A.I. Scientists Are Here. But Is Progress Accelerating? | EP 170

I’m Kevin Roose, a tech columnist at the New York Times.

I’m Casey Newton from Platformer.

And this is Hard Fork. This week, FutureHouse CEO Sam Rodriques joins us in the studio to separate the hype from the reality of AI science.

Well, Casey, it’s time for some science.

Yeah, give me a second, Kevin. I’m just going to put on my lab coat here, get out my Bunsen burner, and see what you’ve got cooking for us today.

I have been obsessed with this question of what AI is and isn’t doing for science and scientific discovery. Obviously, this is something we hear a lot about from the leaders of the big AI companies. People like Dario Amodei, Sam Altman, and Demis Hassabis have all been saying things in recent months about how close they believe we are to solving new scientific problems, curing diseases, and fixing the climate with all of these new AI tools that they’re building.

Some of that is obviously hype, or at least has the sort of markings of hype. But there’s actually a lot of real stuff going on in AI and science that I just do not feel personally qualified to evaluate.

Yeah. I would also say that science has become one of the main ways that the leaders of these tech companies want us to evaluate them, because whenever one of their models does something horrible, the message we basically get back in response is, “Don’t worry, we’re about to cure cancer. Just hang on tight. I know that this chatbot might be driving you to madness, but if you could just give us a few more releases, we’re going to do some really good stuff.”

Yes. This is something that we’re also hearing now from the U.S. government. The Genesis Mission was announced by the White House just before Thanksgiving. That is what they’re calling a dedicated, coordinated national effort to unleash a new age of AI, accelerate innovation and discovery, and solve the most challenging problems of this century.

I thought the Genesis Mission was just them trying to get Phil Collins to play the White House Christmas party. [Laughter]

I guess not. So today we have brought in a bona fide scientist to help us understand which of the scientific discoveries and possibilities out there are real and which are not. We need an expert with a broad focus, someone tracking the impact of AI not just on biotech or drug discovery but across the different sciences. And, Casey, we have found the perfect person.

Let’s hear about him.

Sam Rodriques is the co-founder and CEO of FutureHouse and Edison Scientific, which are San Francisco-based organizations. I guess it’s both a nonprofit and a for-profit.

Have I heard that before? [Laughter]

Yes. Come back when he has his board coup.

FutureHouse is the nonprofit. Edison Scientific is the for-profit that spun out of it. I’ve been to their office in Dogpatch. It’s really fun. It feels like a wacky mad scientist lab. They’ve got all these lab machines that I don’t understand, and people running around in lab coats. They’re all talking about AI, and it just feels like a cool place to be.

They are building what Sam calls an AI scientist, which is an AI agent that can do parts of the process of scientific research. Sam is also himself a scientist. He has a Ph.D. in physics from MIT, and before he launched FutureHouse, he spent several years running an applied biotech lab. So he has seen this stuff happening from a couple of different angles.

Today we want to talk to him about what he is up to, but also get his vision of the entire landscape. Tell us what is working, what isn’t, where’s the hype, and where’s the real stuff. Sam has a lot to say about it.

I think it’s fair to say that Sam is on the more optimistic end of the spectrum of beliefs about what AI will do for science. But as you’ll hear in our conversation, he’s more skeptical than some of the most optimistic people who are claiming that we’ll cure all disease in 5 or 10 years.

Yeah, if you’ve been craving a little bit of cold water for the wildest projections, he has some of that to offer you.

So let’s bring him in. Sam Rodriques, welcome to Hard Fork.

Hello. Thank you.

We have brought you here today to be our science expert, our guide to the biggest recent AI-powered breakthroughs that are happening in science. This is an area that I understand in an ambient way is important and that there are big things happening, but neither of us are scientists, although I did make a killer baking-soda volcano in elementary school.

We have so much to talk about today, but before we get into some of the particulars, I want to ask you about the project that you’ve been working on. Last month, the commercial arm of your nonprofit, which is called Edison Scientific, launched a new AI scientist called Kosmos that you say can accomplish work equivalent to 6 months of a Ph.D. or postdoctoral scientist in a single run of the model. Tell us about how Kosmos works and where that 6-month number comes from.

