Researchers have proposed a new approach to coordinating agents in complex multi-agent systems. The method, called HASSUM, estimates uncertainty using semantic entropy and density, which measure the trustworthiness of answer semantics rather than output probabilities. This allows for adaptive orchestration decisions, such as verifying outputs or selectively reprompting agents. The approach was evaluated on several benchmarks and shown to improve robustness and trustworthiness in agentic AI systems. According to the authors, semantic uncertainty is a practical signal for improving performance in these systems. --- Why it matters: This work matters because it addresses a fundamental challenge in developing large-scale multi-agent systems: coordinating agents under uncertainty. The proposed approach can be integrated into various agent architectures and has the potential to improve robustness and trustworthiness in AI systems that require complex reasoning. Source: https://arxiv.org/abs/2608.14707