Different minds. One shared project.

Most multi-AI setups fail the same way: each model has a clever take, then the thread scatters and nobody owns the merge. What I want instead is boring and checkable: One brief (output, constraints,