What Can Component-Replacement Evidence Establish? A Critical Scoping Review of Local Decisions in LLM Agents
cs.AI
Submitted: 2026-09-30
Updated: 2026-09-30
Sources
- To Call or Not to Call: A Framework to Assess and Optimize LLM Tool Calling
- Ares: Adaptive Reasoning Effort Selection for Efficient LLM Agents
- TraceCompiler: Skill-Guided Mining and Compilation of LLM Agent Traces into Mostly Deterministic Workflows
- Field-Aware Agent Skill Retrieval
- Skill Is Not Document: Query-Conditioned Compatibility for LLM Agent Skill Routing
- Beyond Top- k Skill Retrieval: Diversity-Aware Skill Routing for LLM Agents
- Fail-Fast, Restart-Smart: Early Failure Prediction and Restart for SWE Agentic Tasks
- Agents Don't Paginate: First-Chunk Selection for LLM Tool Responses
- Are Online Skill and Memory Modules Always Worth Their Tokens? A Budget-Constrained Study of Web Agents
- Calibration Is Not Control: Intervention Value for LLM-Agent Oversight
- When May an Agent Stop? Evidence-Carrying Termination for Tool-Using LLMs
- How to Speculate about Uncertainty in Agentic Coding? A Draft-Model Gate Method
- Task- and Session-Level Model Routing: A Common-Interface Hybrid Evaluation of Four Open-Source Routers Across Four Benchmarks
- The Replay Gap: Static Evaluation of Model Switching in LLM Agents Scores the Wrong World
- Scores Are Not Decisions: Cost-Aware Stopping for Tool Acquisition in LLM Agents
- TabAgent: A Framework for Replacing Agentic Generative Components with Tabular-Textual Classifiers
- AgentRouter: Heterogeneous Model Routing for Cost-Optimal Multi-Step Agentic Workflows
- LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems