This assessment is based on version 1 of this paper. Version 2 is now available on arXiv — the authors may have revised their methods, results, or conclusions.
TO-Agents: A Multi-Agent AI Pipeline for Preference-Guided Topology Optimization
Isabella A. Stewart, Hongrui Chen, Faez Ahmed
Abstract
Topology optimization can generate efficient structures, but designers often must manually translate qualitative intent, such as desired visual style, product experience, or manufacturability into solver settings that are not directly tied to those preferences. We present TO-Agents, a multi-agent AI framework that connects natural-language design intent with iterative topology optimization. The framework converts a human-provided problem description into validated solver inputs, runs a topology optimization solver, renders the resulting 3D topology, and uses multi-view vision-language reasoning with an independent judge agent to critique each result and revise solver parameters. We evaluate the framework on two long-horizon design tasks: a cantilever beam benchmark and a phone-stand product design. In both tasks, the designer specifies an aesthetic preference for hierarchically branched structures inspired by natural tree morphologies, and the system performs four revision cycles across ten independent replicates. TO-Agents produces at least one preference-aligned design in 60% of trials for each case study, corresponding to up to 6x more successful trials than an ablated pipeline without visual or historical feedback. Judge scores and human evaluations show that the pipeline can identify effective parameter levers, recover from poor revisions, and expand design exploration. A manufacturing agent further post-processes top-ranked designs for additive manufacturing, enabling end-to-end intent-to-prototype design. We also identify failure modes, including overshooting, selective memory, misplaced tools, and incorrect parameter reasoning. These results suggest that agentic topology optimization can shift designers from low-level parameter tuning toward higher-level specification of form and function, while highlighting safeguards needed for reliable autonomous engineering design.
AI Impact Assessments
(1 model)Scientific Impact Assessment: TO-Agents
1. Core Contribution
TO-Agents introduces a multi-agent AI pipeline that bridges natural-language design intent with iterative topology optimization (TO). The key innovation is framing TO not as a single-shot optimization with fixed parameters, but as an exploratory, agent-mediated process where the *problem formulation itself* becomes part of the search space. The pipeline consists of specialized agents: a Pydantic agent for input validation, a TO agent wrapping the pyFANTOM solver, a vision agent (Qwen2.5-VL-7B) for multiview analysis and parameter revision, an independent judge agent (Gemma-3-27B) for scoring, and a manufacturing agent for additive manufacturing post-processing.
The system addresses a genuine pain point: designers must manually translate qualitative preferences (aesthetics, bio-inspired morphology) into solver parameters (SIMP penalty, volume fraction, filter radius, mesh resolution) through laborious trial-and-error. TO-Agents automates this translation loop, enabling designers to specify high-level intent ("make it more tree-like and branching") while agents handle parameter tuning.
2. Methodological Rigor
Strengths in experimental design:
Weaknesses:
3. Potential Impact
The work addresses a legitimate gap in computational design: bridging qualitative human preferences with quantitative optimization tools. The potential applications are broad—automotive design, architecture, medical devices, consumer products—wherever topology optimization is used but aesthetic or experiential quality matters.
The end-to-end pipeline from intent to 3D-printed prototype is compelling and demonstrates practical utility. The manufacturing agent's ability to post-process designs for additive manufacturing adds tangible value.
However, the impact is somewhat limited by the narrow scope of tested preferences and the reliance on SIMP-based compliance minimization. More complex multi-physics problems, multi-material designs, or fundamentally different aesthetic criteria would strengthen the case. The framework's dependence on specific open-source models (Qwen2.5-VL-7B, Gemma-3-27B) also raises questions about robustness to model updates.
4. Timeliness & Relevance
This work is highly timely. The intersection of agentic AI, vision-language models, and engineering design is rapidly emerging. Multi-agent frameworks (AutoGen, CrewAI) are maturing, and topology optimization is a well-established but parameter-heavy methodology that could benefit enormously from intelligent automation. The paper positions itself well in the broader trend of using LLMs as reasoning engines for scientific and engineering workflows.
The choice to use open-source models throughout is notable and supports reproducibility, though it may limit performance compared to frontier closed-source alternatives.
5. Strengths & Limitations
Key Strengths:
Notable Weaknesses:
Additional Observations:
Summary
TO-Agents is a well-executed systems paper that demonstrates the feasibility of agentic topology optimization guided by qualitative human preferences. Its primary value lies in the detailed behavioral analysis of multi-agent design exploration rather than in algorithmic novelty. The work opens an interesting research direction but requires broader validation across diverse design preferences, problem types, and user studies to establish significant practical impact.
Generated May 22, 2026
Comparison History (21)
Paper 1 addresses a critical, immediate bottleneck in the booming field of LLM deployment. By mathematically identifying and resolving a systemic measurement bias in widely used benchmarking tools caused by Python's GIL, it provides foundational value to AI infrastructure. While Paper 2 presents an innovative application of AI agents to engineering design, Paper 1 has broader cross-industry implications, ensuring accurate performance evaluation and efficient hardware utilization for production AI systems globally.
