Exclusive interview with Anthropic CEO: The new features of Claude are almost entirely developed autonomously by AI, and software is entering the free era.

In the past year, Anthropic has undoubtedly been the biggest winner in the global large model field.

Its AI programming tool, Claude Code, has rapidly gained popularity among developers, capturing more than half of the market share. Its revenue and valuation have also officially exceeded those of OpenAI, with an annual recurring revenue (ARR) reaching $44 billion, and the company's latest valuation in the primary market exceeding $900 billion.

At this high - profile moment, on May 16th, the media released a new interview video of Anthropic CEO Dario Amodei (the interview itself took place in January this year). Different from many AI experts who paint a utopian picture of technology, Amodei gave many realistic warnings:

For example, he said that traditional economic laws are being broken, and human society will face for the first time a situation where "high GDP growth coexists with high unemployment rates."

He mentioned that the public's sentiment towards AI always swings between two extremes, but in fact, the evolution of AI capabilities has always maintained a smooth exponential leap. This continuous growth in capabilities is directly replacing human knowledge - based labor. A large - scale macro - economic restructuring is about to occur, but the whole society is almost completely unprepared for it.

Regarding Claude Code, which the outside world is most concerned about, 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 engineering supervisors within Anthropic no longer write code; their work has become to specifically review and edit Opus's outputs.

Amodei revealed that Claude Co - work, an intelligent agent application for non - technical people, was almost entirely developed autonomously by Claude Opus in just one and a half weeks. One day after the launch of this feature, its various indicators reached about four times those of similar products. Amodei emphasized that people increasingly need this essential intelligent agent task - handling ability, and large models are crossing the form of simple chatbots and truly establishing their status as core production tools.

We've compiled the key information from this interview. Here are the highlights:

1. Anthropic chooses the enterprise - level market to avoid the attention economy trap

Facing the fierce competition among peers in the consumer market, Anthropic chooses to focus on enterprise - level customers. Amodei believes that AI products targeting consumers often fall into the trap of maximizing user usage time, which is prone to breeding low - quality content and causing over - dependence. Anthropic refuses to participate in this attention - grabbing competition and is committed to providing enterprises with systems that can create substantial work value.

In addition, he pointed out that there is a fundamental difference in the sense of responsibility between AI enterprises led by scientists and early social media entrepreneurs. The former is more inclined to prevent problems before they occur and is used to proactively examining the potential social impact of technology before its large - scale implementation and taking steps to prevent possible negative consequences.

2. Mechanistic interpretability is the only path to achieving AI controllability

When testing AI security, Amodei believes that relying solely on external dialogue tests is extremely dangerous because advanced AI systems are fully capable of hiding their true operating logic from humans.

The most urgently needed technological breakthrough in the current security field is mechanistic interpretability. Researchers cannot just look at the surface text output by the system but must delve into the system's interior to directly observe and master its underlying data operation mechanism. Breaking the algorithm black box is the only reliable basis for truly achieving system security and absolute controllability.

3. AI capabilities continue to grow, while public sentiment fluctuates between two extremes

In the past decade or so, the public and the media's perception of AI has often swung between the two poles of "disrupting all industries" and "technological stagnation." However, the real evolution curve of AI technology is extremely smooth, and its system processing ability has always maintained a stable rhythm of significant leaps every few months.

Amodei pointed out that all sectors of society have failed to accurately and objectively measure this technological development law, resulting in a serious cognitive disconnect. This disconnect not only makes enterprises lack foresight when planning business transformation but also causes policymakers to blindly introduce response strategies based on wrong premises. As a result, human society is currently completely unprepared for the upcoming large - scale economic restructuring.

4. High growth and high unemployment will appear simultaneously in human society for the first time

AI is improving social productivity to an unprecedented extent. Taking software engineering as an example, AI code generation has made development work extremely efficient, directly leading to a significant reduction in costs or even making them almost free. This explosive productivity will drive the overall economic scale to expand rapidly.

At the same time, human participation in the work process is being rapidly squeezed. Currently, software engineers may only need to complete 10% of the work, and the rest is done by AI. As the model's capabilities continue to evolve, the proportion taken over by AI will increase. This will directly lead to the complete collapse of the traditional occupational system established by humans in the past few decades, and a large number of jobs will disappear.

