Another team, protecting children in sports wanted to prevent harm rather than respond to it after the fact. They were clear that the adults around a child — coaches and parents had to be part of the design. That principle became the architecture: separate spaces for each group, and a first phase small enough to be safe.

A global gender-data organization came to us for advice on an AI-powered index. They hold one of the most comprehensive pictures of women's lives anywhere, but drawing insight from it took time and technical skill most advocates don't have. They proposed one approach; we told them where it would break. A language model asked to act as a calculator will quietly round a figure or rank countries in not-quite the right order, and for an organization whose credibility rests on its data, a wrong number handed to a journalist is the one outcome worth designing out. We recommended a hybrid instead: plain language in, exact numbers computed straight from their own data, every figure traceable to its source.