Claude vs. Codex vs. Opencode: a practitioner's notes

Not a benchmark. A working log of how these tools actually behave when you're trying to ship something real — what they're good at, where they break, and what that tells you about the underlying architecture.

// ongoing