Human-Like Budget Reasoning: Imitating Cognitive Strategies for Resource-Constrained Tool Use

Research Question: Does modeling agent planning after human cognitive heuristics for resource allocation (e.g., satisficing, loss aversion) under tool-call budgets yield more robust and interpretable behaviors than purely algorithmic budget awareness?

Hypothesis: Agents augmented with human-inspired budget reasoning heuristics will demonstrate improved task efficiency, adaptability, and transparency, especially in ambiguous or open-ended tasks, compared to agents using only the Budget Tracker or BATS.

Experiment Plan: Conduct user studies to characterize how humans solve information-seeking tasks under explicit resource constraints (limited web searches/time). Formalize extracted strategies (e.g., early stopping, escalation, fallback heuristics) as agent policies. Implement these heuristics in tool-augmented agents and compare with Budget Tracker/BATS baselines on benchmarks like MCPVerse and ToolMind. Evaluate performance, resource use, and collect user preference ratings for interpretability.

References:

• 1. Liu, T., Wang, Z., Miao, J., et al. (2025). Budget-Aware Tool-Use Enables Effective Agent Scaling.
• 2. Lei, F., Yang, Y., Sun, W., & Lin, D. (2025). MCPVerse: An Expansive, Real-World Benchmark for Agentic Tool Use. arXiv.org.
• 3. Chen, Y., Le, R., Xing, Y., An, Z., Chen, Z., Zhao, W. X., Song, Y., & Zhang, T. (2025). ToolMind Technical Report: A Large-Scale, Reasoning-Enhanced Tool-Use Dataset.