Large language models (LLMs) — AI systems trained on vast text corpora to predict and generate language — have become the most broadly applicable AI technology in architectural practice. Unlike specialized tools for specific tasks, LLMs operate through natural language, making them accessible to any architect who can describe what they need. Their applications in AEC range from the mundane (drafting specification sections, summarizing meeting minutes) to the transformative (reasoning about design alternatives, querying BIM models conversationally, generating parametric code from descriptions).
LLMs as Architectural Assistants
The most immediate application of LLMs in architectural practice is as writing assistants for the substantial documentation burden that architects carry: project descriptions, tender submissions, specification sections, regulatory submissions, client reports, meeting minutes. An LLM provided with the relevant project data — from the BIM model, from meeting notes, from the regulatory framework — can produce accurate first drafts that the architect reviews and refines, rather than writing from scratch.
This application alone — documentation assistance — can save practicing architects 30–50% of their time on certain project phases. The savings compound over a practice: documentation templates, specification libraries, and report structures developed with AI assistance become reusable assets.
LLMs for Parametric Code Generation
The capability that most expands computational design access is LLM-based parametric code generation: describing a Grasshopper definition, a Python script for Rhino, or a Dynamo definition in natural language, and receiving working implementation code. This removes the syntax barrier that has kept parametric design a specialist skill.
The professional reality: current LLMs produce code that works 60–80% of the time for well-specified parametric problems. The architect must be able to review the code for logical errors and modify it when needed. This requires understanding what the code should do, not necessarily how to write it from scratch — a much more accessible skill level. MIAW Module F0 builds this foundational literacy explicitly.
Conversational BIM and Knowledge Retrieval
One of the most architecturally specific applications of LLMs is conversational query to structured data: asking questions in natural language about a BIM model, a regulatory database, or a project knowledge base, and receiving answers that integrate information from multiple sources. 'What structural members in the north wing are within 2 meters of the revised floor plan boundary?' 'Which specifications sections need updating for the change from concrete to steel frame?' 'What does the building code require for the egress path through the lobby?'
These questions today require navigating complex software interfaces or searching through documents manually. With an LLM connected to the relevant data sources via MCP (Model Context Protocol) or RAG (Retrieval-Augmented Generation), they become one-step natural language queries.