Shiqi Wang

Jianyuan Guo, Zhiwei Hao, Chengcheng Wang et al.

LLM-based agents mark a shift from passive question answering to active task completion: they perceive environments, invoke tools, maintain state, and act over extended horizons. As agent systems have evolved from prompt engineering to workflows and context engineering, harness engineering, and agent-native training with co-evolution, a central question has become increasingly important: where does the bottleneck in agent performance reside, in the foundation model, in the execution harness, or in the coupling between them? This survey examines LLM-based agents through a model-harness lens. We first clarify the functional definition of agents and the implementation view of an LLM-based agent as a foundation model coupled with an execution harness. We then analyze the limits of model-centric scaling, trace four paradigms of agent engineering, and decompose the execution harness into six coupled runtime responsibilities: observation, context, control, action, state, and verification. Using this decomposition, we map task properties and domain pressures to harness configurations, review benchmark and evaluation practices, and synthesize model-harness evidence on how runtime design affects long-horizon task completion, efficiency, and reliability. Finally, we identify open challenges in value-aware evaluation, safety, harness generalization, and model-harness co-evolution. Rather than treating agents as models with auxiliary tools, this survey argues that agent quality -- including success, efficiency, safety, and generalization -- emerges from the interaction between model capability, runtime infrastructure, task structure, and evaluation design. A collection of papers discussed in this survey is provided in https://github.com/ggjy/Awesome-Agent-Engineering.

Vijay Viswanathan, Shiqi Wang, Devamanyu Hazarika et al.

Despite their widespread use, the role of reward models in shaping reinforcement learning is poorly understood. Reward models offer a tempting promise: they automatically estimate response quality in the absence of verifiers or human judges. Unlike "verifiable rewards" which typically produce binary scores, reward models typically produce continuous scores, allowing them to be sensitive to fine-grained differences in responses. However, we show this apparent strength is a serious weakness: many popular reward models are oversensitive, assigning different scores to equally good responses. Theoretically, we show that seemingly perfect reward models can be highly oversensitive; empirically, this oversensitivity can lead to bad policies. In place of existing notions of "reward model accuracy," we propose evaluating reward models using distinct measures of "discriminative ability" and "specificity" (the complement of oversensitivity). As a solution, we describe a training-free algorithm that uses Monte Carlo dropout on any neural reward model to produce discrete reward clusters. Theoretically, we prove there exist discretizations that reduce oversensitivity at minimal expense of discriminative ability; empirically we show, in both controlled and natural RL settings, that discretizing rewards leads to less reward hacking and better policies than training on the original rewards.

Liqiang He, Yingwen Zhang, Riyu Lu et al.

Optimal frame-level quantization parameter (QP) allocation remains a persistent challenge in modern video encoders. The fixed-QP scheme widely adopted in practical systems is inherently content-agnostic, while classical Lagrangian rate-distortion optimization (RDO) methods often suffer from inaccurate multiplier settings. In this paper, we explore the use of large language models (LLMs) to automatically design RDO heuristics for frame-level QP adaptation. We construct a closed-loop evolutionary framework in which the LLM iteratively proposes RDO heuristics as algorithmic ideas with executable code, and these candidates are evaluated directly through encoding with the Fraunhofer Versatile Video Encoder (VVenC), where each heuristic acts as a scoring function that compares different QP choices based on the encoding statistics of past frames and current candidates. Experimental results across multiple test sets show that the evolved heuristic achieves promising rate-distortion improvements over both the fixed-QP scheme and the Lagrangian baseline. Further analysis reveals that the LLM can autonomously discover an adaptive heuristic that penalizes QP fluctuations via entropy-based terms, providing new insights into the design of RDO algorithms