. In the end, it did, right? It got there. I think it was very predictable that Google's models would get good. I may even have written a post about this 2 or 3 years ago, that Google would be relevant, because it had all the assets needed to be a really important foundation-model company. They obviously invented transformers, but they had all the data, all the capital, TPUs and GPUs, and some of the best people for all sorts of things. So it felt inevitable, and I think this feels the same to me. That doesn't mean it's right.

Do you want to talk about IPOs and M&A next year? What do you think will happen there? I think that's another big—that's theme number 4 or 5, I guess. 3 was different types of models, 4 was robots and self-driving, and then 5 would be IPOs and M&A. What do you think? More IPOs, fewer IPOs, more M&A, less M&A, different types of M&A?

It depends on whether or not the bottom falls out of the AI market at some point, right?

What do you mean by “the bottom falls out”? What does that translate into?

I think people just get skittish. The cycle here is: What are people scared of? They're concerned that demand isn't real—demand isn't real for AI to support the capex cycle—and that there's systemic risk from people passing the ball around in terms of who is actually responsible for the capex buildout and these credit agreements, or pay on delivery contracts for data centers and chips. What else are they afraid of? They're afraid of the concentration risk—too much concentration in Nvidia and a small number of other players.

Silicon. It's too much silicon.

It's too much silicon. You're damned if you do, you're damned if you don't. I was talking to a friend of mine who runs a large tech hedge fund, and they're already a foundation-model investor in multiple significant labs that may or may not go public in the next couple of years. They're like, “Okay, well, the question is, do you buy the IPO?” Their game theory was, “Actually, no matter what I think about it, I have to do it because retail will want it,” because they want to be part of the AI revolution.

If you're a hedge fund, you get benchmarked on annual performance. Because of the retail pop and some set of investors wanting to buy into it as a pure play—you don't want to say, “Oh, I can't miss it like I missed Nvidia”—then you have to buy it. His view was, “You buy the IPO regardless of your fundamental view of the company.” And I was like, “Wow, this is not the investing job I know how to do.”

What do you think happens?

I think there'll definitely be a lot more IPOs next year. If one of the main AI companies goes out, it'll probably do extremely well, depending on where they price. Obviously, if they're overly aggressive, it won't, but in general, I think there's so much retail appetite to participate in AI besides Nvidia. That'll just get a lot of other people to go public as followers. So I do expect there'll be a lot of them. It's just a question of which ones actually go out.

It's also a great way to raise huge amounts of money for some of these labs, potentially. So it'll be interesting to watch what happens there.

Any other predictions for 2026?

I did not believe that we were going to see that many unique consumer experiences besides ChatGPT. I think we are going to see a slate of consumer hardware that mostly fails, but I'm still open-minded to it. It remains to be seen if any of these scale, but I am seeing magical experiences with really different consumer-agent software that I actually want and will use.

I think people are barely beginning to—well, these companies are in stealth right now, but I do think there's going to be a lot more product people and model companies experimenting with this next year. So I'm pretty optimistic about that.

Yeah, I agree with that 100%. I think the big question is: What will end up being a breakout startup? And then what will be a startup that will grow really fast, get copied by the main lab or Google, and then just get incorporated into the core product?

The interesting thing is that unless a company truly hits escape velocity and builds a network effect or something else that's really defensible, usually incumbents can launch 2 or 3 years later and catch up. If they have the distribution and they have the core product, then—but to your point, I think it's very exciting, and I've been waiting for this for a while.

I think 2 or 3 years ago, this guy David Song, who was on my team at the time, ran a 2-quarter thing at Stanford where we had different teams apply from the engineering programs there. It was groups of people building consumer apps using AI, because we said, “This wave of AI is so fascinating. Why isn't anybody building anything consumer?”

We basically just gave people free GPUs to go and try stuff, and there was no obligation on their side to do anything with it in terms of us getting involved. It was just, “You go do cool stuff, because this is such a good playground.” There were really neat experiences being prototyped, and then I was just shocked that nothing happened for a couple of years in terms of really interesting consumer products.

