← Back to General Economics
econ.GN

Why AI agents quietly become expensive to manage

Muhammad Zia Hydari, Raja Iqbal, Narayan Ramasubbu

May 27, 2026

AI agents that reason, call tools, and adapt create two new problems: accumulated design debt (poorly patched prompts and workflows) and ongoing "stochastic tax" (the cost of keeping probabilistic behavior reliable). These differ fundamentally from traditional technical debt because agents act through real tools and workflows, not just return predictions. The authors propose dashboards and governance controls to make both visible and manageable.
Published as Governing Technical Debt in Agentic AI Systems arXiv:2605.29129
Read the original paper →