AI chat, except it isn't one line. Fork any message into its own branch, ask three versions of the same question at once, compare the answers, and push what survived back into the parent. You pick the model per branch.