OK, the new @tuple integration with @claudeai is kinda bonkers. Having Claude Code listening on a Tuple call and proactively spawning sub-agents and chiming in on a pairing session—while still a little creepy—is remarkably useful. I'll say something like "I think X is only used in Y module" and move on, and then 30s later Claude Code will pop up with a note about another usage I forgot about. Or we'll be processing several million records via an artisan command to test how it'll handle real-world volume and Claude's just in the background spot checking rows and chiming in with optimizations to make the command more performant. If you had told me a year ago that I'd be piping the raw transcript of my Tuple calls into an LLM, I would have laughed in your face. But holy smokes is it so useful. 😅