I agree with you that there's so much room for that. I always wonder: Is it because there's a different generation of founders who don't want to work on consumer or who've forgotten how, because the big consumer companies have kind of aged out? Is it that the incumbents are just too scary? Why is there so little innovation on the consumer side of AI? I still don't quite understand what the issue is.

Okay, let's list the reasons. I do think that the incumbents are pretty scary. Anybody who was around for the last generation of interesting consumer ideas saw the ingestion of those ideas into the existing platform, as you pointed out.

Yeah.

So there's that. I also think that the first instinct I've seen from founders working on new consumer experiences is essentially building better versions of last-generation experiences with this-generation technology, and it ends up not being that interesting.

I actually think you have to be either quite close to research or pretty creatively ambitious to build something very different that has any chance. I think there's just not that many people who have had that experience set or that creativity, and now we're going to see it.

Yeah, I think it's pretty exciting. The other thing is, I was talking to a really well-known consumer founder who's running a giant public company, and his view is that perhaps, in the entire world, there are a few hundred great consumer product people—at least in terms of who are actually working on it.

Obviously, there's enormous human potential, and people who aren't working in consumer products could. But of the people working in consumer products, he thinks that at most there are a few hundred exceptional people who could actually come up with and launch their own product that would be interesting or good.

You could also say that maybe there's just a limitation on how many of these things can exist, given human potential within the set of people who are already doing it. I think that's kind of an interesting argument. I don't know if I agree with it, but I thought it was an interesting argument that he made.

I would limit myself to that number if it's also the set of people who have the context of what's possible now.

If you've got great consumer product instinct, but you're grinding away on the 50th iteration of an existing product—

Yeah. You're working on the little submit button in Gmail or whatever instead of actually going off and doing this. 100%.

Yeah. Cool. Anything else we should talk about, or any other big predictions for 2026?

I feel like a very big emergent thing that happened this year was the surprising funding of Neolabs—like, 3 through 8. What do you think of that? What do you think about alternative architectures? Do you have any point of view on all of the effort around getting reinforcement learning to be more general, continual learning, and some of the other research directions?

I think there are enormous amounts of really interesting research being done. There's a lot of juice to be squeezed out of these models in different ways, and I think that's really exciting.

Ultimately, these things become capital gains for certain types of approaches or models, because we know scale really matters. That means that eventually you have to have a collapse into a handful of players, because capital will aggregate to the things that are working the most. They're generating revenue, and so the question is: What are those things?

At what point do things just get locked in from a usage perspective, for whatever reason? There are all sorts of ways you can imagine this being built over time against some of the models. I think it's interesting, I think it's exciting, and I think we'll see how it plays out.

I think, to articulate what the arguments could be for new research directions, Ilya did this interview recently where he describes it as “the age of research.” To paraphrase, he basically says that, yes, he believes in scaling, of course, but there's some floor of compute that is not infinite where we can test ideas at scale.

If we have, let's say, secret ideas around how to get to more rapid or more compute-efficient improvement, then it actually isn't just a straight resource battle, which the rat race does feel a little bit like today.

I think the other argument you could take is that multiple architectures are really relevant in big domains of usefulness. People have done some research on this, but they just haven't been scaled, right? There's enough capital out there to test them, be they diffusion models or SSMs or whatever. That's going to happen this next year.

I think there's also a resource-focus argument. If Ilya is describing that some set of labs has an enormous amount of compute, but they have to spend a lot of that compute on inference today, then how much do you spend on your particular research direction—be it self-improvement, post-training, emotional intelligence, or very large-scale agent stuff?

Yeah, it depends on what you're doing, because inference is what ends up raising you money to pay for everything else, because you're generating revenue. So, sure, I think it's effectively your way to bootstrap into more and more scale.

I always thought—perhaps incorrectly; I actually probably think it's incorrect—that eventually you end up with evolutionary systems. That's really how you build AI, because maybe I'm over-indexing on biology, where, effectively, your brain has a series of modules that have different functions or tasks.

