The abstract of a new arXiv paper (2610.20394) addresses a practical problem: AI agents are being adopted rapidly in software development, but their token costs are climbing, and teams lack clear guidance on how to allocate spending. The authors frame this as a cost-optimization question for mixed teams of humans and agents.

Their proposed method is simulation-based, meaning they model different spending strategies and task assignments to find an optimal division. The stated objective is to determine how teams should divide spending—presumably across tasks and between human effort and agent usage—though the abstract provided does not include results or specific recommendations.

Because only the context and objective are available, the article cannot report findings. The contribution so far is the framing and the simulation approach itself, which could help teams move from intuition to data-driven budgeting for AI tools.