When Agents Meet Electric Bus Fleet Operations: Pricing Behavior, Trade-offs, and Policy Implications in an Aggregator Framework

summary

The gist

Agentic systems are changing how complex operational tasks are coordinated by connecting heterogeneous data sources, introducing a new paradigm for simplifying tasks, connecting datasets…

In short

This paper introduces an agentic framework for coordinating electric bus fleets with power grids. It uses supervisory agents to handle disturbances and adapt pricing strategies based on whether the goal is profit or operational flexibility. The results show that choosing the right coordination mode significantly impacts cost savings for the fleet, leading to policy recommendations for regulated V2G markets.

Key concepts

Agentic Aggregator Framework
A system where specialized AI agents supervise an optimization model. These agents detect changes (disturbances), decide when to re-optimize, and adapt pricing based on profit or operational needs. It connects the bus fleet, grid, and market data to simplify complex scheduling decisions.
Profit-based Mode
A coordination strategy where the system prioritizes maximizing revenue for the aggregator. This mode allows pricing agents to propose higher charging prices when the PTO buys energy and lower compensation when it exports energy, focusing on financial gain.
Operational-based Mode
A strategy prioritizing flexibility participation for the Power Take-Off (PTO). This mode favors lower charging prices and higher V2G compensation multipliers. It aims to reduce the PTO's exposure to electricity price markups and return more value directly to the bus fleet.
Prompt Sensitivity Analysis
This study investigates how changing the instructions (prompts) given to an AI agent affects its behavior. It shows that prompt design is a governance variable, meaning how you ask the agent influences whether it pursues profit or operational objectives.

Terminology used across episodes

This episode discusses

The paper

When Agents Meet Electric Bus Fleet Operations: Pricing Behavior, Trade-offs, and Policy Implications in an Aggregator Framework · Read on arXiv

Department of Civil Engineering at McGill University · INESC Coimbra, University of Coimbra

DOI: 10.1016/j.trc.2026.106105

Transcript

Introduction to the show: ident: AI Radio. Generated commentary on the latest Artificial Intelligence papers.

Tom: We're looking at the paper: "When Agents Meet Electric Bus Fleet Operations".