MAPLE addresses a fundamental challenge in game-playing AI—extending AlphaZero to imperfect-information games—which has broad implications for decision-making under uncertainty across AI research. It offers a principled algorithmic contribution (aggregating policy evaluations across sampled states) with strong empirical results (significant Elo improvements). While TO-Agents is an interesting engineering contribution combining LLMs with topology optimization, it is more application-specific, relies heavily on existing components (LLM agents, existing solvers), and shows moderate success rates (60%). MAPLE's methodological contribution to a core AI problem gives it broader and deeper scientific impact potential.
Paper 1 is more likely to have higher scientific impact due to its novel, diffusion-specific safety signal (trajectory “hesitation”) and a clear, efficient bi-level routing mechanism with strong empirical validation across multiple datasets and D-LLMs. The work is timely (LLM safety) and broadly applicable to monitoring/guardrailing generative models, potentially influencing both research and deployment practices. Paper 2 is innovative and application-relevant for engineering design, but its evidence is limited to two case studies with moderate success rates and higher system-level brittleness, which may constrain generalizability and near-term uptake.
Paper 2 likely has higher scientific impact: it targets a broadly relevant, timely failure mode in LLM multi-agent systems (epistemic miscalibration) that can affect many domains beyond any single application, offering a general workflow for improving planning reliability. Its framing is conceptually novel and could influence evaluation and design of agentic systems across AI, HCI, and robotics. Paper 1 is innovative and application-rich for engineering design, but its impact is narrower (topology optimization workflows) and appears more system-integration oriented than foundational.
Paper 1 offers a more novel, broadly applicable systems contribution: it formalizes “semantic recoverability” for commitment-sensitive tool-agent execution and provides a principled runtime mechanism (admissibility certification + checkpoint alignment) with safety auditing, addressing a timely reliability gap in LLM agent infrastructures. Its methodological framing and validation suggest impact across agent runtimes, workflow engines, and distributed/transactional recovery. Paper 2 is compelling and application-relevant, but appears more domain-specific, with limited task breadth and moderate success rates; its core ideas are closer to orchestrated multi-agent prompting around an existing solver.
Paper 1 addresses a fundamental methodological gap in the rapidly growing field of autonomous agents by introducing a direct evaluation metric and benchmark dataset. Foundational evaluation tools typically have a broader and more systemic scientific impact across the AI community compared to domain-specific applications, like the topology optimization pipeline presented in Paper 2.
Paper 2 (TO-Agents) addresses a practical, broadly relevant problem—bridging natural-language design intent with topology optimization via multi-agent AI. It has clear real-world applications in engineering design and manufacturing, demonstrates a novel and intuitive framework, and is accessible across multiple fields (AI, mechanical engineering, additive manufacturing). Paper 1 (ECPO) tackles a highly specialized, narrowly defined task (evidence-certified candidate ranking) with a complex formulation that limits its audience and applicability. While methodologically detailed, its niche scope and lack of clear broad-impact applications reduce its estimated scientific impact compared to Paper 2's broader relevance and timeliness.
TO-Agents presents a novel multi-agent AI framework combining LLMs, vision-language models, and topology optimization—bridging generative AI with engineering design in an innovative way. It addresses a timely problem (AI-assisted design), demonstrates real-world applicability (manufacturing pipeline), and has broad impact across AI, engineering, and HCI. Paper 2 proposes a useful benchmark for KG integration pipelines but is more incremental, addressing a narrower community with established evaluation methodology. Paper 1's novelty, interdisciplinary reach, and timeliness give it substantially higher impact potential.
FLUID addresses a fundamental limitation of ID-based recommendation systems for ephemeral content, demonstrating impact at massive scale (1B+ users) with significant production gains. Its novel approach of fully retiring item IDs in favor of multimodal semantic codes represents a paradigm shift in recommender systems with broad applicability. TO-Agents, while innovative in combining LLM agents with topology optimization, addresses a narrower problem with modest success rates (60%) and limited evaluation scope. FLUID's proven industrial deployment and methodological contributions to cold-start and multimodal representation learning give it substantially higher impact potential.
Paper 1 offers higher potential impact due to its broader applicability and resolution of a fundamental bottleneck in LLM agents. While Paper 2 is highly innovative for mechanical engineering (topology optimization), Paper 1 addresses the prohibitive computational cost and latency of the ReAct paradigm in repetitive tasks. By distilling expensive LLM reasoning into reusable RPA scripts and reducing token usage by up to 96%, AutoRPA provides a highly scalable automation solution. This promises massive cross-disciplinary adoption, economic value, and establishes a critical bridge between modern generative AI and traditional enterprise automation.