Claude Code on the Web: A Solid v1 That Outshines Cursor

Read Articleadded Oct 28, 2025

The author has been using Claude Code on the web extensively and finds it effective despite its early-stage feel. It supports async development workflows by creating branches, PR-based diffs, and a teleport command to continue locally, and it works well via the iOS app. Compared to Cursor’s similar feature, Claude’s greater stability and polish make it the better experience.

Key Points

  • Claude Code on the web is a simple v1: start a thread, it runs in a container, creates a branch, and you view changes via a PR.
  • You can continue locally by using a claude --teleport <uuid> command that syncs the branch and preserves the thread.
  • The author uses it as a self-updating to-do list across multiple projects, returning later to review and ship changes.
  • iOS app availability enables on-the-go questions and answers that wait for the user.
  • Compared to Cursor’s earlier similar feature, Claude Code feels more stable and polished; product quality drives the author’s preference.

Sentiment

Overall, the sentiment is **mixed but cautiously optimistic** regarding the future of AI coding agents. While the article highlights Claude Code on the Web's 'solidity and dependability,' the Hacker News discussion reveals a strong divide. Many users express enthusiasm for the transformative potential of these tools, but there's a clear preference for GPT 5 Codex among some for its code quality, contrasted with significant frustration over Claude Code's (particularly the CLI's) performance, stability, and feature limitations.

In Agreement

  • Claude Code's incremental, careful, and narrow-scope approach is preferred by some over competitor models like Codex, which can 'run off' or make excessive changes.
  • The web-based workflow and mobile integration of Claude Code enable efficient 'vibe coding from an iPad' and support self-executing to-do lists, leveraging Git for memory and pull requests for human review.
  • Claude Code can be configured to commit changes in the user's name, addressing a specific workflow preference and making it feel more like a team member.
  • The general utility and transformative power of AI coding tools, including Claude Code, are highly praised for enabling rapid development and project updates.

Opposed

  • GPT 5 Codex (often via Codex CLI) is often perceived as superior in raw code quality, capable of 'one-shotting' problems and consistently producing working code, contrasting with Claude Sonnet/Opus models' occasional errors.
  • Claude Code (especially the CLI version) is heavily criticized for significant bugs, memory leaks, high CPU usage, tool call errors, infinite loops, and context leaks, leading to a 'broken' and unreliable user experience.
  • Anthropic's engineering is questioned, with Claude Code's performance problems attributed to inefficient I/O operations and constant serialization/deserialization to a large, shared JSON state file.
  • The overall workflow loop (container creation, cloning, changes, testing, PR) for Claude Code is considered too slow for anything beyond trivial tasks, and it lacks support for devcontainers, custom environments, and essential tools.
  • Claude Code's automatic creation of public Git branches for speculative work is a concern, as users prefer a manual authorization step for pushes to public repositories.
  • While Claude is good for 'vibe coding' with poor prompts, some argue its 'collapsed distribution' causes it to disobey detailed prompts or struggle with atypical scenarios, placing it behind GPT Codex in overall capability.
Claude Code on the Web: A Solid v1 That Outshines Cursor