The Loop Simulator is a five-minute interactive walkthrough of a real product moment. You run the same moment twice.
The first pass is the default: one prompt, one output, a human editing whatever comes back. The second pass is the same moment run as designed collaboration — two AI thinking partners holding deliberately different roles, and a human directing the work while it happens rather than approving it at the end.
You leave with three working drafts. There is no signup and nothing to install.
The point of running it twice is that the difference is not speed. Both passes are fast. Speed is table stakes now.
The difference is specificity, and where human judgment enters. In the default pass, judgment arrives at the end, as editing. In the designed pass, judgment sets the direction at the start and steers continuously — so the output is specific to this problem rather than plausible for any problem.
The two AI roles in the simulation have different mandates. One widens the option space — alternatives, framings, things you hadn't considered. The other narrows it — synthesis, structure, a draft you can actually use.
Assigning both jobs to a single AI collapses the distinction, and you get confident, average answers: neither wide enough to surprise you nor sharp enough to ship. Separating divergence from convergence is one of the few structural choices that reliably changes output quality.
The simulator is the applied companion to The Human-AI Loop, a published methodology for designing human-AI collaboration: Test → Build → Codify → Share.
It is one of several working artifacts in a two-year public practice. Related pieces: