AI in BIM and construction
Where does AI actually help today?
Language models, computer vision, prediction and generative design: what each does well today in design, construction and operations, how MCP connects assistants to BIM tools, and how to use them safely.
Understand
What it is, in plain language
- Pedia · WorkflowsAI in construction and BIM (what works today)A practical map of where AI helps across design, construction and operations today, where it is still a demo, and the questions to ask before using it on a project.
- Pedia · SoftwareMCP (Model Context Protocol) for BIM toolsThe open protocol that lets AI assistants use tools and data, and what MCP servers for Revit, Autodesk's cloud, IFC and other BIM software make possible, and risky.
- Pedia · WorkflowsGenerative and computational designDefining a design problem as goals, rules and variables so software can generate and compare many options, and what it needs from BIM data to be useful.
- Pedia · WorkflowsProgress tracking with models and reality captureMeasuring what has actually been built, against the model and the programme, from site updates, scans, photos or drones, so delays show up in weeks rather than months.
- Pedia · SoftwareAutodesk Platform Services (APS)Autodesk's cloud APIs (formerly Forge) for reading, viewing, converting and automating models and project data in Autodesk's cloud, including Autodesk Construction Cloud.
Learn
Short lessons with a quick check
- Lesson · 3 min · quizAI in BIM: what works todayLanguage models, computer vision, prediction and generative design each have real uses today. The useful question is which task, with which data, checked by whom.
- Lesson · 3 min · quizMCP and AI assistants in BIM toolsMCP lets an AI assistant call a BIM tool's actions: read elements, set parameters, create sheets. Powerful for questions and reviewed edits, risky without permissions and checks.
How it works
The process, at a glance
- 1You ask“Which fire doors on Level 3 have no rating?”
- 2Assistant plansChooses the server’s tools to call
- 3MCP server actsCalls the software’s API: get_elements
- 4Results returnElements and properties, as data
- 5You checkSpot-check before trusting or approving changes
Common mistakes
What goes wrong most often on real projects.
- Using public AI tools with confidential project documents.
- Treating confident answers as correct answers.
- Buying an 'AI' tool before identifying the task and the data.
- Letting an assistant edit live models without review.
Is it in place on your project?
Tick what is done — it counts towards the checkpoint on your learning path.
0 / 4 in place
Keep going