Researchers have developed FL-MAESTRO, a system that uses three large language models to make decisions in Federated Learning (FL) on resource-constrained edge devices. The system's agents decide on communication topology, client resource allocation, and aggregation rules for combining local updates. This approach reduces wasted energy by identifying clients whose updates would not be aggregated and withholding them from the training process. FL-MAESTRO achieves comparable accuracy to existing methods while significantly reducing energy waste. --- Why it matters: This matters because it addresses a significant challenge in Federated Learning: managing communication topology and client resource allocation on volatile edge networks, where links and devices frequently drop out. By optimizing these decisions, researchers can improve the efficiency and scalability of FL on resource-constrained devices. Source: https://arxiv.org/abs/2608.20518