Anthropic launched a web application for its AI coding assistant, Claude Code, on Monday, allowing developers to create and manage multiple AI coding agents directly in the browser. This marks an expansion of Claude Code from a command-line tool to a multi-platform product.
The web version of Claude Code is now available to subscribers of Anthropic, including users of the Pro plan at $20 per month, as well as those on the Max plans priced at $100 and $200 per month. Users can access the feature by visiting the claude.ai website (the same site as Anthropic's consumer chatbot) and clicking the "Code" tab, or through the Claude iOS app.
This launch is the latest attempt by Anthropic to move Claude Code beyond a command-line interface tool. By deploying Claude Code on the web, Anthropic hopes that developers will use AI coding agents in more scenarios.
The market for AI coding tools is becoming increasingly competitive. While Microsoft's GitHub Copilot once dominated this space, Cursor, Google, OpenAI, and Anthropic now all have high-performance AI coding tools, many of which already offer web versions. Despite this, Claude Code can be considered one of the most popular products. Since its broad release in May, the number of users of Anthropic's flagship coding tool has increased tenfold, and the product currently contributes over $500 million in annualized revenue for the company.
Cat Wu, a product manager at Anthropic, told TechCrunch that she attributes much of Claude Code's success to the company's AI models, which have become a favorite among developers in recent years. However, Wu also said that the Claude Code team intentionally "adds some fun" into the product.
Wu said that Anthropic will continue to deploy Claude Code in more places, but the terminal may still be the main hub for its AI coding product. "Looking ahead, one of our focuses is to ensure that the CLI product is the smartest and most customizable way to use coding agents," Wu said. "But we will continue to bring Claude Code to wherever developers are, serving them where they are. The web and mobile versions are important steps in this direction."
Anthropic claims that 90% of the code in the Claude Code product itself was written by its AI models. Wu, who used to be an engineer, said that she now rarely sits at the keyboard writing code, mainly reviewing the output of Claude Code.
Early AI coding tools worked like autocomplete tools, completing lines of code as developers wrote them. But the new generation of agent-based AI coding tools, including Claude Code, allows developers to launch agents that work autonomously. This shift has made millions of software engineers more like managers of AI coding assistants in their daily work.
This change is not welcomed by all developers. A recent study found that some engineers actually slowed down when using AI coding tools like Cursor. Researchers believe one factor could be that the engineers in the study spent a lot of time prompting and waiting for AI tools to complete tasks, rather than dealing with other issues. AI coding tools also face challenges in large, complex codebases, so engineers might spend a lot of time handling incorrect responses from AI models.
Despite this, companies like Anthropic continue to push forward with the development of AI coding agents. Anthropic CEO Dario Amodei predicted a few months ago that AI should soon be able to write 90% of the code for software engineers. Although this may be true within Anthropic, such a shift could take longer to materialize in the broader economy.
From a product strategy perspective, the launch of the Claude Code web version lowers the entry barrier. While command-line tools are familiar to many developers, the web and mobile versions can cover more use cases, such as quick code reviews, temporary debugging, or coding needs in non-developer environments. This multi-platform strategy helps increase product usage frequency and user engagement.
The $500 million in annualized revenue indicates that Claude Code has become an important revenue source for Anthropic. Considering that the product was only broadly released in May, this growth rate is impressive. However, this revenue is likely primarily driven by enterprise customers and heavy users, rather than the mass market.
The data mentioned by Wu about "90% of the code being written by AI" highlights the capabilities of AI coding tools, while also raising discussions about the future role of software engineering. If engineers' primary job shifts from writing code to reviewing AI outputs, the required skill sets and workflows will undergo fundamental changes. This shift presents particular challenges for junior engineers, as they may lose opportunities to learn by writing code themselves.
Findings from research showing that some engineers experience a decrease in efficiency after using AI tools are worth noting. This suggests that AI coding tools do not always improve efficiency in all scenarios. In complex codebases, tasks requiring deep understanding of business logic, or decisions involving architecture, AI tools may increase rather than reduce workload. Engineers need to spend time understanding AI-generated code, verifying its correctness, and integrating it into existing systems, which could offset the benefits of automation.
From a competitive perspective, the challenge for Claude Code is how to maintain differentiation in a market where features are increasingly similar. When multiple companies provide high-performance AI coding assistants, marginal differences in model quality may not be enough to sustain a competitive advantage. Product experience, integration ecosystems, pricing strategies, and brand awareness will become even more important. Anthropic's good reputation in the developer community is an advantage, but whether this advantage can translate into long-term market leadership remains to be seen.
Amodei's prediction about "AI will write 90% of the code" represents a vision, but the path to realization is unclear. Even if AI can generate most lines of code, the core challenges of software development—understanding requirements, designing architectures, making trade-off decisions, and debugging complex problems—will still require human judgment and creativity. AI coding tools are more likely to change how engineers work, rather than completely replace their roles.
