Anthropic brings Claude Code to the web for browser-based coding
The new web version could turn your browser into a full coding workspace, where AI agents can write, edit, and manage code in real time.
Something interesting is happening in software development. One that’s moving us away from writing code to managing it through intelligent systems that understand what we mean and act on it.
That’s the idea behind Claude Code, Anthropic’s AI coding assistant that’s now available on the web. It started as a quiet command-line experiment for power users, but Anthropic is opening the doors to everyone. With the new browser version, you can connect your GitHub repositories, describe what you want built, and Claude handles the implementation in real time.
The difference isn’t in what it can write, but how it works. Earlier tools like GitHub Copilot behaved like advanced autocomplete, finishing your thoughts as you typed. Claude Code belongs to a new generation of autonomous tools that can run multiple sessions across repositories, file pull requests, and summarize their own work when done. You can also assign tasks like bug fixes, backend updates, or refactors, then focus on design or architecture while Claude handles the routine.
Right now, Claude Code for the web is in beta, available to Pro ($20/month) and Max ($100–$200/month) users. Its mobile version is coming soon, starting with the iOS app, which will let you monitor sessions or even start new ones while away from your desk.
Security-wise, Anthropic seems determined not to repeat the mistakes of early AI developer tools. It says every Claude Code session will run in a sandboxed cloud environment with restricted access to files and networks, meaning the model can’t wander into unauthorized areas of your codebase.
It also adds that all Git interactions will happen through a secure proxy, so Claude can only touch approved repositories, and developers can even define which domains it connects to.
But the story gets even more interesting when you learn that 90% of Claude Code’s own codebase was written by Claude Code, at least according to Anthropic's product manager, Cat Wu. It’s both impressive and slightly eerie that the product is helping to build itself.
That autonomy comes with trade-offs, though. In older versions, Claude asked before almost every change. The new one acts more confidently, which is great for speed but raises the stakes for review. It’s still early days, but the idea of codebases that maintain and extend themselves no longer feels like science fiction.
Anthropic says Claude Code’s user base has grown tenfold since May and now contributes more than $500 million in annualized revenue, part of its broader climb toward a $170 billion valuation.
Ultimately, maybe the most fascinating part isn’t that Claude writes code faster. It’s that developers are learning to direct, monitor, and correct, not just type. Software isn’t being written less; it’s being written differently.