Yeah, exactly. I’ll start out by saying that when I got that 6-month number, my reaction originally was, “There is no way that this is true,” right? We’ve now measured it in a bunch of different ways. I can walk you guys through that.

Just to take a step back: We’ve been working for 2 years on figuring out how to build an AI scientist. The concept here is that there’s so much more science that we can do than we have scientists, right? So how do we scale up science?

The thing that happened with Kosmos that is pretty cool is that Kosmos is the first thing that I think we’ve made that actually feels like an AI scientist when you’re working with it. You go in, give it a research objective, and it goes away and comes back with insights that are really deep and interesting and sometimes wrong, but about 80 percent of the time right. That’s similar to if you ask a human to go away and do something and come back: A similar percentage of the time, it’s right. It’s a new experience working with it, and that’s very exciting.

The 6-month number specifically: We measured this with a bunch of academic collaborators—scientists who had done science previously that they had not published yet. We gave the same research objective and the same data set to Kosmos and asked it to go away and make new discoveries. It came back having found the same things that the researchers had found overnight.

Then you ask the researchers how long it took them to find this in the first place, and they would say 3 months, 5 months, 6 months. That’s where the number comes from. It’s the amount of time that it took them to come up with the finding.

Let me ask you a couple of questions so I can ground myself here. Is this tool a box you type into like the other chatbots? If so, what is powering it? Did you build your own model from scratch? Did you fine-tune another company’s model?

Yeah. It is indeed a box that you type into. You give it a research objective. It’s not a chatbot; it runs for 12 hours or so before eventually coming back to you with its findings.

In terms of how it’s built, we build on top of a bunch of different language models from OpenAI, Google, and Anthropic. In any given run, we use models from all the different providers. We also have our own models for specific tasks that we’ve trained internally, where those models are much better for the specific tasks that we train them on than the models that the frontier providers make.

The key insight in Kosmos is this use of what we call a structured world model. One of the main limitations with AI systems today is that they’re limited in the length and sophistication of the task that they can carry out before they go off the rails—forget what they’re doing and are no longer on task.

What we figured out was a way to have them contribute to a world model that gets built up over time and basically describes the full state of knowledge about the task they’re working on. That means we can orchestrate hundreds of different agents running in parallel and in series, and have them all working toward a coherent goal. That was the real unlock.

Another thing that I found interesting about Kosmos is the cost. This model costs $200 per prompt.

Yeah.

Every time you give it a task, you’re paying $200. Why is it so expensive?

It uses a lot of compute. That’s the fundamental answer: It uses a lot of compute.

Give us a sense of how much.

An individual run from Kosmos will write 42,000 lines of code and read 1,500 research papers on average. If you run Claude, it might write a few hundred lines of code. That gives you some sense of how much compute is going into this.

Have you ever had a scientist whose cat walks across the keyboard and accidentally hits Enter and all of a sudden spends $600?

This is a problem. The thing that you have to understand is that if you are a scientist and you go and do an experiment, you get some data back.

You're going to spend $5,000 or $10,000 gathering that data. What scientists want is the absolute best performance that they can get. Scientists who have used Kosmos generally come back to me and are like, “They can't believe we're only charging $200 for it.”

Right?

And, you know, I will say: $200 right now is a promotional price. We actually have to eventually charge more.

It's going up. So get those prompts in before Christmas. [laughter]

Exactly. But really, if you have to spend thousands of dollars gathering the data, the cost at the end of the day is not the limitation. We do have to be very generous with refunds because people make mistakes. I made a typo, right? It sucks.

So what you just mentioned about the tests that you all ran to figure out how long this thing could run for and how much time it was saving scientists, that's about replicating existing research that's out there. But a lot of what we hear from the people who are running these big AI labs is the possibility that pretty soon AI will start making novel scientific discoveries. It will start doing things that existing scientific methods and processes can't do. How close are we to that?

That's already happening, actually. If you go and read the paper that we put out about Kosmos, we put out 7 conclusions that it had come to. Three of them were replications of existing findings; 4 of them are net-new contributions to the scientific literature, like new discoveries.

And of those, what's the most impressive?

One of the ones that we really like is about the human genome. It contains millions of genetic variants—differences between different people's DNA—that are associated with disease. For the most part, we know that a variant is associated with a disease, but we have no idea why.

We gave Kosmos a bunch of raw data about a huge number of different genetic factors: what the variants are, what proteins bind near the variants, all these kinds of things. We asked it, for type 2 diabetes, to identify a mechanism associated with one of these variants.

It came back and identified a variant that was not in a gene. Kosmos identified that this is actually somewhere where a different protein binds. It was able to identify what protein binds and what gene is being expressed, and connected that to the actual mechanism of the gene SIRT1, which is involved in the pancreas in secreting insulin.

Right.

So in this case, is what I'm hearing that your model was able to do some very fancy reasoning over some existing data and identify something that no other human scientist had gotten around to and might not have for a really long time?

Yeah, that's right.

Okay. I think science generally consists of deciding what data to gather, gathering that data, and then drawing conclusions. At this point, basically, it's step 3 that Kosmos is aimed at, and there's—

You left out step 0, which was getting the Trump administration to unfreeze your funding, [laughter] but everything else was right.

Yeah.

So what happens when you get a discovery like this from Kosmos? Do you have to then go validate it? Do you hand it to a team of researchers who then have to make sure it works? What happens next?

Yeah, absolutely. You have to go and validate it. In the paper, we describe how we went and validated that particular variant.

In general, when people are using it, literally, when you run a Kosmos run, the first thing you have to do is understand what it's telling you, because it has just done something that scientists think is 6 months' worth of work. You're going to sit there for a long time just reading and understanding it.

Once you've read and understood it, then yes, indeed, you're going to run various experiments, do your own analysis, cross-reference things, and try to convince yourself that this is true. Based on what your research objective is, you'll decide the next steps. In this case, I think it's probably unlikely that there's a new drug target from this particular finding, but you could run this on other findings. Eventually, maybe you find a new drug target and start a drug program.

One concern that I've heard people express about models like Kosmos is that this is just not where the roadblocks are. The reason that we don't have more AI-discovered drugs and designed drugs out there curing diseases is not actually because we don't have the research methods to discover them. It's because you have to go to trials, recruit human subjects, get FDA approval—all that stuff takes a lot longer than the actual discovery of the drug.

What problems are models like these helping to solve in our scientific process right now?

I really agree that the bottleneck at the end of the day in solving medicine is basically clinical trials. The easiest way to see this is if you look at the number of diseases that we know how to cure in mice. It's astronomical because, obviously, you can just run experiments, and in humans, things are slow.

That said, if you think that every experiment being run right now by pharmaceutical companies—every clinical trial—is optimally planned and optimally conceived, given the full state of knowledge, you are off your rocker. There's no way. Those experiments cost hundreds of millions of dollars.

The question is: We do have to run clinical trials, so how do we make sure that those experiments are the best experiments we could possibly be running, given all the knowledge and data we have? There's so much data that has insights in it waiting to be found, but we simply don't have people to go and find them. That's ultimately going to feed into better experiments and better trials.

Well, then I'm curious how you see your tool fitting into the workflow of today's scientist. Is it the sort of thing where I've completed my experiments and now I want some help doing some analysis? Is it that I have all these old experiments that I only did a little bit of analysis on, and I'm curious if I can squeeze any more juice out of them? What other ways are you seeing AI being really good right now for a working scientist?

Going back to me in 2019, which is when I was wrapping up my PhD, I had this gigantic data set, and I wanted to graduate because I was a PhD student, which meant that I was making $40,000 a year or something. There were a ton of great opportunities to go out and not be a PhD student anymore.

So I spent 6 months literally just sitting at my desk trying to analyze the data and draw conclusions, reading papers. That's where Kosmos fits in right now. You would just take that data set, give it to Kosmos, and it comes up with a lot of findings.

