A researcher compared five AI agents with different roles and contexts against five identical clones using the same model (Opus 5) and budget on a 14-night virality forecasting task. The diverse agents outperformed the clones on 9 of 14 nights with a better pooled Brier score (0.0225 vs 0.0275), suggesting that multi-agent networks compound context rather than just compute.