Can AI Co-Design Structures? An AI Agent for Early-Stage Structural Design
Early-stage structural design is an iterative, creative process that balances form, force, and material. This paper proposes a proof-of-concept workflow connecting a large language model agent to a COMPAS-based structural modeling API via a Model Context Protocol (MCP) server, enabling natural-language generation and editing of semantic building elements. By producing explicit graph-based artifacts rather than generic meshes, the work aims to reduce modeling friction and move toward human-AI co-design in structural engineering.
The paper presents an architecturally sound but deliberately preliminary proof-of-concept. It successfully demonstrates that an LLM agent can drive a domain-specific structural API to generate editable building models from natural language, including progressive modifications. However, the evaluation is limited to five qualitative examples with no formal metrics, correctness guarantees, or comparison against baselines, making it a tentative first step rather than a validated system.
The strongest aspect is the architectural decision to reject generic text-to-3D meshes in favor of semantically rich structural objects. By using COMPAS graph data structures with explicit Beam, Column, Slab, and Joint classes, the generated artifacts remain inspectable, serializable, and compatible with downstream AEC workflows. The progressive modeling sequence in Figure 4 also convincingly illustrates how stateful agent interaction can mimic real iterative design practice.
The evaluation is purely anecdotal: five hand-picked prompt examples, one editing sequence, and one progressive sequence, with no quantitative measures, robustness tests, or user studies. The authors openly acknowledge that the generated models lack structural correctness, code compliance, and analysis readiness.
More critically, the demonstration in Figure 3 reveals concrete geometric failures—an LLM-induced slab overlap—that exposes the agent's limited spatial reasoning and underscores the need for careful human verification.
The qualitative evidence supports only the claim that language agents can invoke structural API calls, not that they reliably produce good or correct designs. Comparisons to text-to-3D and CAD-generation literature are fair in identifying the need for semantic objects over meshes, but the paper does not empirically compare its approach against script-based generation, parametric templates, or even manual modeling effort. Without such baselines, the practical advantage of the proposed workflow remains unquantified.
Reproduction would be severely hindered by the absence of released code, prompts, hyperparameters, or the exact MCP server configuration. The agent relies on the proprietary Google Gemini 3.1 Pro API, and the paper discloses no details about temperature, system prompts, or retry logic. While COMPAS is open source and MCP is an open protocol, an independent researcher cannot replicate the demonstrations without significant engineering guesswork.
Recent advances in large language models and agentic workflows offer a new paradigm to early-stage structural design translating design intent into structural models. Although text-to-3D approaches have demonstrated that natural language can be used to produce visual, mesh-based geometric models, structural design requires domain-specific geometric arrangements and semantically rich data types. This paper introduces a proof-of-concept workflow for prompt-to-structure generation using an AI agent connected to a COMPAS-based structural modelling API through a Model Context Protocol (MCP) server. Through controlled modelling operations, the agent creates and modifies a graph-based structural data model composed of explicit building elements. The resulting artifacts can be inspected, modified, serialised, and further connected to downstream AEC workflows. The approach is demonstrated through selected prompt-to-structure examples and by progressive and editable design sequences that resemble typical iterative structural design workflows. The results are preliminary, but demonstrate the possibility of combining natural language, agentic workflows, and domain-specific structural data structures as a first step toward AI-assisted structural co-design.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.