Anthropic brings Claude Code to your browser

You no longer need to live inside a terminal to let Anthropic's AI write your software.

Anthropic just rolled out Claude Code on the web in a beta research preview, bringing its viral coding assistant directly to the browser and the Claude iOS app. The web version is available now to Pro subscribers paying $20 a month, as well as Max tier users paying $100 or $200 monthly.

The tool has been an absolute cash cow for Anthropic. According to product manager Cat Wu, Claude Code has grown tenfold in users since its broader launch in May and now generates over $500 million in annualized revenue. Anthropic also claims that 90% of Claude Code itself was written by its own models, with Wu admitting she rarely writes code by hand anymore and mostly spends her time reviewing AI outputs.

Cloud sandboxes and parallel coding

Taking Claude Code out of the terminal unlocks something local environments struggle with: true parallel execution.

Instead of watching a single CLI prompt work through one task at a time, developers can now connect their GitHub repositories and fire off multiple coding jobs across different repos simultaneously from a single browser tab. Each session runs inside an isolated, Anthropic-managed cloud sandbox with strict network controls and a secure Git proxy.

You can watch Claude work in real time, steer it if it starts heading in the wrong direction, and let it configure network access to download packages like npm for testing. Once finished, Claude automatically creates pull requests with clear change summaries.

While the web and mobile apps make delegating tasks far more accessible on the go, Wu noted that the command line will remain Claude Code's primary home base for deep customization.

Why it matters

This shift fundamentally changes the developer's day job from typing code to managing an autonomous engineering fleet.

If you are a solo developer drowning in bug backlogs or maintenance chores, spinning up three cloud agents in parallel to handle well-defined backend fixes is an obvious win. It lowers the barrier to entry significantly by removing local environment friction, and being able to check a PR from your phone is genuinely useful.

Still, there is a catch. Recent studies have pointed out that engineers sometimes work slower with agentic tools because they spend excessive time writing prompts, waiting on generation, and untangling errors in complex codebases. Running parallel agents could easily multiply your code review headache if Claude hallucinates subtle bugs.

With Cursor, OpenAI, and GitHub all racing to dominate autonomous workflows, Anthropic had to make this move. Putting agents in the cloud is smart infrastructure play, but keep in mind that cloud sessions still share your overall Claude Code rate limits. For messy legacy codebases, human oversight is still non-negotiable.