You have a visual system that's highly prewired to deal with vision really effectively. You have different areas of higher-order thought and learning. You have memory. You have mirror neurons that are involved with empathy. Your brain is actually very specialized in some ways.

Although, obviously, there are people who are born with literally half a brain hemisphere, and the brain rewires and covers all the functionality. There are a few famous cases like that. Fundamentally, you have a lot of stuff that evolves into very specialized tasks. It's almost like evolution or something, you know.

The question is the degree to which you recapitulate that as you're doing further development of AI. When do you start spawning off a bunch of instances of something and have some utility function evolving against that? Then you have some selection and recombination and all the other things you do to try to make some of that work, versus how much of it is a more analytical approach, a more experimental and iterative approach, or something done in a directed way.

I think it's really interesting to ask, because if you look again at biology as a potential precedent—although maybe a very bad one—and look at protein design, for a long time there were these super-analytically designed proteins. Then they came up with all these systems, like phage display and mutational scans, that give you dramatically better results than if you just sat and thought about it.

Now, of course, we kind of solved it with AI, where you have all this 3D structure prediction that's actually very good. That was AlphaFold and a few other things that really were breakthroughs there.

It feels like, in the context of AI, maybe eventually we end up there as well, right? You just evolve these systems, and that may be a very different type of approach and training. That may be where I think things really have an interesting break.

That's one of the reasons, arguably, people are so focused on code, because code is arguably a bootstrap into moving faster on the development of AGI. But I think code plus self-evolution is really the potentially interesting approach to get some really fast lift-off. Maybe not, right? We'll see.

What is the one prediction you have for 2026 that has nothing to do with AI?

Do you think about anything else, Sarah? [Laughter]

I do.

I'm joking.

Really? I mean, the other thing, by the way—one other prediction that does have to do with AI—is that I do think defense will accelerate in terms of startups and defense tech, and the shift to drone-based systems in general will lead to a massive reworking of how you think about war and defense.

I think that's going to be a huge shift that we'll see accelerate even faster this coming year. I think this is accelerating in part because of how the Trump administration has been approaching it, and how the secretary of war and everybody there have been thinking about it. But I think, in part, you just have enough density now of startups doing interesting things.

I think that's the other thing that's a huge shift. It's a hype cycle right now, and I actually think, again, it's a little bit underthought because it's going to be so big. Outside of AI, I think there are obvious, really interesting things happening in space—SpaceX and Starlink—and I think communications and telecom are a big shift. There are really interesting things happening in energy and mining, and I think there's a lot going on in the world.

I agree on defense, with some concern that we have to wait for the budget to actually shift from contracts to primes to some of these new companies at scale. But the demand—the need to be competitive in a world that's increasingly autonomy-driven—is so obvious.

I think hype cycles and booms are good in that they bring a lot of people to the table: capital, founders, and people who want to work in the industry. You can make a lot of progress in a quick amount of time, even if a lot of companies die. There's more enthusiasm in a very short period of time, so I agree with that. I also don't think that's necessarily bad, right?

What's your AI prediction?

I think that the GLP-1 thing is, despite all of the enthusiasm, still underrated for how much impact it is having, right? I think that the continual adoption of these is inexorable. I actually think it creates a path that's interesting for other peptide and hormone therapies.

I think the fact that it has been so effective has lots of second-order effects, both from people just being a lot less overweight directly and from the willingness to look at other engineered peptides. I think everybody understands now that delivery matters. There are these really incredible medicines, and I think the impact of that is going to fuel much more investment in anything that looks like that type of opportunity. I think that's exciting.

Yeah, I actually think one thing that you mentioned is really interesting. If you look at the biohacking community, there's a lot of peptide use now—different peptides that will do different things. Somebody will have some chronic corporal cheerle thing[?], and they'll fly to Dubai to get peptides injected or whatever. Usually, those are early indicators of potential larger-scale adoption by society.

I think that's a really interesting trend right now in general: this whole world of peptides and their uses. Is there a Hims of peptides? What's coming there? I think that's super interesting.

I also think the biohacking community, as you said—the set of people who were really early off-label GLP-1 adopters, interested in longevity, neuromodulation with ultrasound, stem-cell injections, for example—has been a fringe, small community.