Jane: Here's the gist of the paper. Agentic systems are changing how complex operational tasks are coordinated by connecting heterogeneous data sources, introducing a new paradigm for simplifying tasks, connecting datasets, and automating processes<ref:2606.26400#pg2> Electric bus fleets provide a relevant test case for this paradigm<ref:2606.26400#pg2> This paper proposes an agentic aggregator framework that streamlines this decision environment by coupling an optimization-based electric bus scheduling model with supervisory agents for disturbance detection, tariff adaptation, and schedule evaluation<ref:2606.26400#pg2> The results show that agentic aggregation can support adaptive fleet-grid coordination by maintaining feasible schedules, activating re-optimization selectively, and improving the use of charging and V2G flexibility<ref:2606.26400#pg2> However, they also reveal a critical trade-off: the same agentic capability that reduces operational complexity can extract value from the PTO when configured around profit-oriented pricing<ref:2606.26400#pg2> Framework Architecture The proposed methodology is organized around the separation between agentic supervision and optimization-based feasibility enforcement<ref:2606.26400#pg2> The framework is organized around three coupled components: the fleet and charging system, the grid and market environment, and the multi-agentic aggregator layer that coordinates the interaction between them<ref:2606.26400#pg2> On the fleet side, the framework receives operational information, including bus state-of-charge (SOC), battery limits, charger availability, trip assignments, departure and arrival times, and disturbance information related to timetable delays and route-energy deviations<ref:2606.26400#pg2> On the grid and market side, the framework receives electricity prices, power limits, and V2G flexibility opportunities<ref:2606.26400#pg2> The aggregator layer acts as the coordination interface between the fleet and the grid<ref:2606.26400#pg2> In the day-ahead (DA) workflow, the Pricing Agent translates the selected coordination mode into structured pricing guidance, and the Evaluator Agent assesses the resulting schedule against economic and operational criteria<ref:2606.26400#pg2> Coordination Modes and Value Allocation Two coordination modes are compared: a profit-based mode that prioritizes aggregator revenue and an operational-based mode that prioritizes PTO-compatible flexibility participation<ref:2606.26400#pg2> In the profit-based mode, the Pricing Agent is allowed to propose higher charging-price multipliers when the PTO buys energy and lower V2G compensation multipliers when the PTO exports energy<ref:2606.26400#pg2> In the operational-based mode, the Pricing Agent therefore favors lower charging-price multipliers and higher V2G compensation multipliers, reducing the PTO’s exposure to electricity-price markups and returning more value to the fleet when it provides flexibility<ref:2606.26400#pg2> The agentic layer can shift between these modes through prompt configuration alone, with no change to the optimizer, the contract, or any externally visible system parameter<ref:2606.26400#pg2> Disturbance Response and Economic Outcomes The RT evaluation examines how the multi-agentic aggregator responds when observed operating conditions diverge from the accepted DA reference plan, s∗DA<ref:2606.26400#pg2> Across the fourteen matched RT scenarios, operational-based operation lowers PTO cost in every case relative to profit-based operation<ref:2606.26400#pg2> The largest PTO reductions occur under P + 50 (-116.07 EUR), C-All 5–25 (-62.48 EUR), C-Seq (-55.35 EUR), and P + 25 (-55.05 EUR)<ref:2606.26400#pg2> The results show that the coordination mode, not the disturbance, governs how flexibility value is split between the aggregator and the PTO<ref:2606.26400#pg2> Prompt Sensitivity Analysis The prompt-sensitivity analysis further shows that agentic pricing behavior changes with prompt configuration, making prompt design a reproducibility and governance variable<ref:2606.26400#pg2> Under profit-based behavior, prompt design strongly affects the agent’s ability to express the intended aggregator objective<ref:2606.26400#pg2> The FS+CoT rows therefore correspond to the accepted profit-based and operational-based DA files used in the main DA analysis<ref:2606.26400#pg2> These experiments support the interpretation that the proposed framework is not only an optimization wrapper, but an agentic decision system whose behavior depends on how economic objectives are communicated to the agent<ref:2606.26400#pg2> Policy Implications The aggregator domain cannot be left to unconstrained market behavior if electric bus fleets are to participate in V2G markets on sustainable terms<ref:2606.26400#pg2> Three findings established above support this conclusion<ref:2606.26400#pg2> These findings motivate three regulatory measures for V2G markets involving public fleets: (i) bounds on aggregator tariff margins, or mandated value-sharing rules, so that PTO cost exposure cannot exceed a defined level regardless of the aggregator’s objective<ref:2606.26400#pg2> (ii) transparency requirements covering the coordination mode and, for agentic systems, the prompt configuration and constrained-output specification that govern pricing behavior, so that the chosen objective is auditable<ref:2606.26400#pg2> and (iii) reporting standards requiring disclosure of realized tariff vectors, trigger decisions, and value allocation<ref:2606.26400#pg2> The paper proposed a multi-agentic framework for electric bus fleet-grid coordination under DA and RT operation<ref:2606.26400#pg2> The contribution is therefore a structured decision pipeline that separates disturbance detection, tariff adaptation, and schedule acceptance while preserving the transparency of an optimization-based scheduler<ref:2606.26400#pg2> The results show that prompt design affects whether the Pricing Agent expresses the intended economic behavior<ref:2606.26400#pg2> Future agentic energy-management studies should therefore report prompt paradigms and constrained-output assumptions alongside optimization inputs and results<ref:2606.26400#pg2> The main methodological lesson is that RT fleet-grid coordination should report not only aggregate cost and revenue, but also the trigger timing, accepted tariff vectors, SOC trajectories, and V2G exchange pathways through which disturbances reshape operational outcomes<ref:2606.

