What an AI agent for construction actually does (and what it doesn't)

How is an agent different from a chatbot or a CAD plugin?

The difference is who holds the pencil. With an agent, you describe the result you want and check it; the agent does the steps in between.

What does Stud's agent do, step by step?

1. Reads the plan you attached: DXF, PDF, plan image or sketch.
2. Builds the building in 3D: levels, walls, openings, floors and roofs.
3. Checks the model after each step and tells you what still needs a look.
4. Gives the quantities read off the model.
5. Splits it into panels and frames them, in light gauge steel or timber, when you ask.
6. Prepares the production files: machine files, IFC (the open BIM exchange standard) and production PDFs.

You see each step as it happens: the 3D model updates while the agent works.

Why does it ask instead of guessing?

Because a wrong dimension costs more than a question. A plan view rarely shows wall heights or thicknesses. An agent that fills them with "standard" values gives you a model that looks right and prices wrong.

What doesn't it do?

• Price the job. It gives quantities; your rates make the price.
• Sign off the structure. It does not run structural calculations or check building codes.
• Read every plan perfectly. Clean, dimensioned plans work best. Busy real-world sets mean more questions and a closer check.
• Decide for you. When a choice is yours (the roof, the system, a dimension), it asks.

How do you stay in control?

• You approve each step. By default the agent stops at the building and proposes the next step: panels, then framing.
• You point at what to change. Click an element in the 3D view and write the change.
• It keeps your requirements. It notes the requirements of a project in its brief and checks every model against them.
• It works your way. Your instructions apply to every project, and your skills hold your methods.