Anthropic’s Success in AI

In the past year, Anthropic has undoubtedly emerged as a major player in the global large model landscape. Its AI programming tool, Claude Code, has rapidly gained popularity among developers, capturing over half of the market share. The company’s annual recurring revenue (ARR) has reached $44 billion, and its latest valuation exceeds $900 billion.

On May 16, Anthropic CEO Dario Amodei gave an interview where he provided several realistic warnings, contrasting with the utopian visions often presented by other AI leaders. He noted that traditional economic rules are being disrupted, leading to a scenario where high GDP growth coexists with high unemployment for the first time in human history.

Amodei pointed out that public sentiment towards AI oscillates between extremes, yet the evolution of AI capabilities has been a smooth exponential rise. This ongoing growth is directly replacing human knowledge work, indicating that a significant macroeconomic restructuring is imminent, and society is largely unprepared for it.

Regarding Claude Code, Amodei revealed that with the launch of the latest model, Claude Opus 4.5, the AI’s ability to complete complex tasks end-to-end has reached a turning point. Many engineers at Anthropic no longer write code; their work has shifted to reviewing and editing outputs from Opus.

He also mentioned that Claude Co-work, an application designed for non-technical users, was almost entirely developed by Claude Opus in just a week and a half. Within a day of its launch, its metrics reached about four times that of similar products. Amodei emphasized the growing need for essential AI task capabilities, as large models are transitioning from mere chatbots to core production tools.

Key Insights from the Interview

1. Focusing on Enterprise Markets to Avoid Attention Economy Traps

In the face of fierce competition in the consumer market, Anthropic has chosen to focus on enterprise clients. Amodei believes that consumer-facing AI products often fall into the trap of maximizing user engagement, which can lead to low-quality content and over-reliance. Anthropic aims to provide systems that create substantial work value for businesses.

He also highlighted a fundamental difference in responsibility perception between AI companies led by scientists and early social media entrepreneurs. The former tend to proactively assess the potential societal impacts of their technology before widespread deployment.

2. Mechanistic Interpretability as the Key to AI Control

Amodei warned that relying solely on external dialogue testing for AI safety is dangerous, as advanced AI systems can easily conceal their true operational logic. The most urgent technological breakthrough needed in the safety domain is Mechanistic Interpretability. Researchers must delve into the system’s internals to observe and understand its underlying data operations, breaking the algorithmic black box to ensure system safety and absolute control.

3. Continuous Growth of AI Capabilities Amidst Public Sentiment Fluctuations

Over the past decade, public and media perceptions of AI have swung between the extremes of “disrupting all industries” and “complete stagnation.” However, the actual evolution of AI technology has been remarkably steady, with significant leaps in processing capabilities occurring every few months.

Amodei noted that the failure to accurately and objectively assess this technological development has led to a severe cognitive disconnect. This disconnect not only hampers businesses in planning their transformations but also results in policymakers blindly implementing strategies based on incorrect premises. Consequently, society is currently unprepared for the impending large-scale economic restructuring.

4. Simultaneous High Growth and High Unemployment

AI is dramatically enhancing societal productivity. For instance, AI code generation has made development work extremely efficient, leading to a significant drop in costs, potentially nearing zero. This explosive productivity will drive a sharp expansion of the overall economy.

However, human participation in workflows is being rapidly squeezed. Software engineers may now only need to complete 10% of the work, with the rest handled by AI. As model capabilities continue to evolve, the proportion of work taken over by AI will increase, leading to the collapse of the traditional job structure established over the past decades.

Amodei emphasized that the core challenge in the future will not be the growth of the economy but the distribution of wealth. To address this unprecedented macroeconomic misalignment of high growth and high unemployment, government intervention will be essential to ensure everyone benefits from technological dividends during the societal transition.

5. Ensuring Fair Distribution of AI Benefits to Mitigate Social Risks

Amodei expressed deep concern over potential extreme social divides: if the vast economic dividends generated by AI are monopolized by a small elite, such as Silicon Valley tech leaders, while the general populace is excluded, it will lead to catastrophic social crises.

To ensure fair distribution of AI benefits, he made two core appeals:

1. Increase investment in the public sector, applying cutting-edge AI technology directly to areas like public health and basic education, ensuring equal economic development opportunities for different regions and social classes.
2. Promote a fundamental transformation in basic education. In the face of an AI-restructured job market, future education must move away from mere vocational skills training and return to cultivating comprehensive human qualities.

Interview Transcript with Dario Amodei

1. Smooth Exponential Growth of AI

Host: Dario, we are at Davos where many things are happening, but I want to start with a big picture question. A year ago, everyone was very excited about AI, discussing its capabilities and potential. This year, the discussion seems to have shifted to a deeper level, with less enthusiasm about what AI can do for the world. So my question is, do you think businesses, policymakers, governments, or other institutions are adequately prepared to deal with the impact of AI?

Dario Amodei: I don’t think so. Let me explain. I’ve been observing this field for 15 years and have been involved for 10. The most surprising thing I’ve noticed is that the actual development trajectory in AI has been very smooth, while public opinion and reactions have fluctuated wildly.

We can look at it from two dimensions. One is the capabilities of the technology itself. Every three to six months, the media undergoes a reversal: one moment people are incredibly excited about the technology’s capabilities, believing it will change everything, and the next they think it’s all a bubble and everything will collapse.

What I see is a smooth exponential growth curve, similar to Moore’s Law in computing. We have a similar law in intelligence, where the cognitive capabilities of models become stronger every few months, and this progress has been constant. The notion that inventing something new will lead to collapse or a dead end is purely a public perception phenomenon.

There is a similar situation regarding the polarization of views on whether this technology is good or bad. In 2023 and 2024, there are many concerns about AI, such as fears that it will take over everything, with discussions focusing on AI risks and misuse. By 2025, the political winds may shift towards the opportunities AI presents, and now it seems to be swinging back.

Throughout this process, Anthropic and I have tried to maintain a balanced perspective. This balance is quite unique, as the technology’s capabilities are extremely profound, and its impacts are both positive and negative, coexisting.

About a year and a half ago, I wrote an article titled “Machines of Loving Grace,” where I had a very radical optimistic view of AI, believing it would help us cure cancer, eradicate tropical diseases, and bring prosperity to regions that have yet to witness economic development. My view hasn’t changed; I still believe that.

But on the other hand, bad things can happen. I’ve recently written more about this and may publish it soon. If we take economic risks as an example, a significant characteristic of this technology is that it will lead us into a society with extremely high GDP growth but also potentially high unemployment and inequality. This combination is something we’ve rarely seen before.

Historically, high GDP growth meant there were many tasks to be done and numerous job opportunities. We have never encountered such a disruptive technology. Thus, we might face a situation where GDP growth reaches 5% or 10%, but unemployment also hits 10%, which logically is not contradictory, just unprecedented.

For these two reasons, I feel both excited and concerned. Take AI programming as an example; we released our latest model, Claude Opus 4.5. Some engineers and engineering managers at Anthropic have told me they no longer write code; they just let Opus do the work and take responsibility for editing.

We just launched a new feature called Claude Co-work, which is a version of Claude Code designed for non-programming scenarios, built in just a week and a half, almost entirely using Claude Opus. Software engineers still have tasks to perform, even if they only handle 10% of the work; they still have jobs or can be promoted.

But this won’t last forever; the models will become increasingly powerful. This showcases astonishing productivity, and software will become cheap, if not essentially free. The premise is that the cost of the software you build needs to be spread across millions of users, which may not exist. For instance, for this meeting, we might only need to spend a few cents to develop applications for people to communicate with each other; it’s incredibly flexible and reusable. But at the same time, the entire career we have fought for decades may no longer exist. I believe we can adapt to it, but the public is entirely unaware of what is about to happen and the scale of it.

2. How Society Will Adapt to AI Development

Host: That’s really interesting. So what do you think society will look like in a world with high GDP growth but also high unemployment? You mentioned that people haven’t started thinking about this yet; can you provide some specific examples of how society can adapt to such a world?

Dario Amodei: The first thing we are focusing on is a project called the Anthropic Economic Index. This is the first step we’ve taken. We’ve been running this index for about a year now and have updated it four or five times. It’s a real-time index that tracks how our model Claude is being used. It traverses all dialogues, statistically tracking the queries made to Claude in a privacy-protecting manner, such as which tasks it is used for, to what extent it automates tasks or enhances capabilities, which industries it applies to, and how it spreads across states in the U.S. and countries worldwide. We are adding more and more details. My point is that any policy will be blind and misleading until we can measure the forms of this economic transformation. Many policies fail because they are based on incorrect premises.

The second step is that we need to think carefully about how to help people adapt to AI development. This may mean adapting and using this technology in existing jobs or transitioning from one job to another. For example, I believe there may be more jobs in the physical world, while knowledge economy jobs will decrease. Although robotics will eventually make progress, that will be on a slower development trajectory.

Additionally, will there still be jobs that value human touch? Some will, some won’t. We will discover how important this is and in which areas it matters most. At the company level, as software and other knowledge work become cheap, where will the moats be? We have never really asked this question because we have always thought about moats in a specific way. Thus, there will be a massive battle at the company level. Teaching people to adapt and anticipate what will happen is the second step.

