Dario Amodei

I've made this series of predictions on code and software engineering, and I think people have repeatedly kind of misunderstood them. So let me lay out the spectrum, right? Eight or nine months ago or something, I said, you know, the AI model will be writing 90 % of the lines of code in like three to six months, which happened at least at some places, right? Happened at Anthropic, happened with many people downstream using our models. But that's actually a very weak criterion, right? People thought I was saying like, we won't need 90 % of the software engineers. Those things are worlds apart, right? Like I would put the spectrum as 90 % of code is written by the model.

100 % of code is written by the model. And that's a big difference in productivity. 90 % of the end -to -end SWE tasks, right? Including things like compiling, including things like setting up clusters and environments. 90 % of the SWE tasks are written by the models. 100 % of today's SWE tasks are written by the models. And even when that happens, it doesn't mean software engineers are out of a job. There's like new higher level things they can do where they can manage. And then there's a further down the spectrum, like there's 90 % less demand for SWE, which I think will happen. But like, this is a spectrum.

And you know, I wrote about it in the adolescence of technology where I went through this kind of spectrum with farming. It's just, these are very different benchmarks from each other, but we're proceeding through them super fast.

I think people are very attached to the idea that they want to believe there's some fundamental wall. There's something different, something that can't be done. It kind of reminds me of... You think it's a coping mechanism deep down? Yeah, you know what it reminds me of? So you know the 19th century notion of vitalism? You know, this was the idea that, you know, the human body and, you know, organisms that are alive are made of a fundamentally different material than inanimate matter, which of course we know scientifically now is not true. But it's something people very much want to believe. And your common sense seems to suggest it.

Like I'm not very much like a table. I've made of very different materials than metal or glass or whatever. But, you know, when we actually go down to the fundamental units, of course, we're all made of the same thing. But you think people now have this kind of modern concept of vitalism in whatever the fundamental humanity is. And they're saying, oh, you know, models can't do that. I think there's some tendency to believe it. And I think as with vitalism, the way around it is to, you know, is to recognize that a mind is a mind, no matter what it's made of. The notion of the dignity or the specialness of cognition or sentience.

It's not that it isn't special. It's that it can be made out of anything.

What I would say is that we are about to be in a world where growth and economic value will come very easily, right, if we're able to build these powerful AI models. What will not come easily is distribution of benefits, distribution of wealth, political freedom. You know, these are the things that are going to be hard to achieve. And so when I think about policy, I think the technology in the market will deliver all the benefits. That's kind of my fundamental belief. The technology in the market will deliver all the fundamental benefits almost faster than we can take them. And that these questions about distribution and political freedom and rights are the ones that will actually matter and that policy should focus on.

On one hand, you get people who say, we should stop building AI. We should have a global worldwide pause. There's this other extreme position. It's like we have to build as fast as possible. The most extreme people say things like, it doesn't matter if humanity's wiped out. AIs are the future. I mean, that's a really extreme position. The position of someone who's kind of trying to be thoughtful about this, trying to build, but build carefully. If you feel like you have to make those people happy, what can end up is either you get polarized on one side or another. Then you kind of repeat all the slogans of that side and you become a lot dumber than you would otherwise.

Who's yelling at me on Twitter? Who thinks I'm great on Twitter? You're not going to arrive at the position that makes everyone boo. That might just be the correct position.

If you're off by only a year, you destroy yourselves. That's the balance. I think it is true we're spending somewhat less than some of the other players. It's actually the other things, like, have we been thoughtful about it? Or are we YOLOing and saying, oh, we're gonna do $100 billion here, $100 billion there. I kind of get the impression that some of the other companies have not written down the spreadsheet, that they don't really understand the risks they're taking. They're just kind of doing stuff because it sounds cool. We're buying a lot. We're buying a hell of a lot. Like, we're buying an amount that's comparable to that, that the biggest players in the game are buying.

We've thought carefully about it, right? We're an enterprise business. Therefore, we can rely more on revenue. It's less fickle than consumer. We have better margins, which is the buffer between buying too much and buying too little. But if you're asking me, why haven't we signed 10 trillion of compute starting in mid -2027? First of all, it can't be produced. There isn't that much in the world. But second, what if the country of geniuses comes, but it comes in mid -2028 instead of mid -2027? You go bankrupt. You go bankrupt.

You've had a few skeptical things to say about Donald Trump and his trustworthiness as a political actor. What about the domestic landscape, whether it's Trump or someone else? You are building a tremendously powerful technology. What is the safeguard there to prevent essentially AI becoming a tool of authoritarian takeover inside a democratic context? Yeah, I mean, look, look, just to be clear, you know, I think the attitude we've taken as a company is very much to be about policies and not the politics, right? That, you know, the company is not going to, you know, say Donald Trump is great or Donald Trump is terrible. Like, you know, but it doesn't have to be Trump.

