🤖 AI Summary
This work addresses critical limitations in conventional LLM agent loop paradigms—namely implicit dependencies, unbounded recovery, and variable execution histories—which hinder debuggability and controllability. To overcome these issues, the paper introduces SGH, a structured graph framework that, for the first time, integrates classical scheduling theory into LLM agent execution. SGH explicitly models control flow using a static directed acyclic graph (DAG), cleanly separating planning, execution, and recovery into three distinct logical layers. It further incorporates a strict escalation protocol and formal node state machines to enforce rigorous execution semantics. The framework is systematically evaluated across 70 systems, analyzing trade-offs among controllability, expressiveness, and implementability, while providing formal guarantees of termination and correctness. Seven traceable experimental suites are designed to empirically validate its efficacy.
📝 Abstract
The dominant paradigm for building LLM based agents is the Agent Loop, an iterative cycle where a single language model decides what to do next by reading an ever growing context window. This paradigm has three structural weaknesses: implicit dependencies between steps, unbounded recovery loops, and mutable execution history that complicates debugging. We characterize the Agent Loop as a single ready unit scheduler: at any moment, at most one executable unit is active, and the choice of which unit to activate comes from opaque LLM inference rather than an inspectable policy. This perspective places Agent Loops and graph based execution engines on a single semantic continuum. We propose SGH, Structured Graph Harness, which lifts control flow from implicit context into an explicit static DAG. SGH makes three commitments: execution plans are immutable within a plan version, planning execution and recovery are separated into three layers, and recovery follows a strict escalation protocol. These choices trade some expressiveness for controllability, verifiability, and implementability. Our contributions are fourfold: a scheduler unified framework that applies classical scheduling theory to LLM agent execution and identifies challenges introduced by non deterministic LLM nodes; a trade off analysis of controllability, expressiveness, and implementability across 70 surveyed systems; a formal specification including a node state machine with termination and soundness guarantees; and an attributable experimental framework with a seven group design for future validation. This is a position paper and design proposal. We provide a theoretical framework, design analysis, and experimental protocol, not a production implementation or empirical results.