Right now, you need to go and do a bunch of manual work to validate those findings and so on. Pretty soon, it's going to come with findings and you're going to be like, “Great.”

Sam, I'm curious if you could help give us and our listeners a state of the world of AI science right now. Recently, the White House announced what it's calling the Genesis Mission, which is a federal effort to corral and harness all of these data sets that the federal government is sitting on and use them to do new scientific exploration.

We also have lots of efforts, including yours, and lots of things going on in and around the tech industry and the biotech industry, with people doing AI for materials science. Give us a sense of the lay of the land of what's hot right now in AI science. Where's the effort and money going?

In order to understand the landscape of AI and science, the first thing fundamentally that you have to understand is that AI is about building models. For example, what is a language model? A language model is fundamentally a model of human language.

It just so happens that when you build a model of human language, it learns how to think like a human in some sense, because humans encode their thoughts in language. This is one of the greatest discoveries, certainly in the 21st century, maybe of all time.

Similarly, when we talk about AI and science, what you have to think about is that you are modeling things. That is what AI does. There are 2 fundamental categories: modeling the natural world and modeling the process of doing science.

These things are fundamentally different. The reason to make this distinction is because we are modeling the process of doing science. The other side of the AI-for-science world is building models that can, for example, predict the structure of proteins, generate a new antibody, or create a new organism from scratch—all things that have happened in 2025, where there's a huge amount of momentum.

Of the things happening in the process of modeling the natural world, you mentioned protein folding and novel organisms. What has most excited you as a scientist that you've seen?

It's absolutely what's most exciting right now. I think, without a doubt, it's this trend toward what we call generative models. These are models that can produce examples of proteins or antibodies or whatever that have desired characteristics, basically from scratch. This is a new capability that we have never had before. And it's huge.

I'm curious about the reliability piece as you're running all of these experiments. I saw this going around on social media this week. I reproduced it myself. If you asked Google, “Is 2026 next year?” it said, “No, 2026 is not next year. It is the year after next.”

So, in such a world, Sam, some people might get concerned at the idea that we're now entrusting AI with all of our data analysis. How much time are scientists having to spend going back and essentially rechecking the work of the AIs? And what kind of tax does that place on their work?

Yeah, this is very funny. Look, you have to spend a lot of time going back and checking. But, to be clear, this is true regardless of whether an AI does it or whether you ask a friend to do it. If you're going to publish a paper, you damn well better go back and check it and be sure that you are confident.

And it's never going to be 100%, right? The best you're going to do is get to a place where it is similarly good to if you were doing it yourself, which is not 100% because you're not infallible. And checking the work is always going to be faster than producing it in the first place.

Yeah.

Right. By a lot.

A lot of our biggest scientific breakthroughs in history have come from these kinds of strange accidents, these moments of serendipity. Penicillin starts growing in a Petri dish, or we discover, “Oh, my God, this is great.”

Does AI preserve that kind of serendipity, those kinds of accidents, or does it optimize them away?

Yeah, this is a great question, and the fact of the matter is, we just really don't know yet. This is going to be a really important core question that a lot of people are asking. What's your intuition on that?

I think they probably will—

They probably will—

They probably will preserve it.

My understanding is that, basically, the window was left open on some agar with no antibiotic in it. Obviously, they didn't have antibiotics if this was the discovery of the first one, right? So the window was left open with some agar, and some spores flew onto it and began growing. They observed that the bacteria was inhibited, right?

That's a mistake. Someone screwed up, right? And that mistake led to something fantastic. You will have mistakes. I think that will be preserved.

But in the meantime, scientists should always leave their windows open. You never know what's going to happen.

Seriously, though, when you get first-year graduate students in academia, they have no idea what to do. And that is a huge source of scientific progress, because they just do the most random, kooky stuff that no one who knows anything would ever think to do. And it's actually really important.

You almost want your AI scientist model to hallucinate a little bit, so that it doesn't lose that—

Add noise, right? We talk about this as just adding noise. This is actually important for biological evolution also, right? The genome has a lot of noise, and that's how evolution randomly comes up with new stuff.

