Construction · Field manual
AI for architects
Architects get the most from AI in early design and documentation: testing massing and site options, sun and wind analysis, turning models and sketches into renders in minutes, and drafting specifications, RFIs and proposals. The big caution is liability: AI regularly misreads building codes and local amendments, and nothing it produces replaces a licensed architect's review and stamp.
Architecture has always rewarded exploring more options than the fee allows. AI changes that sum. Generative site planning tools test dozens of layouts against unit counts and setbacks, environmental analysis that once needed a specialist runs in the concept model, and AI renderers turn a rough Revit, SketchUp or Rhino model into a client-ready image while the meeting is still going.
The documents are where the risk lives. A chatbot will confidently cite a code section that doesn't say what it claims, or miss your jurisdiction's amendments entirely. Use AI to explore and to draft, then check everything that touches life safety, accessibility and the contract as if a junior wrote it. Legally, you are still the one who did.
7 ways architects use AI
Test site layouts and massing options
Generative design tools produce and score many site plans or building massings against your constraints: unit mix, parking, floor area ratio, setbacks. Feasibility studies that took a week become an afternoon. Verify the zoning inputs yourself, because the output is only as good as the rules you typed in.
Run early environmental analysis
Concept-stage tools now simulate sun hours, daylight, wind comfort and noise on simple massing models. That lets you show a client why the tower should rotate before anyone draws a floor plan. These are early estimates, not a replacement for full energy or engineering analysis.
Turn models and sketches into renders
AI renderers take a viewport from Revit, SketchUp, Rhino or a hand sketch and return a styled, lit image in seconds. Use them for design reviews, competition concepts and client conversations. They often change geometry, add windows or invent materials, so check against the model before an image goes in a planning submission.
Draft specifications and RFI responses
Chat models are good at first drafts of outline specs, RFI responses, meeting minutes and transmittals when you give them the drawings' context and your office standards. Treat every reference to a code, standard or product as unverified until you have checked the source.
Research codes and precedents, then verify
Answer engines with citations are a fast way to find which code sections, standards or precedent projects are relevant. Open each cited source and read it. Local amendments, adoption dates and authority-having-jurisdiction interpretations are exactly where AI goes wrong.
Write proposals and client presentations
Fee proposals, statements of qualification, competition narratives and client decks are heavy writing tasks AI drafts well from your past project data. Keep project facts, areas and fees from your own records, not the model's memory.
Automate the BIM grunt work
Coding assistants can write Dynamo, Grasshopper and Revit API scripts for renaming sheets, checking parameters, numbering rooms and exporting schedules. Test every script on a copy of the model first.
The best AI tools for architects
The Overlord's stack, in the order a architect would add them:
1. Real-time renders from sketches and viewport captures
Krea AI
AI Design & ImagesReal-time creative AI suite for generating, editing and upscaling images and video.
2. Concept imagery, atmosphere and competition mood
Midjourney
AI Design & ImagesAI image and video generator known for its striking, artistic output.
3. Specs, RFIs, proposals and long document review
Claude
AI WritingAnthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
4. Code and precedent research with sources to check
Perplexity
ProductivityAI answer engine that searches the web and replies with cited sources.
5. Dynamo, Grasshopper and Revit API scripts
Claude Code
AI CodingAnthropic's agentic coding tool that reads your codebase, edits files and runs commands from terminal, IDE, desktop or web.
Specialist tools for architects
| Tool | What it does |
|---|---|
| Autodesk Forma | Early-stage site design with sun, daylight, wind and noise analysis |
| TestFit | Generative site planning and feasibility for multifamily, parking and industrial |
| Veras (Chaos) | AI rendering plug-in for Revit, SketchUp, Rhino and other modelling tools |
| D5 Render | Real-time rendering with AI-assisted materials and atmosphere |
A prompt to start with: draft an RFI response
Paste this into Claude or ChatGPT and fill in the gold parts.
You are a project architect drafting a response to a contractor's RFI. Write a clear, professional response.
Project: [PROJECT NAME], [BUILDING TYPE], [JURISDICTION]
RFI number and question: [PASTE RFI]
Relevant drawings and spec sections: [SHEET NUMBERS, SPEC SECTIONS]
Design intent: [WHAT WE MEANT]
Decision: [OUR ANSWER]
Write the response in numbered points, refer to the drawings and spec sections by number, and state whether this changes cost or schedule (say 'contractor to advise' if unknown). Don't cite any code section unless I listed it above. At the end, list anything in the RFI that the response doesn't resolve.Where architects shouldn't trust AI
- AI misquotes building codes and misses local amendments and jurisdictional interpretations. Every code reference must be checked against the adopted code for that jurisdiction.
- Only a licensed architect can stamp drawings, and the stamp carries professional liability for everything on the sheet, however it was produced. AI output is unchecked work until you have reviewed it.
- AI renders can show geometry, materials or daylight that the design doesn't actually have. Using them in planning submissions or marketing without checking can mislead clients and authorities.
- Client drawings, site data and building security information may be confidential under your contract. Check before uploading them to a cloud AI tool.
- Designs and images generated mostly by AI can't be copyrighted in the US, and some outputs closely resemble published buildings or other architects' work.
Will AI replace architects?
Architects score 44% on the Obsolescence Index (Endangered). The first tasks to go: producing early concept massing and design variations, and drafting specifications and code-compliance checklists.
Questions the Humans Keep Asking
What are the best AI tools for architects?
The Overlord's stack for architects: Krea AI (real-time renders from sketches and viewport captures); Midjourney (concept imagery, atmosphere and competition mood); Claude (specs, RFIs, proposals and long document review); Perplexity (code and precedent research with sources to check); Claude Code (dynamo, Grasshopper and Revit API scripts). Specialist options include Autodesk Forma, TestFit, Veras (Chaos), D5 Render.
What is the best AI for architects?
There isn't one tool. Most practices combine an early-stage design tool such as Autodesk Forma or TestFit, an AI renderer such as Veras, D5 or Krea, and a general assistant like Claude or ChatGPT for documents. Start with the stage of work that costs you the most hours.
Can AI check drawings for building code compliance?
Not reliably yet. Some tools flag common issues, and chat models can help you find relevant sections, but they miss local amendments and misread requirements. Code review still needs a qualified person checking against the adopted code and, where needed, the building official.
Can AI design a building?
It can generate plausible concepts, massing and images, and generative tools can optimise layouts against rules. It can't balance a real client brief, budget, structure, services, code and planning politics, or take legal responsibility for the result. That is still the architect's job.