AI-Native Architectural Design

A real estate architectural practice delivers around 50 full design packages a year with a team of roughly 13, working in Archicad from concept through detailed design. The architects are not slow — the workflow around them is. Site and planning information is assembled by hand from several portals, detailed architectural design absorbs two to three months of largely production work, structural and MEP engineering sit in separate organisations and systems, and the architect acts as the manual integration layer between all of them. On the standard permit route a project spends roughly 15–19 months between brief and building permit; only 4–6 of those months are office-controlled, the rest is authority review outside anyone's control. Within the controllable time, five forms of value leakage recur: production effort, rework of 10–15%, manual multidisciplinary coordination, human quality checking and fragmented information.

The workflow is redesigned around AI rather than assisted by it. The site and planning rules become a machine-readable development envelope; the developer brief becomes design rules and optimisation criteria. AI generates and evaluates concept alternatives across saleable area, unit mix, parking, daylight, structural logic, MEP feasibility, compliance and indicative cost. The architect selects the direction. Architectural, structural and MEP information develop as one connected building model; a proposed change is impact-assessed across disciplines and propagated only after professional approval; clashes, inconsistencies and rule issues are flagged continuously; and the coordinated package is issued with version control, traceability and explicit sign-off. Initial targets: concept design same day instead of 3–5 days, a detailed package reviewed in 1–3 business days instead of two to three months, preventable error below 1%, and annual throughput roughly doubling with the same core team — targets to validate through implementation, not results yet claimed.