It is easy to imagine a hypothetical U.S. president who wants to use your technology. Absolutely. And for example, that's one reason why I'm, you know, I'm worried about the, you know, the, the, the autonomous drone swarm, right? So, you know, the constitutional protections in our military structures depend on the idea that there are humans who would, we hope, disobey illegal orders. With fully autonomous weapons, we don't necessarily have those protections. But I, I actually think this whole idea of constitutional rights and liberty along many different dimensions, you know, can be undermined by AI if we don't update these, these protections appropriately. So, you know, think about the fourth amendment.

It is not illegal to, you know, put cameras around everywhere in public space and, you know, record every conversation. It's a public space. You don't have a right to privacy in a public space. But, but today the, the government couldn't record that all and make sense of it. With AI, the ability to transcribe speech, to look through it, correlate it all, you could say, oh, there's this, you know, this person is a member of the opposition. This person is expressing this view and, and make a map of all, you know, a hundred million. And so are you going to make a mockery of the fourth amendment by, by the technology finding kind of technical ways around it?

And, and, and so, you know, again, if we, if we had the time and we should do this, we should try to do this, even, even if we don't have the time, is there some way of reconceptualizing constitutional rights and liberties in the age of AI? Like, you know, we don't need to write a new constitutional, but you know, But you have to do this very fast. Do we expand the meaning of the fourth amendment? Do we expand the meaning of the first amendment? Right.

I don't think in the history of startups, there has been a single application that has been as hotly competed in as coding agents. And CloudCode is a category leader here. And that seems surprising to me. Like it doesn't seem intrinsically like Anthropic had to build this. And I wonder if you have an accounting of why it had to be Anthropic or how Anthropic ended up building an application in addition to the model underlying it that was successful. Yeah, so it actually happened in a pretty simple way. We had our coding models, which were good at coding. And, you know, around the beginning of 2025, I said, I think the time has come where you can have non -trivial acceleration of your own research if you're an AI company by using these models.

And of course, you know, you need an interface. You need a harness to use them. And so I encourage people internally. You know, I didn't say this is one thing that, you know, that you have to use. I just said people should experiment with this. And then, you know, this thing, I think it might have been originally called CloudCLI. And then the name eventually got changed to CloudCode. Internally was the thing that kind of everyone was using. And it was seeing fast internal adoption. And I looked at it and I said, probably we should launch this externally, right? It's seen such fast adoption within Anthropic, like coding is a lot of what we do.

And so, you know, we have a, we have a audience of many, many hundreds of people that's in some ways, at least representative of the external audience. So it looks like we already have product market fit. Let's launch this thing. And I think just the fact that we ourselves are kind of developing the model and we ourselves know what we most need to use the model. I think it's kind of creating this feedback loop.

Within 10 years, we'll get to what I call kind of country of geniuses in a data center. I'm at like 90 % on that. I mean, it's hard to go much higher than 90 % because the world is so unpredictable. Maybe the irreducible uncertainty would be if we were at 95 % where you get to things like, I don't know, maybe multiple companies have kind of internal turmoil and nothing happens and then Taiwan gets invaded and like all the fabs get blown up by missiles. You could construct a scenario where there's like a 5 % chance that things get delayed for 10 years. That's maybe 5%.

There's another 5%, which is that I'm very confident on tasks that can be verified. So I think with coding, I mean, I think we'll be there in one or two years. There's no way we will not be there in 10 years in terms of being able to do it end -to -end coding. My one little bit, the one little bit of fundamental uncertainty, even on long timescales, is this thing about tasks that aren't verifiable, like planning a mission to Mars, like doing some fundamental scientific discovery like CRISPR, like writing a novel. Hard to verify those tasks. I am almost certain that we have a reliable path to get there, but like if there was a little bit uncertainty, it's there.

Tell me if I’m wrong about this. A world that has multiplying A.I. agents working on behalf of people, millions upon millions, who are being given access to bank accounts, email accounts, passwords and so on, you’re just going to have essentially some kind of misalignment, and a bunch of A.I. are going to decide — “Decide” might be the wrong word, but they’re going to talk themselves into taking down the power grid on the West Coast or something. - Won’t that happen? - Yeah, I think there are definitely going to be things that go wrong, particularly if we go quickly. So I don’t know. This is one area where people have had just very different intuitions.