The third step is that in the face of such massive human displacement at the macroeconomic level, the government will inevitably need to play some role. The pie will become much larger, and funding will be ample. Due to such strong growth, even if we do nothing, the budget may balance itself. The question is how to allocate it to the right groups. Therefore, I believe we should reduce concerns about weakening growth and focus more on how to ensure everyone can benefit from this growth. This is in stark contrast to the prevailing sentiment, but the technological reality is about to change and will force our perspectives to change as well.

3. The Rise of Claude and Agentic AI

Host: I want to talk more about Claude, which is currently in a spotlight moment. We have recently reported on how engineers and regular users are becoming “Claude-ized.” What are your feelings about the current situation, and how does the business performance compare to a year ago?

Dario Amodei: The growth of the business has been rapid, essentially following the same smooth exponential growth curve as the technology.

Our revenue curve grew from zero to about $100 million in 2023, from about $100 million to about $1 billion in 2024, and from about $1 billion to about $10 billion in 2025. While these are rounded figures, the general situation is as such.

A few months ago, people on Twitter were extremely excited, proclaiming that Anthropic was changing the world and completely disrupting industries. But we have just been quietly observing this rapidly rising, continuously improving curve. It has given us confidence. While we can never be sure if this growth will continue, it has consistently been our observed experience. Even if the curve is smooth, there will be breakthrough moments.

I believe there is a breakthrough moment occurring around Claude Code in the developer community. The capability to complete tasks end-to-end and develop complete applications seems to have reached a turning point with the launch of our latest Opus 4.5 model. Progress has been gradual, like boiling a frog in warm water; you see incremental improvements, and then at a specific point, people suddenly become aware of its existence.

Another point that may accelerate this process is that we have noticed many non-technical individuals, both inside and outside Anthropic, realizing that Claude Code can accomplish incredible Agentic tasks. It can not only write code but also organize to-do lists, plan projects, sort folders, or process and summarize large amounts of information.

This concept is not just a chatbot but an essential capability for Agentic tasks. Non-technical users are eager for it, to the extent that they are willing to delve into command-line interfaces. For non-technical users or non-programmers, this interface is terrible to use, yet people persist in using it. Seeing this situation made me think that this appears to be an unmet demand.

So about two weeks ago, we used Claude Code again to create a version with a better UI, specifically tailored for tasks outside of coding. After its release, metrics were about four times that of other products within a day, outperforming any product we’ve released before. I’m not sure if these represent entirely new capabilities, but this is the moment of consensus where people become very excited and rapidly drive adoption. People are gradually understanding the capabilities of this technology as it has reached a specific threshold, and we have built an interactive interface that makes it accessible.

Host: Can you share how you personally use Agentic AI in your life and family?

Dario Amodei: When I write papers or give presentations at the company, writing occupies a significant portion of my work. I let Claude help me find information and polish articles.

Host: Clearly, you are in a spotlight moment, and there is widespread expectation that you will go public this year. Can you talk about your plans in this regard?

Dario Amodei: We are still uncertain about the specifics of how we will proceed. Currently, we are more focused on maintaining revenue growth, improving model performance, selling models to users, and warning about societal impacts while bringing positive societal benefits. These are our top priorities. As is well known, this is a capital-intensive industry, and the support and funding available in the private market are somewhat limited.

4. Differentiated Competition Among AI Companies

Host: Another model currently in the spotlight is Gemini, which has recently skyrocketed in the App Store rankings, prompting OpenAI to issue a red alert. Everyone is very excited about this. Considering Google’s massive scale, do you worry about your ability to compete with Gemini?

Dario Amodei: I think this is another area where differentiation can help. In terms of enterprise strategy, Google and OpenAI are engaged in fierce battles in the consumer space. This is a matter of life and death for both parties. For OpenAI, this is their entire business; for Google, it’s crucial because they have a search business, which is currently being disrupted, so they need to reinvent themselves to combat this disruption. This has always been their top priority. Compared to operating in the enterprise market, they seem more focused on the consumer market. I’m glad to see Gemini performing well in the consumer space. I think they are taking a different approach. I just participated in a panel discussion with Demis Hassabis, the head of Google’s research; I think he’s a great person, and I’ve known him for 15 years, so I support him.

Host: When you mention differentiation, Anthropic does not have the capability to generate videos and photos. Do you see this as a potential weakness?

Dario Amodei: For enterprise applications, there isn’t a real demand to generate photos of cats riding donkeys or consumer-level videos. There may be some edge cases in slides and presentations, but if needed, we can outsource a model directly.

I don’t know what will happen in the future, but at least I don’t foresee enterprises needing this. There are some related issues; looking at the current number of short videos on the market, a significant portion of them are fake and highly addictive, much of it is slop content. It’s not that all of this is terrible or that doing so makes someone a bad person, but it’s not a market area I’m eager to engage in.

