// compare

Two products. One rulebook.

Every comparison is a permanent URL with its own card. Same five questions, same published rules, same day — so the difference you see is the products, not the method.

01

Worth opening with

5 questions people already have
  1. Claude Code vs Codex 5/5 · 5/5 Two terminal agents with the same shape. The difference is where the leash sits.
  2. Cursor (agent mode) vs Windsurf Cascade 5/5 · 4/5 Two editors that both ship an agent mode and disagree about who presses go.
  3. Operator vs Manus 4/5 · 5/5 Two computer-use agents, two very different answers on supervision.
  4. v0 vs Bolt.new 5/5 · 2/5 Two prompt-to-app tools that market the same promise and score three points apart.
  5. CrewAI vs LangGraph 5/5 · 2/5 Two frameworks — judged as frameworks. Neither is a shipped agent, and that is the point.

Framework comparisons are labelled as such. A framework that fails a shipped-agent criterion is not a defective product — it is a different kind of thing, and the method says so.