AI agents often execute delegated tasks in parallel while a single overseer monitors several of them. In practice an agent may lack required information such as an unstated preference, constraint, or specification. It can either ask the human for clarification or search for the information itself, which consumes time and may fail. We introduce a stylized model where each agent, when facing missing information, can choose to search, ask, or continue parts of its task that do not depend on the missing data, and we formulate the coordination problem as a dynamic program. We first show that the decision to ask depends on more than the agent's own uncertainty: an agent with additional work may benefit from asking earlier, and the order in which the human answers questions depends on all pending queries, so a per‑agent ranking is suboptimal. To address this we propose the Pairwise Coordination Policy (PCP), which uses exact solutions of one‑task and two‑task subproblems to estimate how pairs of tasks delay each other when sharing the human. Theoretically, PCP is exactly optimal when at most two requirements remain unresolved, and its cost is at most twice the optimum when every unresolved requirement must be resolved by the human. We evaluate PCP on 864 synthetic cases and in a case study with LLM agents. On the synthetic benchmark PCP’s mean gap to the optimal cost is 0.07% and it reduces cost by an average of 18%–31% compared to three standard ask‑when rules. In the LLM case study all tasks were completed correctly and PCP achieved the lowest mean cost across all tested settings.\n\nReview: The paper rigorously uncovers the intertwined nature of ask‑timing decisions in multi‑agent settings with limited human attention, and the proposed PCP demonstrates near‑optimal performance both analytically and empirically, offering a practical scheduling tool for real‑world AI systems.