There's a protein that's just totally random and doesn't do anything. Then one day, all of a sudden, oops, it does something, and that's great, right?

What do you make of the leaders of the big AI labs, people like Demis and Dario and Sam Altman, who are saying AI is going to allow us to cure all diseases, or most diseases, within the next decade or two?

A decade is crazy. I think—and I'm happy to take a very strong stance on this, because if I'm wrong, it's a great thing, right? If I'm wrong, everyone wins. But a decade is crazy.

Why is it crazy?

Because of the reason we were talking about before: You have to run clinical trials, right? If we had a drug right now that prevented aging—completely halted aging in humans between the ages of 25 and 65 or something—you would not know for 10 years, because you can't detect in humans in that age range whether or not they're aging for at least 5 or 10 years. You don't detect from 1 year to the next that you're aging. So you won't know if the thing is working.

I don't know. Some people at my 10-year high school reunion were already looking pretty bad.

Hate to say it.

I did say 25. Twenty-five. Fair enough. Fair enough.

But, right, we have to conduct experiments. Those experiments will take time. Now, 30 years, I think, is very plausible. We don't know what is going to be possible. We don't know if it's possible to halt aging. We don't know if it's possible to cure all diseases or whatever.

But between now and 30 years from now, I think you should expect to see a humongous leap forward in terms of—

I want to drill in on that a bit, though, because I think some people might hear that and say that this is essentially a regulatory issue, that we just don't have the FDA set up to measure this. I'm curious about the experimental side of it, though, right? Because my understanding is we don't really have enough biologists to run all the experiments that we might. We might not have the funding to fund the experiments. And you did raise the point that some of these experiments actually take a long time to run, right?

So what are all of the factors that, in your mind, are just going to make it so hard to—

You have to go and, even supposing you have a molecule that you want to test in a human and you know which humans you want to test it in, you have to go and make it, right? Humans are big. They require a lot of it. You have to make sure it's high enough grade that you can actually put it into a human.

You have to find the patients, which means forming relationships with doctors and actually waiting until you have enough patients who are willing to do it. For many diseases, there just aren't that many patients, and so finding the patients is hard, right? Then you actually have to dose them. You have to wait and see what happens, right?

Even with no regulation, it would be slow.

There's no AI shortcut for almost any of that, at least not right now.

No. What AI will allow us to do is discover a lot of things where we already have the information to discover them. We just haven't figured that out yet.

The other thing that AI researchers sometimes talk about, which is probably not reasonable, is that you should not expect that you're one day going to get GPT-7 and just ask it how to cure Alzheimer's and it will just tell you.

My expectation is that there is not enough knowledge, right? We do not have enough knowledge to solve it in principle, even with infinite intelligence. With infinite intelligence, there would still be some things that are just not known about the world, where we have to conduct the experiments to see.

You'll be able to plan the best possible experiment given everything that's known. But you will not just be able to de novo figure it out, right?

Casey, I took Latin. That means “from new.”

Oh, thank you. Thank you. That's saved me a step. This isn't quite science per se, but I'm curious what you make of this, Sam. All of the big AI labs are obsessed with math.

With winning the International Math Olympiad, with putting up a gold-medal score, with solving these unproven math theorems. And I have a take about this, which is that I believe this is because these labs are filled with people who were themselves competitive math athletes in high school, took part in the IMO, and did pretty well.

And a lot of those people think that AGI will just sort of be a slightly smarter version of them. But I'm curious: Why are these places so obsessed with math as being one of the first places that they want to make a lot of progress?

There are 2 reasons. I think one of the reasons is exactly what you just said. It's just familiar, right? But the other reason is that you can measure progress, right?

Ultimately, what drives progress in machine learning—a big part of what drives progress—is benchmarks. With math, you can tell whether or not your proof is right, and there's kind of an infinite number of things to go improve. So it's just really easy to tell whether or not you're getting better. Things like the IMO just present great opportunities.