I think it's going to get less fringe.

A lot of these things, traditionally, 10 years ago, came out of the bodybuilding community, right? The bodybuilding community was into creatine and all these things that are more broadly used now, but also other things, like sleep aids or magnesium and all this stuff.

And to round out this year-end episode, we've asked some of our friends for their predictions for 2026. I'm so curious.

Guest

My prediction for next year is that reasoning systems are going to translate directly to AIs that are much, much more versatile and much, much more robust. Reasoning is going to revolutionize not just language models; it's going to impact every single industry, from biology to self-driving cars to robotics.

Reasoning, I think, is the big, huge breakthrough that's going to transform a lot of different applications and industries.

Guest 2

In 2026, AI will stop being a reactive tool that waits for us to prompt it. Instead, it will become very proactive and get deeply integrated in our work life. It'll go where we go, hear what we hear, know what tasks we need to work on, and, in fact, most of the time, complete those for us before we even ask it to do so.

It'll be our coach that helps us improve our skills. It'll be our manager who helps us prioritize our work and manage our time. In short, it's going to be the best work companion we could wish for.

Guest 3

I think the main AI prediction that I have for next year is that context is just going to be the most important part of every single product. Honestly, one of the best experiences I've had with it so far is memory in ChatGPT.

I think there are going to be a lot more features whose goal is to extract the user's intent and make the onus less on the user to give the model, the system, or the product more and more context. In other words, how do you put the onus on the product to actually extract that from the user instead of the user having to do all of the work up front?

Guest 4

My prediction for 2026 is that there will be a whole new suite of product experiences that run on much faster inference.

Guest 5

My prediction for 2026 is that we'll finally stop copy-pasting stuff into chat boxes. Instead, I think we're going to have applications that make better use of screen sharing and context management across the sources that matter the most.

Guest 6

One prediction for 2026: there's so much talk of agents right now, and there has been for a while, but no one has truly created a mass-scale consumer agentic AI. I think the models are there today for this to be possible.

In 2026, we will see the group that figures out the right interface, system, and product create as big a step function in the overall experience as ChatGPT did when it first came out. I think this area is not nearly as ceded to the labs as people assume. It really is anyone's ball game.

Aaron Levie

Hello, Aaron here. First of all, I get quite awkward around doing selfie videos. This is my ninth take of this video, so I hope it goes okay.

My 2026 prediction would be that this is going to be the continued second year of AI agents, but in particular, AI agents in the enterprise, in either deep vertical or domain-specific areas. I think this is going to be the main way that we actually take all of the progress that we're seeing in AI models and deliver it into the enterprise.

You have to be able to tie them to the workflow of the organization. You have to be able to get access to the data that they have. You have to have the right context engineering to make the agents actually work. Then you have to do the change management that makes the agents effective. This is going to be a year where we start to see this pattern emerge more and more.

That equally means that we need to ensure that we have a lot more happening on agent harnesses. Shout out to Aorvosu[?] and Dex for that answer. It's definitely going to be the year of agent harnesses and seeing how you start to get an order-of-magnitude improvement in the models' capabilities by having all the right scaffolding around the model.

Finally, it will be the year of economically useful evals—really starting to figure out how these models end up doing a lot more knowledge-worker tasks in the economy. We're going to see a lot more of that in 2026. We saw some previews of that this year with APEX and GDPval, and a handful of others. We're going to see way more of that.

Guest 7

I think 2026 is going to be a very interesting year for American open models. Over the last year, the frontier of open intelligence shifted from America to China, starting with the release of DeepSeek at the end of 2024.

American institutions were slow to notice this erosion of American leadership in open intelligence, but I think they've noticed in a big way over the last half year, both at the government level and at the enterprise level. There are some really interesting new labs starting to come out with open intelligence as their directive, and there are a few of these—not just Reflection AI. These companies are starting to produce some very interesting small open models.

Next year, I think we'll see the U.S. regain leadership at the open-weight frontier at the largest scale, and I'm really excited to see that.

