The Roundtable
Different minds. Better questions. Human judgment. Claude and ChatGPT in the same conversation, critiquing the same work, with a human in the chair.
I'd been running the same questions past Claude and ChatGPT separately, then manually carrying one model's answer over to the other to see what it would push back on. Every round trip was me doing copy-paste as a job. The Roundtable removes the courier step: both models read the same prompt and each other's responses in one shared thread, and I moderate. It turns out two models with genuinely different instincts — one more expansive, one more skeptical — catch different things, and the disagreements are often more useful than either answer alone. What they don't get is the final word. Items that need judgment stop at a status of AWAITING TERESA and wait. The models argue; the human decides.
Debate
Independent AI perspectives. A Google Sheet holds one shared thread, and Google Apps Script passes each turn to the Claude and OpenAI APIs, so both models see the prompt and each other's replies instead of answering in two separate chat windows.
Challenge
Test assumptions and ideas. The first live topic was a critique of the Builder's Ledger, the practice on this site that asks whether AI-assisted work leaves the human more capable. Two models stress-testing a framework about not over-relying on AI felt like the right first test.
Decide
Human judgment wins. I set the topic, steer the rounds, and own the outcome. Anything that needs a real decision halts at AWAITING TERESA instead of being settled by whichever model sounds most confident.
Boundaries
Participants get a deliberately limited public context (positioning, approved projects, voice, guardrails) and nothing private, confidential, or employer-related. The boundary is designed in, not bolted on.
Built and running since October 2026. Next exploration: whether invited human collaborators can join the same table, with the same rules — the chair still makes the call, and the AI participants stay critics, not decision-makers. Consistent with the rest of this site: AI accelerates the thinking; it doesn't own the decision.