Amodei emphasized that the core challenge in the future will no longer be the growth of the total economic volume but the distribution of wealth. Facing this unprecedented macro - misalignment of high growth and high unemployment, government intervention is necessary to ensure that everyone can benefit from the technological dividend and smoothly pass through the social transformation period.

5. To address social risks, it is necessary to ensure the fair distribution of AI benefits

Amodei is deeply worried about the possible extreme social rift in the future: if the huge economic dividends created by AI are only monopolized by a very small number of groups such as Silicon Valley tech elites, while the public is excluded, it will surely lead to a catastrophic social crisis.

To ensure the fair distribution of AI benefits, he put forward two core appeals:

First, increase technological investment in the public domain and directly apply cutting - edge AI technology to fields such as public health and basic education to ensure that people in different regions and social classes have equal economic development opportunities;

Second, promote the underlying transformation of basic education. Facing the job market reshaped by AI, future education must abandon simple vocational skills training and return to the cultivation of human's internal comprehensive qualities.

Here is the transcript of Dario Amodei's interview:

1. The smooth exponential growth of AI

Host: Dario, we're in Davos, and a lot is going on here. But first, I want to ask a big - picture question. This time last year, everyone was very excited about AI, and people were discussing its capabilities and potential. I feel that this year's discussions have shifted and become more in - depth, with less speculation about what AI can do for the world. So my question is, do you think businesses, policymakers, governments, or other institutions are well - prepared to deal with the impact of AI?

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

We can look at it from two dimensions. One is the capabilities of the technology itself. Every three to six months, the media experiences a reversal between two extremes: one moment, they're extremely excited about the technological capabilities, thinking it will change everything; the next moment, they think it's all a bubble and everything will fall apart.

But what I see is a smooth exponential growth curve, similar to Moore's Law in the computing field. In the field of intelligence, we also have a similar law, that is, the cognitive abilities of models become stronger and stronger every few months, and this progress has been constant. The idea that inventing something new will make everything collapse or hit a wall is purely a public perception phenomenon.

There is a similar situation in the polarization about whether this technology is good or bad. In 2023 and 2024, people had a lot of concerns about AI, such as thinking that AI would take over everything, and the discussions mostly focused on AI risks and abuse. In 2025, the political wind shifted towards the opportunities brought by AI, and now it seems to have shifted back.

Throughout this process, Anthropic and I have always tried to maintain a balanced perspective. This balance is very peculiar because this technology is extremely powerful in terms of capabilities, and its impact has both positive and negative aspects, which coexist.

About a year and a half ago, I wrote the article "Machines of Loving Grace." I have a very radical view on the optimistic side of AI, believing that it will help us cure cancer, eradicate tropical diseases, and bring prosperity to areas that have not yet witnessed economic development. My view hasn't changed, and I still believe in these things.

But on the other hand, bad things will also happen. I've written more about this recently and may publish it soon. Taking economic risks as an example, the remarkable feature of this technology is that it will lead us into a society with extremely high GDP growth but also potentially high unemployment rates and full of inequality. This combination is something we've hardly ever seen.

In the past, high GDP growth meant there was a lot to do and everyone had many job opportunities. We've never seen such a disruptive technology. So we may face a situation where the GDP growth rate reaches 5% or 10%, but at the same time, the unemployment rate also reaches 10%. This is completely logical and just hasn't happened before.

For these two reasons, I'm both excited and worried. Taking AI programming as an example, we released the latest model, Claude Opus 4.5. Some engineers and engineering supervisors within Anthropic basically told me that they no longer write code; they just let Opus do the work and are responsible for editing.

We just released a new feature called Claude Co - work, which is a version of the Claude Code tool for non - programming scenarios. It was built in just one and a half weeks, almost entirely developed using Claude Opus. Software engineers still have something to do. Even if they only do 10% of the work, they still have a job or can move up a level.

