Anthropic today launched Claude Code on the web, a browser-based interface that lets Pro and Max subscribers dispatch coding tasks to Claude on Anthropic-managed cloud infrastructure. The research preview is accessible from claude.ai and the Claude iOS app; users connect GitHub repositories, describe what they need, and Claude implements changes in isolated sandbox environments with network and filesystem restrictions. Each session runs in its own sandbox, and git interactions pass through a secure proxy that limits Claude to authorized repositories.

The launch extends Claude Code beyond the command-line interface, where it has become a favorite among developers since its broader release in May. Anthropic says the product has grown 10x in users and now accounts for more than $500 million in annualized revenue. Anthropic says the web interface is especially effective for answering questions about how projects work and how repositories are mapped, bugfixes and routine, well-defined tasks, and backend changes, with real-time progress tracking, automatic pull request creation, and change summaries. Anthropic product manager Cat Wu said the team will continue to put Claude Code in more places, though the terminal will likely remain its home base.

The mobile release is an early preview, and the company is collecting feedback to refine the experience. Security is a central pitch: every task runs in an isolated sandbox, and custom network configurations let users specify which domains Claude can reach, such as allowing npm package downloads during test runs.

The practical questions are already circulating among developers: how much time gets spent reviewing rather than writing code, and whether parallel agents actually reduce the backlog or just add review overhead. One recent study found some engineers were slower with AI coding tools, spending time prompting and waiting. Anthropic’s claim that 90% of Claude Code is written by its own models only sharpens the debate about how far delegation can go.