What Do CAE Simulation Agents Really Need Beyond a Generic Harness?
Abstract
Computer-aided engineering (CAE) simulation is among the largest and most demanding areas of engineering, where setting up a solver such as OpenFOAM, FEniCS, or COMSOL takes real expertise. Large language model (LLM) agents promise to turn a natural-language request into a working simulation, and recent CAE agents add simulation-specific machinery: multi-agent decomposition, domain retrieval, and scripted reflection. That machinery suited weak base models; modern harnesses already supply multi-turn reasoning, tool use, and execution feedback. We ask what a CAE simulation agent still needs beyond a generic harness. With information access and repair budget held fixed, a single-agent harness matches or beats multi-agent specialized systems (FoamBench 96.4\% vs.\ 88.2\%). Ablations trace this to capabilities the harness already provides: execution-feedback repair lifts FoamBench from 71.8\% with no repair round to 96.4\%, while scripted reflection adds nothing. The one input that still helps is domain knowledge supplied as solver tutorials, our largest measured gain (80.9\% to 96.4\%).
Related
- Reinforcement Learning to Harness Approximation Errors for Long-Time Quantum Simulation
- CP-Agent: A Harness-Engineered Agent for Crystal Plasticity Simulation Workflows
- Beyond-RTA magnetoconductivity of multifold semimetals under generic orientations of electromagnetic fields
- Beyond circular and equatorial orbits: Bayesian constraint on vector charge with the adiabatic inspirals of generic Kerr geodesics
- Rethinking Social Simulation in Open-Population Systems: From Persistent Agents to Communication Processes
BibTeX
@misc{shi2026what,
title = {{What Do CAE Simulation Agents Really Need Beyond a Generic Harness?}},
author = {Jiasheng Shi and Tianhan Zhang},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.03718},
eprint = {2609.03718},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.03718},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - What Do CAE Simulation Agents Really Need Beyond a Generic Harness?
AU - Jiasheng Shi
AU - Tianhan Zhang
PY - 2026
DA - 2026/09/03
PB - arXiv
DO - 10.48550/arxiv.2609.03718
UR - https://arxiv.org/abs/2609.03718
AB - Computer-aided engineering (CAE) simulation is among the largest and most demanding areas of engineering, where setting up a solver such as OpenFOAM, FEniCS, or COMSOL takes real expertise. Large language model (LLM) agents promise to turn a natural-language request into a working simulation, and recent CAE agents add simulation-specific machinery: multi-agent decomposition, domain retrieval, and scripted reflection. That machinery suited weak base models; modern harnesses already supply multi-turn reasoning, tool use, and execution feedback. We ask what a CAE simulation agent still needs beyond a generic harness. With information access and repair budget held fixed, a single-agent harness matches or beats multi-agent specialized systems (FoamBench 96.4\% vs.\ 88.2\%). Ablations trace this to capabilities the harness already provides: execution-feedback repair lifts FoamBench from 71.8\% with no repair round to 96.4\%, while scripted reflection adds nothing. The one input that still helps is domain knowledge supplied as solver tutorials, our largest measured gain (80.9\% to 96.4\%).
ER -