Replaces heuristic ReAct-style agent loops with a mathematical framework based on control theory to prevent LLM agents from over-deliberating or using excessive tools.
April 1, 2026
Original Paper
The Triadic Cognitive Architecture: Bounding Autonomous Action via Spatio-Temporal and Epistemic Friction
arXiv · 2603.30031
The Takeaway
Current LLM agents often fall into 'infinite loops' or waste compute because they lack a sense of time and cost. This architecture uses an HJB-motivated stopping boundary to decide when the value of further information is outweighed by the cost of delay, significantly improving agent efficiency in time-sensitive tasks.
From the abstract
Current autonomous AI agents, driven primarily by Large Language Models (LLMs), operate in a state of cognitive weightlessness: they process information without an intrinsic sense of network topology, temporal pacing, or epistemic limits. Consequently, heuristic agentic loops (e.g., ReAct) can exhibit failure modes in interactive environments, including excessive tool use under congestion, prolonged deliberation under time decay, and brittle behavior under ambiguous evidence. In this paper, we p