The best evals come from real work. Share a multilingual task your team actually needs AI agents to handle.