A thought experiment, staged as a panel.

What-If turns a single hypothetical into a structured debate. You ask a question — a policy, a technology shift, a counterfactual — and an AI moderator convenes a panel of stakeholders to argue it out, live.

The three phases

Every discussion moves through three explicit phases: first-order (the immediate aftermath), second-order (how systems and behaviors adapt over months and years), and third-order (the structural picture a generation out). This is the simplest frame for thinking about consequences that doesn't collapse into either "it'll be fine" or "it'll be terrible."

The panel

Panelists are auto-suggested by the model based on your question — chosen to actually disagree, not to nod. You can mute any panelist (their turn is skipped) and peek at the exact system prompt the model uses to roleplay them. The prompts are viewable, not editable: this is a window into the machinery, not a sandbox.

The depth lever

Beginner keeps things plain-language. Intermediate is the default — substantive but readable. Expert assumes domain literacy and invites the panel to reach for frameworks, numbers, and edge cases.

A caution

Panelists are simulated. They will sound confident. They are sometimes wrong. Use this the way you'd use a well-read friend riffing on a question over dinner: as a thinking tool, not as settled truth.

Feedback

Found a bug, a missing perspective, or a panel that went off the rails? Tell us — every simulation also has a 1–5 star feedback box, and you can always email ramesh@marcelle.in. Real notes from real users are how this gets sharper.