Title and authors: Jane: So, to get into that coordination aspect, let's talk about who wrote this. The paper is titled "When Agents Meet Electric Bus Fleet Operations: Pricing Behavior, Trade-offs, and Policy Implications in an Aggregator Framework," and the authors are Jônatas Augusto Manzolli, Ali Eslami, and Luis Miranda-Moreno.

Tom: They’re from McGill University and INESC Coimbra at the University of Coimbra. The title tells you they aren't just doing some abstract optimization; they're focusing on pricing behavior and what that means for policy when using these agentic systems in real bus fleets.

Lu: The authors are tackling a very practical problem: how to structure the interaction between a fleet, the grid, and the aggregator so that it works well operationally while also considering economic trade-offs.

Meng: I wonder if they really managed to balance those two things—the operational feasibility and the economic profit—without one completely overshadowing the other in their model.

Lalam: They are looking at how agents can change the way we share flexibility value between the aggregator and the PTO, which is a key area for real-world deployment.

The paper's summary: Jane: So, what’s this framework actually doing in detail? It proposes an agentic aggregator framework that connects an optimization-based electric bus scheduling model with supervisory agents for three main jobs: detecting disturbances, adapting tariffs, and evaluating the schedule.

Tom: It streamlines the decision environment by linking data from the fleet—like battery state-of-charge, charger availability, and trip assignments—with info from the grid and market signals like electricity prices.

Lu: The framework works by having these three components interact: a fleet side with operational info, a grid side with market info, and this multi-agentic aggregator layer that coordinates everything in between.

Meng: So for me, the most important part is that this coordination layer is what translates the abstract grid needs into concrete charging and V2G actions that the actual buses can take.

Lalam: It’s not just about scheduling; it’s about defining how the economic value of flexibility gets split between the aggregator and the PTO based on these continuous market and grid signals.

The paper's improvements: Tom: They compare two main coordination modes: a profit-based mode that prioritizes making money for the aggregator, versus an operational-based mode that prioritizes flexibility participation compatible with what the PTO needs.

Jane: In the profit-based mode, the pricing agent can suggest higher charging prices when it buys energy or lower V2G compensation when it sells energy. That’s a big difference in how they value different types of flexibility.

Lu: Then there’s this crucial comparison: an operational-based mode favors lower charging prices and higher V2G compensation, which reduces the PTO’s exposure to price markups and returns more value to the fleet when it provides flexibility.

Meng: That sounds like a significant improvement for a service provider because it directly addresses their cost exposure based on how they participate in the grid.

Lalam: And what I like about this is that you can switch between these two modes just by changing the prompt configuration, without having to change the core optimizer or any external contracts, which makes it very flexible for deployment.

Conclusion: Jane: So wrapping up, they've shown how this agentic aggregation supports adaptive fleet-grid coordination by maintaining feasible schedules and activating re-optimization selectively. They also look at how prompt design affects the pricing agent’s behavior, which is important for governance.

Tom: The main finding is that the coordination mode, not just any single disturbance, governs how flexibility value gets split between the aggregator and the PTO across things like delays or price changes.

Lu: They also found that prompt design really matters; how you communicate those economic objectives to the agent directly affects whether it expresses profit-based or operational-based behavior.

Meng: From a deployment standpoint, this means we need transparency around which mode is active and what prompt configuration is being used so that we can actually audit the value allocation for the PTO.

Lalam: And to bring it home, they suggest that if electric bus fleets are going to participate in V2G markets sustainably, there needs to be regulation around aggregator tariff margins or mandated value-sharing rules so that the PTO cost exposure stays within a certain level regardless of what the aggregator’s objective is.

Jane: So, in short, this paper on "When Agents Meet Electric Bus Fleet Operations: Pricing Behavior, Trade-offs, and Policy Implications in an Aggregator Framework" gives us a structured way to manage bus fleets using agents that can adapt their pricing strategy based on whether they are trying to maximize profit or support the operational needs of the fleet.

More episodes

← Home