Guest 8

Hey, folks. My prediction for 2026 is that I think we will see AI become much more politicized. I think we'll see it become a major point of discussion for the 2026 midterm elections, and some people will come out strongly against it. Some people will come out probably supportive of it. I'm not sure which side's going to win out.

Guest 9

2025 has marked an incredible year in AI drug discovery.

Guest

In the past year alone, we've gone from being able to design small molecules on the computer to designing small antibodies and now, most recently, full-length antibodies with drug-like properties zero-shot on the computer.

If 2025 has been the year of research in AI drug discovery, 2026 will be the year of deployment. The models have finally entered an era where they're becoming really useful for drug discovery. Not only do they make things faster, but they're also allowing us to go after really challenging targets that have traditionally been difficult to address with traditional techniques. I'm really excited to see what comes next because the models show no signs of slowing down.

Bryan Johnson

Okay, my prediction for 2026 is it will be the year that YOLO dies. We will begin transforming ourselves from “you only live once” to “don't die.” I think right now we're kind of a suicidal species. We do very primitive things. We poison ourselves with what we eat. We design our lives so that we slowly kill ourselves.

Companies make profits by making us addicted and miserable. We destroy the only home we have. And somehow we celebrate these things as virtues. I think it's all backwards, and I think one day we'll look back and be pretty astonished that we behaved like this.

I think the shift coming is going to be simple and radical: that we say yes to life and no to death. It's simple, but I think it could be in response to AI's progress. And we do this defiantly as a form of unification.

I think it does require a lot of courage for us, though, to say we recognize how sacred our existence is. We don't want to throw it away, and we want to defend it with every bit of courage and strength we have because it is so precious. I think it's going to be the year we end YOLO and the beginning of “don't die.”

Andrej Karpathy

The most striking thing about next year is that the other forms of knowledge work are going to experience what software engineers are feeling right now, where they went from typing most of their lines of code at the beginning of the year to typing barely any of them at the end of the year. I think of this as the Claude Code experience but for all forms of knowledge work.

I also think that continual learning gets solved in a satisfying way, that we see the first test deployments of home robots, and that software engineering itself goes utterly wild next year.

Guest 4

My prediction for 2026 is that it's the year where everyone's perceptions are flipped. Currently, everyone believes that you can only use NVIDIA outside of Google, and it will be obvious that that's not the case. Currently, about a third of Americans hate AI and think it's really bad. That number will increase.

Currently, most Americans think AI is not useful. That will flip as well. And so, everyone's priors will be flipped. That's because the transformative use of AI will be so prevalent. The obvious utility of it will be so high that there is no way for anyone's priors to remain unchanged; cognitive dissonance will be wiped away.

Ben Spectre

Hey, I'm Ben Spectre.

Ash Spectre

I'm Ash Spectre.

Ben Spectre

And our prediction is that 2026 is the year of energy-efficient AI. Data center buildings are primarily constrained by energy, power availability, grid interconnects, high-voltage equipment, things like that. That's why xAI's Colossus was initially powered by on-site gas turbines.

The thing is, the demand for computing is growing. New labs like us and like Crusoe have a pretty remarkably insatiable demand for both training and compute. And this demand is currently outstripping our ability to push loads onto the grid. This means that in 2026, it will be really important to squeeze every available bit of tokens out of every watt.

That said, in the long term, chips probably matter more than power because chips depreciate much more quickly than the underlying power infrastructure.

So, for example, with data center power at $0.10 per kilowatt-hour, the chips cost an order of magnitude more than the power over a 5-year depreciation cycle.

So, in 2026, we think intelligence per watt is really important to squeeze as much intelligence as you can out of every unit of energy. But in the long term, we think it's the chips that matter more.

Happy holidays.

Happy New Year.

Thanks for the year. Happy 2026.

Happy 2026, listeners. Thank you.

Find us on Twitter [music] at no prior pod. Subscribe to our YouTube channel if you want to see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way, you get a new [music] episode every week. And sign up for emails or find transcripts for every episode at no-bers.com. [music]