Cost-Conscious Autonomous Agents for Scalable Intelligence Systems
Year: 2026
Type: conference-paper
Abstract: The swift emergence of autonomous agents that utilize large language model (LLM) technology has helped to reveal cost as a critical performance bottleneck due to high token costs, recursive reasoning, latencies, as well as the cost of external tool utilization. Although reactive solutions utilizing cost-aware orchestration, model routing, and market-based coordination have been proposed to help address these performance cost factors, the solutions largely address cost as an external programmatic constraint, which means agents are still economically ignorant.This paper presents a framework for economically self-aware AI agents that considers costs, budgets, and utility in the agent's decision-making process. The proposed framework allows an agent to estimate costs, assess task utility under specific budget limits, and adjust its reasoning depth. It can also delegate tasks or wisely refuse tasks that may lead to negative expected value. We will discuss an agent architecture that incorporates economic awareness into task intake, planning, and execution with feedback. Based on this, a formal cost-utility formulation helps in choosing the best strategy when resources are limited. This work promotes a move away from externally controlled cost systems towards an autonomous agent that can manage accuracy, speed, and cost, supporting scalable and sustainable intelligent systems.
Institution: AISSMS Institute of Information Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7214939898