Electric bus charging and vehicle-to-grid coordination require timely responses to operational disruptions and electricity market conditions. However, evidence remains limited on when agentic supervision improves optimization-based coordination and how its pricing objectives distribute flexibility value between aggregators and public transport operators (PTOs). This study evaluates an agentic aggregator combining three language-model agents for disturbance detection, tariff selection, and schedule evaluation with a mathematical optimizer that enforces fleet operating constraints. A bus depot case study examines day-ahead and real-time operation under synthetic delays, energy deviations, price shocks, and combined disturbances, comparing two coordination objectives (profit-based and operations-based). To assess its value, the agentic layer is compared with a matched rule-based workflow and a two-stage stochastic program. In day-ahead operation, operations-based coordination yields a PTO cost of EUR 116.08/day and an aggregator margin of EUR 2.41/day; profit-based coordination increases aggregator revenue to EUR 20.91/day while raising PTO cost to EUR 138.41/day. In real-time operation, operations-based coordination yields lower PTO costs in all 14 disturbance scenarios; profit-based coordination increases aggregator revenue in all 14 while raising PTO cost by EUR 54.12 on average, against EUR 31.64. Across both benchmarks, agentic supervision shows no uniform economic dominance: under profit-based coordination, it generates higher aggregator revenue in two of three cases and lower revenue in the third, while operations-based outcomes remain equivalent within solver tolerance. Its advantage is limited to converting unstructured operator messages into scheduling inputs and to time-varying pricing. Deployment requires transparent pricing objectives, auditable tariff decisions, and explicit value-sharing rules.
