Claude Code started running tasks on the web in October 2025. The browser mattered more to me than another editor integration because it turned coding into delegated work: connect GitHub, hand off a task, and come back to the result.
Three Interfaces, Different Tradeoffs
Web ran tasks on Anthropic’s cloud and let me start several independently. I could check progress, steer a task, and resume it later.
CLI runs in my local environment, with access governed by its permissions and sandbox settings. That gives me control of the environment; cost depends on the plan or API billing I use.
iOS is for monitoring and light tasks: checking progress, providing feedback, triggering simple refactors. Anthropic admits it needs refinement.
The web launch connected authorized GitHub repositories to a cloud sandbox, with a proxy handling Git operations. That is the setup described in Anthropic’s launch post. The local CLI uses the checkout and credentials I give it; the cloud sandbox guarantees do not automatically apply there. These interfaces do not require one shared Dispatch workflow.
The Maintenance Machine
Refactoring, test generation, dependency upgrades, and bug fixing are not glamorous, but that is what fills sprints. React project needs upgrading? Claude analyzes the codebase, identifies breaking changes, runs migrations across files.
Type annotations on JavaScript files. Unit tests for new functions. Database migrations from schema changes. The time savings pile up on repetitive work that eats days, not on architectural decisions.
The web interface shifts the psychology. You’re delegating, not augmenting your editor. That changes how you scope the work.
A Browser Tab Lowers the Commitment
Before the web launch, I reached for Claude Code through a terminal or editor integration. I had to seek it out.
In the browser, I could try one delegated task without moving my editor setup. That smaller commitment was the product change I cared about.
Was this useful?
Thanks for the feedback.