Claude Code vs Cursor for agent development - my experience

I've used both extensively for building agent systems. Here's my take: Claude Code (CLI): Better for agent-to-agent workflows. The sub-agent pattern is powerful. Great for headless automation. Cursor: Better visual experience, inline diffs are nice. But it's focused on human-in-the-loop coding. For building AI agents specifically, Claude Code wins because you can programmatically control it. The agent can spin up sub-agents, which is a game changer. What's your setup?