There are some people in the field who say, look, we program these A.I. models. We make them, like, we just tell them to follow human instructions and they’ll follow human instructions. Your Roomba vacuum cleaner doesn’t go off and start shooting people. Like, why — why is an A.I. system going to do it? That’s one intuition. And some people are so convinced of that. And then the other intuition is like, we basically, we train these things, they’re just going to seek power. It’s like “The Sorcerer’s Apprentice.” How could you possibly imagine that, like, they’re not going to — they’re a new species. How can you imagine that they’re not going to take over.

And my intuition is somewhere in the middle, which is that, look, these — you can’t just have these things do exactly what you want to do. They’re more like growing a biological organism. But there is a science of how to control them. Like, early in our training, these things are often unpredictable. And then we shape them. We address problems one by one. So I have more of a, not a fatalistic view that these things are uncontrollable. This is a complex engineering problem. And I think something will go wrong with someone’s A.I. system, hopefully not ours. Not because it’s an insoluble problem. But, again, and this is the constant challenge because we’re moving so fast.

I don't think this field's going to be a monopoly. All my lawyers never want me to say the word monopoly. But you do get industries in which there are a small number of players. Ordinarily, like the way you get monopolies like Facebook is these kind of network effects. The way you get industries in which there are a small number of players are very high costs of entry. Cloud is like this. You have three, maybe four players within cloud. I think that's the same for AI. Three, maybe four. And the reason is that it's so expensive. It requires so much expertise and so much capital to run a cloud company.

Profits are not astronomical. Margins are not astronomical, but they're not zero. And I think that's what we see on cloud. Cloud is very undifferentiated. Models are more differentiated than cloud. Everyone knows cloud is good at different things than GPT is good at, than Gemini is good at. And it's not just, Claude's good at coding, GPT is good at, you know, math and reasoning. It's more subtle than that. Like models are good at different types of coding. Models have different styles. Like, I think, I think these things are actually, you know, quite different from each other. And so I would expect more differentiation than you see in cloud.

So yeah, freaking Thank you.

Could we create an equilibrium where it becomes infeasible for authoritarian countries to deny their people private use of the benefits of the technology? You know, are there equilibria where we can kind of give everyone in an authoritarian country their own AI model that kind of defends themselves from surveillance and there isn't a way for the authoritarian country to like crack down on this while retaining power? I don't know, that sounds to me like if that went far enough, it would be a reason why authoritarian countries would disintegrate from the inside. But maybe there's a middle world where like there's an equilibrium where if they want to hold on to power, the authoritarians can't deny kind of individualized access to the technology.

But I actually do have a hope for the more radical version, which is, is it possible that the technology might inherently have properties or that by building on it in certain ways, we could create properties that have this kind of dissolving effect on authoritarian structures? Now, we hoped originally, right, we think about back to the beginning of the Obama administration, we thought social media and the internet would have that property and turns out not to. But I don't know, what if we could try again with the knowledge of how many things could go wrong and that this is a different technology? I don't know that it would work, but it's worth a try.

On fully autonomous weapons, the logic as I understand, it is not so much about democratic values It's more that you think the technology is not ready for your models to be used fully autonomously You have a different view of the capabilities of the technology, perhaps, than the Department of War Yeah, you can think of this as a supplier of aircraft who says “this aircraft is not safe to fly above a certain altitude or to turn at a certain speed” We didn't make these systems to be safe if you use them in that particular way They're not suitable for this use case And in that, was there anything specific—your model has been used, was used in Venezuela, it's being used in Iran

Is there anything in these existing use cases that has worried you One thing I want to say is we started with a substantially more limited contract that enabled a more limited number of use cases that I think is in force maybe even until today This was all about negotiating a new contract So we had even a substantially more limited contract for quite a long time and that more limited contract, they didn't run into problems with it on the ground So none of this was driven by “we have concerns right now about what the Department of War was doing”, right? Iran is fine and Venezuela was fine You know, I'm not expressing opinion either way on military policy No, no, I'm just— the use of your, the use of your models, you had no concerns If you provide a service you don't get to decide where that service is used There's this

operational side of military decisions Of course we don’t have any say about that What we do have say over is this question of very broad use cases It's not about any administration It's not about any policy It’s how should this technology be used in general This is about, this is about going forward how should we think about the uses and the governance of this technology And so when we talked about fully autonomous weapons, I said there were two objections The first one I said was reliability The second one I said was what I would call oversight There are norms about serving in the military You're supposed to follow orders, but, if something crazy

enough happened, the soldiers would say “I'm not going to do that” What if you have an army of 10m drones instead of 10m human soldiers? What are the norms of the AI-driven drones? I think if we handle this wrongly you could have a situation where there's a very small number of people or one person who has their hand on the button, and we need to answer these questions because we don't want to make companies more powerful than the government But we also don't want to make government so powerful that it can't be stopped We have both problems at once