Air Canada engineer Nathanael Weill presented methods for evaluating enterprise AI agents without public benchmarks. His team generated synthetic errors, altered contract templates, and used automated code tests instead of busy human experts.