Researchers have developed a new method for optimizing the information and tools available to large language model (LLM) agents in mission-critical infrastructure operations. The approach involves identifying optimal harness configurations that match what each task requires with what the harness provides. This is achieved by classifying tasks based on their underlying system representation, ranking harness configurations, and constructing mappings between tasks and harnesses. A new algorithm, map-guided escalation, begins with a task-specific harness and expands only when necessary. The method was evaluated in two scenarios: liquid cooling and power grids. Results showed that the approach can improve accuracy while reducing costs. --- Why it matters: This matters to AI researchers because it addresses a critical challenge in deploying LLM agents in real-world infrastructure operations, where efficiency and effectiveness are paramount. By optimizing harness provisioning, engineers can improve agent performance while minimizing resource waste. Source: https://arxiv.org/abs/2608.17433