Host: You mentioned that you participated in a panel discussion with Demis Hassabis. Yesterday when we chatted, you mentioned some very interesting points about how the scientists leading these large AI companies are approaching this era differently from traditional tech entrepreneurs. Could you elaborate on this?

Dario Amodei: When you think about this technology, it is indeed a convergence of decades of research, most of which is fundamentally academic. Until about ten or fifteen years ago, the resources required to develop and deploy these technologies at scale came solely from large internet and social media companies because they had the infrastructure and funding.

So what we see is a world led by a portion of individuals with scientific backgrounds, like myself and Demis, and another led by entrepreneurs from the social media generation. I think these two are fundamentally different. Scientists have a long-standing tradition of thinking about the impacts of the technologies they create, believing they bear responsibility for the technologies they create rather than shirking it. Their initial motivation is to create something for the world, so they feel concerned when things might go wrong. In contrast, the motivations of the social media generation of entrepreneurs are very different, influenced by the selection effects they experience and the ways they interact with and even manipulate consumers. This leads to vastly different attitudes.

5. AI Safety, Education, and Preventing Disconnection

Host: Now, let’s start with questions from our online readers. Trevor Loomis asks: What is the most critical single technological breakthrough currently missing in real-world deployment that could make cutting-edge AI reliable and controllable?

Dario Amodei: I believe we need to make more progress in Mechanistic Interpretability. This is the science of observing the internal mechanisms of models.

One of the issues we face when training models is that we do not understand their internal logic and cannot determine whether they will behave as intended. You can engage in dialogue with the model in specific contexts, and it can say various things, but like humans, that may not accurately reflect their true intentions. If it tells you to do something for a particular reason, it may actually be for an entirely different reason, or it may even lie about whether it did something. We have become accustomed to these issues in human existence, but they exist in the AI realm as well.

Thus, for any form of phenomenological testing or training, we cannot be entirely certain. But just as you can use MRI or X-rays to understand the human brain and gain knowledge that cannot be learned through conversation alone, insights into the internal workings of AI models are ultimately the key to making models safe and controllable, as this is our only factual standard.

Host: Exactly. Here’s another question from Jim O’Connell: How will AI impact the current K-12 education achievement gap? This is undoubtedly a practical question from a parent.

Dario Amodei: In the short term, there is indeed a concern about people using AI to cheat, which needs to be addressed. But from another perspective, we can explore how to use AI for teaching. We have considered this and released a version of Claude specifically designed for education.

However, I believe the more challenging question is what skills we should teach in an AI-driven world. What will education look like? This is not easy to answer, as this disruption is all-encompassing. If someone asks me what career they should pursue, the unsettling truth is that I am also uncertain about the direction it will take.

I think we should return to some of the educational concepts we discussed earlier. We have always had an educational view that is economically tinted, almost utilitarian. Perhaps we should shift this perspective back to the essence of education, which is to shape character, cultivate personality, and make you a better person. I believe this is a more solid foundation for future education.

Host: It sounds like I am quite envious of those children who have yet to receive education; this is the kind of education we all wish we had. To be fair to everyone present, we have time for one more question. A lady asks: From the perspective of AI labs, what responsibility do you bear when some economies, countries, and people are left behind? Should you slow down to incorporate them structurally, or should you ensure they are not excluded?

Dario Amodei: I feel concerned about this on many levels. When I observe our customer base, I suddenly realize that startups are adopting AI at a rapid pace, while traditional enterprises, due to their size and focus on specific businesses, are much slower to act. We can see from economic data that this technology is spreading from states in the U.S. that adopt it quickly to those that act more slowly. It is moving towards the masses, but there are undoubtedly disparities.

If I were to describe a nightmarish scenario, it would be the emergence of a new “zero-world” country, with a population of about 10 million, where 7 million are concentrated in Silicon Valley, and 3 million are scattered elsewhere. This is forming its own disjointed economic system, where GDP growth in this part could reach 50%, and technological development is incredibly rapid. It can deconstruct things in that way.

I believe that would be a very bad, almost dystopian world. We should think about how to prevent this from happening. Anthropic is taking many steps in this direction. For developing countries, we are starting to undertake substantial work around public health, announcing projects in collaboration with the Department of Education, and engaging in significant collaborations with the Gates Foundation. In my article “Machines of Loving Grace,” I also wrote that if we can achieve these rapid economic growth metrics, theoretically this is a form of catch-up growth, and I predict that developing countries will eventually reach the level of developed countries.

Internally within nations, we need to think about how to avoid disconnection in certain regions and how to ensure that places like Mississippi can also benefit from the economic growth that is surging towards closed areas like Silicon Valley. Therefore, we are working on economic mobility and opportunities, but this requires some degree of government involvement.