Claude Code
by AnthropicAnthropic's terminal-native coding agent. Reads your codebase, writes patches, runs commands, and iterates until tests pass.
$ cat curator-note.md
Claude Code's defining choice is the terminal. Where Cursor and Windsurf re-skin VS Code into an agent host, Claude Code lives in your existing shell — same prompt, same git workflow, same multiplexer. That makes it strangely sticky: once you've delegated a refactor and gotten back a clean diff to review, the awkwardness of "switching to the AI editor" disappears. The MCP integration is also genuinely useful — agents that can read your linear tickets or hit your staging API solve actual end-of-day problems, not toy ones.
Where it falls short is in the long tail of UI affordances that an IDE-integrated agent gets for free. There's no codebase-graph view, no inline ghost text, no "show me where this symbol is referenced" beyond what grep gives you. For exploratory work in a codebase you don't know yet, a real IDE agent still wins. The pricing is also a sharper edge than Anthropic likes to admit — it's easy to burn through tokens on a single big task if you're not paying attention.
Use Claude Code if you already work primarily in a terminal, run a multiplexer, and want an agent that respects your existing workflow rather than imposing a new one. If you're an IDE-first developer who lives in tabs and panels, Cursor or Cline will probably feel more natural. The model quality is the same in both directions; the choice is really about where you want the agent to live.