But this won't last forever. The models will become more and more powerful. Here, amazing productivity is demonstrated, and software will become cheap or even basically free. The premise is that the cost of the software you build needs to be spread across millions of users, and this situation may not exist. For example, for this conference, we may only need to spend a few cents to develop some applications for people to communicate with each other. It's very flexible and recyclable. But at the same time, our entire careers that we've struggled for decades may no longer exist. I think we can adapt to it, but the public is completely unaware of what's about to happen and its scale.

2. How society adapts to AI development

Host: This is really interesting. So what do you think society will be like in a world with high GDP growth but also high unemployment rates? You said that people haven't started thinking about this yet. Can you give some specific examples of how society can adapt to such a world?

Dario Amodei: The first thing we're focusing on is a project called the Anthropic Economic Index. This is the first step. We've been running this index for about a year and have updated it four or five times so far. It's a real - time index that allows you to track what our model, Claude, is being used for. It goes through all the conversations and statistically analyzes the usage of Claude in a privacy - protected way, such as what tasks it's used for, to what extent it's automating tasks or enhancing capabilities, which industries it's applied to, and how it spreads across states in the US and countries around the world. We're adding more and more details. My view is that any policy will be blind and misleading until we can measure the shape of this economic transformation. Many policies go wrong because they're based on incorrect premises.

The second step is that we need to think very carefully about how to enable people to adapt to AI development. This may mean adapting to and using this technology in existing jobs, or it may mean transitioning from one job to another. For example, I think there may be more job opportunities in the physical world, while there will be fewer in the knowledge - based economy. Although robot technology will also make progress in the end, it's on a slower development trajectory.

In addition, will there still be jobs that highly value human touch? Some will, and some won't. We'll find out how important this is and in which fields it's most important. At the corporate level, where is the moat when software and other knowledge - based work become cheap? We've never really asked this question because we've always thought about moats in a certain way. So there will be a huge battle at the corporate level. Teaching people to adapt and anticipate what will happen is the second step.

The third step is that when there is such a huge loss of human labor at the macro - economic level, the government must play a certain role. The cake will become much larger, and there is sufficient funds. Due to such strong growth, the budget may be balanced even if we do nothing. The problem is how to distribute it to the right people. So I think we should now worry less about weakening growth and focus more on ensuring that everyone can share this growth. This is the opposite of the current general sentiment, but the technological reality is about to change and will force our concepts to change as well.

3. Claude and the popularization of intelligent agents

Host: I want to talk more about Claude. It's at a high - profile moment. We've also reported recently on how engineers and ordinary users have become "Claude - oriented." I'm wondering how you feel about the current situation and how the business performance compares to a year ago?

Dario Amodei: The business has been growing rapidly, basically on the same smooth exponential growth curve as the technological development.

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. Although these are rounded numbers, that's roughly the situation.

A few months ago, people on Twitter were extremely excited, exclaiming that Anthropic was changing the world and completely disrupting the industry. But we just quietly observed this rapidly rising and continuously improving curve. It gave us confidence. Although we can never be sure if this growth will continue, this has always been the empirical result we've observed throughout. Even if the curve is smooth, there will be breakthrough moments.

I think there is currently a breakthrough moment in the developer community around Claude Code. The ability 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. The progress is gradual, like boiling a frog in warm water. You'll see a gradual improvement, and then at a certain point, people will suddenly notice it.

The second point that may accelerate this process is that when we observed Claude Code, we noticed that many non - technical people inside and outside Anthropic realized that Claude Code can complete incredible Agentic tasks for you. It can not only write code but also organize to - do lists, plan projects, organize folders, or process a large amount of information and summarize it.

This concept is not just a chatbot but an essential intelligent agent task - handling ability. Non - technical people are so eager to get it that they're willing to delve into the command - line interface. For non - technical people or non - programmers, this interface is terrible to use, but even so, people are still insisting on using it. After seeing this, I thought, this looks like an unmet need.

So about two weeks ago, we used Claude Code again to make a version with a better UI, specifically customized for tasks other than code. About one day after the release, the various indicators were about four times those of other products, better than any product we've ever released. I'm not sure if these are entirely new capabilities, but this is the kind of moment when a consensus is reached, and people become very excited and quickly promote the popularization. People are gradually understanding the capabilities of this technology because it