Anthropic's AI programming tool, Claude Code, is steering the global software development industry into a pivotal turning point. Leveraging its latest "Agentic" capabilities, the product has not only revolutionized code writing but has also achieved explosive growth in commercialization.
On January 22, according to a report by the tech media WIRED, Boris Cherny, head of Anthropic Claude Code, revealed in an interview that with the leap in underlying model capabilities, software development is shifting from manual coding to full AI agency. This transformation is reshaping engineers' work patterns and the entire industry's productivity logic.
According to reports, informed sources disclosed that Claude Code's annualized recurring revenue (ARR) had further increased by at least $100 million by the end of 2025, following the product's announcement in November that its ARR had just surpassed the $1 billion mark. Currently, Claude Code contributes 12% of Anthropic's total ARR of approximately $9 billion, making it one of the company's fastest-growing business segments.
This breakthrough is largely attributed to the launch of Anthropic's latest AI model, Claude Opus 4.5, which is regarded by many developers as a "step-function" improvement in programming capability. Unlike earlier tools limited to code completion, the new model-based Claude Code can understand natural language instructions and autonomously carry out subsequent development. This "agentification" trend has prompted startups like Cursor and Windsurf, as well as tech giants like OpenAI, Google, and xAI, to accelerate their deployments, fiercely competing for a share of this emerging market.
As AI agent capabilities mature, Anthropic is attempting to replicate this model of "shaking up" the software industry in broader domains. The company recently launched a new product called Cowork, aiming to extend AI agent functionality from programming terminals to general office scenarios like file management and software interaction, attempting to ignite a similar productivity revolution in non-programming fields.
The evolution speed of AI programming tools has exceeded expectations. According to the Wired report, from 2021 to 2024, most tools could only offer auto-completion features, suggesting a few lines of code as developers typed. By early 2025, startups like Cursor and Windsurf began launching early "agentic" programming products.
Boris Cherny, head of Claude Code, stated in an interview that Anthropic bet on the future direction of AI capabilities during product design, rather than its actual level at the time. Cherny said:
"We built the simplest thing. The craziest part is that three months ago, we learned half of Anthropic's sales team was using Claude Code every week."
This strategy proved to be forward-looking. Cherny revealed that the proportion of code he personally wrote using Claude Code increased from an initial 5% to 30% last May with the releases of Opus 4 and Sonnet 4. In the two months following the launch of Opus 4.5, 100% of his code was written by Claude Code.
The report stated that Workera ultimately chose Claude Code after testing multiple AI programming tools. Katanforoosh indicated that for the company's senior engineers, Claude Code's performance surpassed competitors like Cursor and Windsurf.
The core advantage of Claude Code lies in its agentic working mode. Cherny explained that unlike a year ago when users primarily utilized chat functions, Claude Code and Cowork have truly achieved agency, capable of using tools, reading system files, and interacting with Slack and Google Sheets.
"This is the golden age for those with short attention spans," Cherny said. The most efficient Claude Code users launch multiple tasks simultaneously, let Claude run each one, and then check the progress sequentially. Cherny himself typically runs five to ten agents concurrently across terminal, mobile, and web interfaces.
This working style is becoming widespread within Anthropic. During an internal review meeting before the product's external release, CEO Dario Amodei asked if employees were forced to use the product. In reality, almost 100% of Anthropic's technical staff frequently use Claude Code, and the team writes 95% of its code with Claude Code.
Anthropic's enterprise customers are exhibiting similar usage patterns. Cherny mentioned that enterprise customers are very similar to Anthropic itself in terms of security requirements and product interaction methods, enabling the company to optimize the product through internal use.
Building on the success of Claude Code, Anthropic this month launched Cowork, an AI agent product aimed at non-programmers. It can manage files on a user's computer and interact with software, without needing to touch a coding terminal.
Cherny described Cowork as "Claude Code for non-programmers." He has already started using Cowork for project management, for instance, having it check if engineers have filled out their work progress sheets and sending reminders via Slack to those who haven't.
"AI agents will be able to handle all the tedious things in life. This is already happening this year in engineering, and I think it will happen in every other field. Agents will be able to handle tasks like filling out forms, moving data between different places, sending emails, etc."
Cherny acknowledged that this shift will have a disruptive impact, requiring a collective industry response. However, he believes it makes work more enjoyable, freeing people from tedious tasks.
The success of Claude Code is attracting more competitors into the AI programming market. Cursor announced in November last year that it had reached $1 billion in annualized recurring revenue; its platform allows users to program using models from Anthropic and other AI labs. According to informed sources, Cursor achieved particularly strong month-over-month revenue growth in December.
OpenAI, Google, and xAI are also accelerating the development of agentic programming products based on their own AI models, attempting to capture a larger share of this market. The core of this competition lies in the capabilities of the underlying AI models, and the launch of Claude Opus 4.5 seems to have given Anthropic a temporary advantage.
For engineers experiencing this transition, Cherny offered a historical perspective. His grandfather programmed using punch cards in the Soviet Union, after which programming evolved through machine code, C, Java, Python, and so on.
"As an industry, we've always been undergoing transformations. It's a continuum of increasing abstraction, and I think agents are just a point on that continuum."
Cherny anticipates that as the learning curve smooths out, effectively using tools like Claude Code and Cowork will become increasingly easier. Anthropic plans to achieve positive cash flow by 2028, and Claude Code, as one of the fastest-growing business segments, is expected to play a key role in achieving this goal.
Comments