By contrast, if you look at some of the biggest breakthroughs recently—the biggest breakthroughs this year in AI for biology—things like Chai Discovery and Nabla Bio coming up with these extremely good models for producing antibodies de novo, right? Huge breakthrough, but ultimately the win for them is going to be when it's approved in a human, and that might be another 5 years or something, right?

Arc Institute putting out, like, the first time anyone has designed an organism from scratch.

They designed a bacteriophage. It’s a kind of virus that infects bacteria. Incredible, right? But it’s just harder to evaluate: How good is it? You’re not going to release it into the wild, and so on. It’s harder to evaluate, whereas the IMO is just super clean. And so I think that’s one thing that we think about a lot: How do we get really clear benchmarks that we can pursue to measure whether or not we’re doing a good job at science?

I have an answer here: International Cancer Curing Olympiad.

I like that.

Should we start this? We can give people a medal if they win.

Let’s get on it, labs.

So when the CEOs or the leaders of these companies make statements about how we’re going to cure all disease using AI in the next 10 or 15 years, or whatever timeline they give, are they doing that because they don’t understand the bottlenecks? These are very smart people. So what are they not seeing, or are they just doing this as a marketing exercise? Is this an attempt to get people excited about AI who might otherwise be freaked out about it? Why are they giving these projections?

No, look, I think reasonable people could disagree. There are lots of reasons why you could argue that the models will get super smart and figure out ways to measure whether or not we’re making progress before you run a clinical trial, and that will increase the iteration cycle. There are reasonable arguments to be made about that: that we’re just not going to do full clinical trials anymore. We’ll just use biomarkers. That’s not crazy, and that’s one way that I could be wrong and maybe in 10 years we do have cures for all diseases.

That’s part of it. Obviously, there’s part of it where they want to hype the thing. Part of it is, does Sam Altman really intimately understand what it takes to go and manufacture—scale up manufacturing for—a small molecule to put into the clinic? Probably not. So there’s a mixture. I don’t think any of it’s in bad faith. It’s just that people are very excited.

There will be a little bit of a collision with reality at some point. We’re going to see exactly where that is. But regardless, the future is going to be awesome, right?

At this moment in 2025, how much do you think AI tools have changed the life of a working scientist? And how different do you expect that will be a year from now?

I think you’d be shocked by the extent to which they have not yet. Scientists in general are extremely conservative people because you never know. If you’re running an experiment, you never actually fully know what is in biology, at least. You usually do not fully understand why the experiment works and why it doesn’t.

There are some things that you’ve inherited from protocols that you’ve run in the past, where it’s like, “We do it this way.” You could go and test it, but there are way too many things to test. So you’re just kind of locked in on your methods. It’s what works, and you just want to do what works. For that reason, biologists just adopt new methods slowly.

I think most labs around the world are still probably doing science the way they’ve done it before and probably will continue to do so for a while. And that’s okay. One place where I think a lot of people are already adopting it is coding, because historically, coding has been a big bottleneck in biology. It’s a huge unlock now that biologists who didn’t know how to code can do a lot of coding using Claude Code, OpenAI’s models, Gemini, and so on. That’s a huge unlock, and I think that’s going to see a lot of adoption quickly.

Literature search is another one. Being able to parse the immensity of the scientific literature is a huge unlock. That’s going to get adopted very quickly. The tools like what we’re building are a little more frontier. We also build literature-search agents that people use, but in general, the tools are a little more frontier. It might take a little longer to adopt, but that’s okay. Ultimately, people adopt them when they see other people using them and getting great results.

Sam, can we play a little lightning-round game here with you? We’re calling this one “Overhyped, Underhyped.” We’ll tell you something, and you tell us whether, in your scientific opinion, it’s overhyped or underhyped. Ready?

Yeah.

Vibe proving. This is when AI systems go out and write math proofs.

Probably, if I had to choose, overhyped. It’s great as a progress driver in AI, and being good at it will probably have implications elsewhere. But is it itself that useful? I’m not sure.

Robotics for AI lab automation.

Robotics for automating AI labs, or—

Yes, or for automating scientific labs.

