Claude Code gets a web version—but it’s the new sandboxing that really matters (arstechnica.com)

Anthropic has rolled out web and mobile interfaces for Claude Code, its agentic AI coding tool that began life as a popular CLI. The new web UI lets developers connect a GitHub repo, give high-level commands like “add real-time inventory tracking to the dashboard,” and watch Claude Code execute steps with live progress updates. It also supports in-flight feedback—so you can suggest or correct behavior while the agent is mid-task instead of cancelling and restarting—and lets you run and switch between multiple sessions via a left-side panel. The mobile client is available on iOS but remains in an earlier stage compared with the polished web experience. What really matters for the AI/ML community is Anthropic’s new sandboxing runtime for Claude Code. Beyond convenience, the sandbox aims to reduce risk and friction by isolating the agent’s execution, constraining what it can modify, and streamlining permission flows; historically Claude Code asked for permission before most changes, which slowed iteration. Safer, lower-friction runtimes make agentic coding more practical for larger teams and continuous workflows, but they also shift attention to how sandboxes are scoped, audited, and integrated with repo permissions. If Anthropic’s approach balances autonomy with tight controls, it could accelerate adoption of agent-driven software development while setting a practical bar for secure, production-ready coding agents.