arXiv:2508.08115v3 Announce Type: replace Abstract: Complex medical reasoning has historically required frontier language models to achieve clinically-acceptable accuracy, creating computational barriers that limit deployment in resource-constrained clinical settings. We present TeamMedAgents, a modular multi-agent framework that translates Salas et al.'s evidence-based teamwork theory into computational mechanisms--shared mental models, team leadership, team orientation, trust networks, and mutual monitoring--enabling Small Language Models to perform multi-step clinical reasoning efficiently. Evaluation across 8 medical benchmarks demonstrates that TeamMedAgents advances the Pareto efficiency frontier by 1-2 orders of magnitude, achieving competitive accuracy at substantially lower token cost than MDAgents, MedAgents, DyLAN, and ReConcile. The framework exhibits the lowest cross-dataset variance among multi-agent approaches, enabling deployment without per-task tuning. Our results establish that theory-grounded coordination mechanisms provide essential scaffolding for deploying efficient medical AI in resource-constrained clinical environments.

arXiv:2509.13471v2 Announce Type: replace-cross Abstract: Large language models (LLMs) show promise for translating natural-language statutes into executable logic, but reliability in legally critical settings remains challenging due to ambiguity and hallucinations. We present an agentic approach for developing legal-critical software, using U.S. federal tax preparation as a case study. The key challenge is test-case generation under the oracle problem, where correct outputs require interpreting law. Building on metamorphic testing, we introduce higher-order metamorphic relations that compare system outputs across structured shifts among similar individuals. Because authoring such relations is tedious and error-prone, we use an LLM-driven, role-based framework to automate test generation and code synthesis. We implement a multi-agent system that translates tax code into executable software and incorporates a metamorphic-testing agent that searches for counterexamples. In experiments, our framework using a smaller model (GPT-4o-mini) achieves a worst-case pass rate of 45%, outperforming frontier models (GPT-4o and Claude 3.5, 9-15%) on complex tax-code tasks. These results support agentic LLM methodologies as a path to robust, trustworthy legal-critical software from natural-language specifications.