Anthropic has expanded Claude Code beyond its command-line roots with new web and mobile interfaces. The bigger shift, however, is not only where developers can use the agentic AI coding tool. It is how much work Claude Code can do with fewer interruptions while still operating inside tighter boundaries.
The new features are available in beta now as a research preview for Claude users with Pro or Max subscriptions.
A web interface for an agentic coding tool
Claude Code is described in the source as an immensely popular command-line interface agentic AI coding tool. The new web version brings that workflow into a browser-based interface and can be connected to a GitHub repository.
Once Claude Code has access to a repository, developers can assign broad tasks rather than only small, step-by-step instructions. The source gives the example request: “add real-time inventory tracking to the dashboard.” From there, the tool begins working and provides progress updates that approximate what it is doing and where it is in the task.
The web interface also supports a newer Claude Code capability: developers can send suggestions or requested changes while the agent is already working. That matters because the previous pattern often required canceling and restarting if the agent began moving in the wrong direction or missed an important requirement.
The interface is also built for parallel work. Developers can run multiple sessions at once, switch between them, and see those sessions listed in a left-side panel.
Mobile access is here, but uneven
Anthropic has also added a mobile interface for Claude Code. The source notes an important distinction: the web interface appears well-baked at launch, while the mobile version is limited to iOS and is at an earlier stage of development.
That makes the web release the more complete access change for now. Developers who already use Claude Code from the command line may see the browser version as a more convenient way to start, monitor, and redirect work across repositories and sessions.
Still, the mobile release signals the same broader direction. Claude Code is no longer only a terminal-based coding assistant. It is becoming an agentic coding environment that can be reached through more surfaces, even if those surfaces are not equally mature yet.
The sandboxing runtime changes the workflow
Alongside the web and mobile rollout, Anthropic has introduced a new sandboxing runtime for Claude Code. The stated aim is to make the experience both more secure and lower friction.
Previously, Claude Code asked permission before making most changes and before many steps in the process. That kept developers involved, but it also meant frequent approval prompts during active work.
The new approach lets Claude Code receive permissions for specific file system folders and network servers. In practical terms, that can mean fewer approval steps while still restricting what the coding agent is allowed to touch or contact.
The network isolation model is a key part of that change. According to Anthropic’s engineering blog, Internet access is only allowed “through a unix domain socket connected to a proxy server running outside the sandbox. … This proxy server enforces restrictions on the domains that a process can connect to and handles user confirmation for newly requested domains.”
Users can also customize the proxy to define their own rules for outgoing traffic. That allows Claude Code to fetch npm packages from approved sources without giving it open-ended permission to communicate with the outside world.
Less friction means more responsibility
For many developers, the sandboxing changes may be more important than the new web or mobile interfaces. The reason is simple: agentic coding tools become more useful when they can continue working without asking for approval at every small step.
But that convenience changes the developer’s job. The earlier, approval-heavy workflow forced close attention to many individual actions. If Claude Code needed permission repeatedly, the developer had more chances to notice whether it was editing the right area, using the right approach, or drifting away from the requested task.
With fewer approval prompts, the review burden shifts later in the process. Developers may get a smoother experience, but they also need to be more disciplined about reviewing the resulting code. A bad call by the agent may be easier to miss if it happens inside a longer, less interrupted session.
That tradeoff is central to the update. Sandboxing can reduce risk from prompt injection and other problems by narrowing file system and network access. At the same time, a more independent coding agent still needs human review, especially when it is making changes across a repository.
What developers should take from the beta
The Claude Code beta shows two changes happening at once. First, Anthropic is making the tool easier to reach through web and mobile interfaces. Second, it is adjusting the safety and permission model so the agent can work with fewer approvals.
For day-to-day development, the second change may have the larger impact. A web app can make Claude Code easier to manage. Sandboxing can change how much autonomy developers are willing to give it.
The useful balance is not simply maximum freedom or maximum restriction. The source describes a model where Claude Code can work inside approved folders and approved network paths, while new domains still require confirmation and the proxy can be customized. That gives developers a way to reduce interruption without removing boundaries altogether.
Claude Code’s web version may be the most visible part of the release. The sandboxing runtime is the part that could reshape how developers actually use the tool: fewer interruptions, more independent sessions, and a stronger need for careful code review when the work is done.