I've been loving Claude Code on the Web (ben.page)

Anthropic’s Claude Code on the web is showing early promise as a lightweight, developer-facing coding assistant that feels polished and dependable. In this v1 experience, each prompt spawns a small container “agent” that works in its own thread and produces a branch you can later open a PR to inspect changes (the PR is currently the only way to see diffs). If you prefer local work, a provided cli command (claude --teleport <uuid>) pulls the branch down so the same conversational thread continues on your machine. There’s also seamless continuity via the Claude iOS app, letting you capture ideas while mobile and pick up results later. For the AI/ML and developer tooling communities the significance is practical: Claude Code reduces friction around small edits and exploratory tasks, effectively acting like a to-do list that executes itself and preserves conversational context across sessions and devices. Key technical implications are the containerized agent model, branch/PR-based visibility for edits, and a teleport workflow for local continuation—patterns that prioritize auditability and developer control. Users compare it favorably to Cursor’s earlier effort, crediting product polish and smoother UX for higher adoption potential despite both being early-stage offerings.