Robotics for automating scientific labs: I think appropriately hyped. It is going to be totally transformative. The technology is not at all there yet. There’s a lot that we need to do, but, yeah, probably appropriately hyped.

AlphaFold 3.

I think that’s an interesting one. I would say probably underhyped, in that I think all of the protein-structure models—there’s a lot of hype around them—but they’re still probably going to be extremely transformative. So maybe I would say probably underhyped. It’s hard. There’s a lot of hype around it, though, so it’s a hard decision to make.

Virtual cells, like we heard from Patrick Collison this summer about what the Arc Institute has done in making a virtual cell.

This is overhyped, but for a specific reason. The models that they’re building at Arc are awesome. They’re doing similar things at NewLimit, Chan Zuckerberg, and many other great companies and organizations. I think calling it a virtual cell is a little overhyped, right? Ultimately, that kind of model models something very specific. Actually building a true virtual cell—being able to simulate a cell in a computer—is an amazing goal. We are very far away from that.

Quantum computing.

Overhyped.

Brain-computer interfaces.

I’m also—oh, man, this one’s really hard. I’m going to say overhyped. I’m a huge believer in BCIs. I think effective BCIs, the way that we imagine them in science fiction, are further out than people imagine. Even Neuralink is making amazing progress.

Yeah, Casey’s got one in his head right now.

It’s on the fritz.

So we’re nearing the end of the year. If we can put you in a bit of a reflective mode, what do you think were the top 3 AI-driven scientific advancements this year?

I think the first one is—this year has been the year of agents. This was the year when people discovered agents. In good faith, I have to put us on that list, also with Google’s AI co-scientist. We’re not the only people who are working on this. Google has been doing a great job, and there are a bunch of other people.

AI agents for science, definitely. And then generative design is just having a huge moment. The other ones would probably be the work that Chai has been doing, the work that Nabla Bio has been doing, and many others on de novo antibody design.

I’m really glad you defined de novo earlier in the broadcast, by the way. It’s come up a lot.

Yes. Sorry. When I say de novo, I just mean that it literally generates it from scratch. You don’t give it anything, or you give it a target that you want it to bind to, and it generates it from scratch. This is huge because the promise that companies like Chai, Nabla, and so on are going after is a world in which you can say, “We know to cure this disease, we have to target that protein.” You click a button, and you have an antibody that you can go and put in humans tomorrow, right? That’s huge.

Or, as I was mentioning before, manufacturing is a pain in the ass. You have to go and manufacture it. But it’s huge. It cuts out an enormous amount of what people had to do previously. So that’s a huge one.

And the third one, I just think, is what Brian Hie, Patrick Hsu, and so on at the Arc Institute have done with generating organisms from scratch.

We know what it means now.

This is our Pee-wee’s Playhouse word of the week this week.

The de novo design of organisms. Is it useful? I don’t know. Is it awesome? Absolutely. It’s such a big breakthrough.

And Sam, what should we be watching for next year? What are you excited about that may be coming down the pipe for 2026?

Honestly, you’re going to see an explosion in agents. Again, it’s going to be the agents that see an explosion. We are right now at the beginning of that S-curve, and that is going to continue, right? I was telling people back maybe a year ago that I thought in 2026, or maybe 2027, the majority of the high-quality hypotheses generated by the scientific community would be generated by us or by agents that are like the ones we’re building.

When I said it in 2024, I thought I was overhyping it, right? But I was just like, it needs some hype. At this point, it may be real.

I think 2026 would be ambitious for that. That's a huge leap—for the majority of the good hypotheses that come out to be made by agents, that's a huge leap—but 2027? Yeah, man. I mean, 2026 is going to be the year when we see these agents start to infiltrate everything: infiltrate labs, infiltrate people's normal lives. I mean, it's already happening.

Cool, yeah.

Well, I look forward to it. Sam, thank you so much for giving us the science education that we clearly didn't get in school.

Yeah, you've really given us some de novo things to think about. I appreciate that.

Good. Thank you guys.

Thank you.