Claude Code: Now in Beta in Zed

Original Article ↗ Hacker News Discussion ↗

Claude Code integration in Zed (features & rough edges)

  • Integration uses Anthropic’s SDK/ACP, so several Claude Code desktop features are missing: no Plan mode, limited slash commands (/compact, /clear, /new, ESC-ESC), no multi-agent support, unclear model switching, and weak context-window management.
  • Users report errors during setup (“can’t load supported slash commands”, initialization failures), though some were quickly patched.
  • Confusion around billing: if an Anthropic API key is present, Zed may bill via API instead of using a Claude subscription, surprising some who burned through API balances.
  • Compared with running Claude Code in a terminal, Zed’s pitch is first‑class diffs, integrated review/rollback, and editor focus tracking edits—but several people say the CLI + editor still feels more reliable today.

AI enthusiasm vs resistance and business model worries

  • Many like Zed’s AI features (agent mode, Claude integration) and see this as a sustainable business path, especially given VC funding.
  • Others dislike “LLM‑infested” tools on ethical or aesthetic grounds, even if features are fully toggleable; they fear AI will dominate roadmap over core editor quality.
  • Strong skepticism about VC funding and eventual “enshittification”; some see capitalism/VC, not “AI itself,” as the underlying problem.
  • Data/ethics concerns: one comment notes that rating AI responses may send entire chat history to Zed, which is seen as risky for proprietary code.

Zed vs VS Code, JetBrains, and others

  • Pro‑Zed: praised for extreme responsiveness, low memory, strong Vim mode, clean design, and being native/not Electron. Many use it as main editor or quick lightweight alternative to heavy IDEs.
  • Anti‑Zed: some find startup and typing laggier than VS Code or even Emacs; others report frequent crashes on Linux and GPU/Wayland issues.
  • Compared to JetBrains IDEs, Zed is seen as “a very good editor” rather than a full IDE: Git UI is basic (no 3‑way merge, limited diffs), test tooling shallow, and multi‑file refactoring weaker.
  • VS Code is defended as “fast enough” with a massive extension ecosystem; critics emphasize Electron latency and bloat, especially on modest hardware.

Plugin ecosystem, autocomplete, and local models

  • Zed’s extension catalog is small (hundreds) and mostly languages/themes, versus tens of thousands for VS Code. Some say the plugin API is too limited for rich UI integrations.
  • Cursor is repeatedly cited as having vastly better AI autocomplete/edit predictions; this is the main blocker for many who otherwise prefer Zed’s core editor.
  • Users want first‑class support for local models (Qwen, Ollama) for both agents and inline completions; partial support exists via custom agents and ongoing PRs, but it’s not yet as polished.
  • Zed’s own autocomplete model (a fine‑tuned Qwen 7B) is open source, which some see as a plus.

UX, configuration, and platform gaps

  • Complaints include: JSON‑only settings without a rich GUI, inflexible panel layout, lack of vertical tabs, and weak Git/merge UI. Some feel the UI is less “balanced” and polished than VS Code.
  • Font rendering on non‑HiDPI displays is a recurring pain point; lack of subpixel rendering and hinting makes Zed look blurry for some users on Linux/Windows‑style setups.
  • Remote SSH development is considered immature: separate configs per remote, crashes, and Claude Code not working over remote yet.
  • No official Windows build: some use unofficial builds, but rough edges deter others.

Standardization (ACP) vs deeper redesigns

  • Several commenters are excited that ACP/MCP could unify agents and editors: any agent (Claude Code, Codex, Gemini, etc.) talking to any IDE, lowering switching costs.
  • Others criticize ACP as a bolt‑on to legacy editor architectures, arguing that real innovation would require shared state layers and rethinking IDEs beyond LSP + Git.