The field has consolidated. A handful of agents now dominate real usage, and they differ more in workflow than in raw capability.
The short version
Claude Code — terminal-first. Strongest at long, multi-step tasks in existing codebases: migrations, refactors, debugging sessions. Extends through skills and MCP servers.Cursor — editor-first. An AI-native fork of VS Code; the lowest-friction way to get agent help while you stay in the driver's seat.Codex CLI — OpenAI's terminal agent, open source, moving fast. Comparable shape to Claude Code with a different model family behind it.Cline, Roo Code, Kilo Code — open-source editor agents with strong communities. If you want to see and control everything the agent does, start here.Zed — a fast editor with agentic editing built in, not bolted on.What the data shows
We track stars, forks and (where published) install numbers for every coding agent in the directory. Two patterns stand out: terminal agents are growing faster than editor plugins, and open-source agents keep pace with proprietary ones on activity, if not on polish.
How to actually choose
Pick by workflow, not by benchmark:
Live in the terminal, work in big repos → Claude Code or Codex CLI.Want AI inside your editor with full control → Cursor, or Cline/Roo if you want open source.Care about speed and a native feel → Zed.All of them speak MCP now, so the tool ecosystem you invest in carries over when you switch.