AI automation · Architecture AI for architects: which practice workflows automation can take over, and what stays with the architect

Typical workflows in an architectural practice

These are typical processes, not client projects. In each, the AI prepares and the architect or project lead checks.

1. Meetings, minutes and action lists

Design team and client meetings produce decisions that have to be recorded and followed up, usually after the meeting from memory and scribbled notes.

• Input: meeting notes or, with everyone’s agreement, a recording, plus the project’s action log.
• AI prepares: a structured draft of the minutes, a list of actions with owner and due date, and updates to the action log, marking anything it could not hear or assign clearly.
• Project lead checks: content and attribution before the minutes are circulated.
• Output: minutes and a refreshed action list, ready to issue after approval.

2. Drawing register and issue sheets

Registers fall behind. Revisions are missing, files are named differently from the register and issue sheets do not match what was sent.

• Input: the drawing register, the exported file list and the issue sheet.
• AI prepares: compares drawing numbers, titles and revisions, flags missing, superseded or misnamed files and drafts the corrected register and the transmittal.
• Information manager checks: every flagged item before anything is issued.
• Output: a checked register and issue sheet, and a list of discrepancies.

3. Site visit reports and snagging

After a site visit, photos, notes and observations have to become a report, and defects have to become a list that can be tracked.

• Input: photos with location tags, voice or text notes and the room or area list.
• AI prepares: assigns each observation to the right location, drafts the report in the practice’s layout and creates the snagging list with photo references, marking observations that need a second look.
• Architect checks: every observation on site knowledge, adds the judgement and issues the report.
• Output: a report draft and a snagging list.

4. Submission packs

Planning and building control submissions fail on missing or inconsistent documents.

• Input: the project documents and the authority’s validation checklist.
• AI prepares: checks that every required document is present, that titles, scales and references are consistent and that statements in the supporting text match the drawings, and drafts parts of the supporting statements from project data.
• Architect checks: the content, the design statements and the compliance claims, and signs.
• Output: a pack with a completeness report. Submission and any certification stay with the architect.

What stays with the architect, and what we do not do

• Design: concept, detailing and the decisions behind them. The AI sees documents, the architect sees the project.
• Compliance and certificates: statements on building regulations, statutory certificates and sign-off are the responsibility of the named professional. Software never signs or submits on anyone’s behalf.
• Copyright and client confidentiality: drawings and client information are put into AI tools only where the client contract and the tool’s terms allow it.
• Fees and contracts: appointments, fee proposals and variations are decided by the principal.
• Decisions about people: we do not build systems that rate or select individuals, such as applicant screening. These are high-risk under Annex III of the AI Act, see risk categories.

Professional and building-safety rules

The table lists the main sources as an orientation. Practices in the UK and Ireland work under different regimes, and Scotland and Northern Ireland have their own rules.

On top of this, the GDPR applies to personal data in project files: a data processing agreement with every provider (Art. 28), servers in the EU or UK and clear access rights per project. Art. 4 of the AI Act asks for sufficient AI literacy among staff, and Art. 50 requires AI chatbots to identify themselves. This overview is general information, not legal advice. Your professional body and your insurer can confirm what applies to your practice.

Systems you probably use

Practices typically model in Revit, ArchiCAD or Vectorworks, draw in AutoCAD, mark up PDFs in Bluebeam, exchange documents in a common data environment such as Asite or Autodesk Construction Cloud and write specifications in NBS Chorus or Word. We connect through exports, the interfaces these systems offer, email and shared folders. The names are for orientation only and do not imply any partnership.

How AI assistants find architects

Clients ask ChatGPT, Gemini or Google’s AI Overviews for “architects for a Victorian terrace extension in Bristol” or “who designs low-energy homes in Galway”. The answer draws on the practice’s own project pages, directories, awards and press, the register of architects and reviews.

• A page per project type or building type: extension, new build, retrofit, conservation, with the typical brief, the process and the planning approach.
• Project pages with substance: location, brief, constraints and outcome. Client names appear only with the client’s consent.
• Register and directory entries that match: practice name, address and architects’ registration, for example in the ARB register and the RIBA or RIAI directories.
• The same details everywhere: website, Google Business Profile and directories.

Is it worth automating? Eight questions

1. Are minutes and action lists written days after the meeting?
2. Does someone reconcile the drawing register with the files by hand before every issue?
3. Do site photos and notes sit in a phone until the report is due?
4. Do submissions come back because a document or a reference is missing?
5. Do you retype data from the model or schedules into reports and statements?
6. Is information on the same project spread over email, the CDE and spreadsheets?
7. Is there a written rule on what may go into AI tools, given client contracts and licence terms?
8. Can you name one task where a draft that needs checking would already save time?

Three or more yes answers usually point to a workflow worth testing. Question 7 is the prerequisite.