Let’s be honest – for design, yes. Architecture, no.
Every architect I know is having a version of this moment as the use of AI becomes more prevalent, so here is my thought to this question. AI will replace a great deal of the design production, and it will do this well and cheaply. What AI cannot replace is architecture itself: the judgment, the responsibility, and the human conversation from which a building’s meaning comes.
Design and architecture have never been the same thing, and the difference between them is about to become more clear.
Design is the production of plausible solutions: a floor plan that works, a facade that complies, a set of drawings that council may approve. AI is already good at making plausible decisions that are based on probability; and it will be better next year than it is today.
Architecture begins where plausibility ends. It is the act of deciding which question deserves an answer on this site, for this family, on this street, and of taking responsibility, with our own name and our registration, for what happens next. AI can generate ten thousand options in an afternoon, but it cannot want one of them, and it cannot answer for one of them. The philosopher Alain de Botton observed that we ask our buildings to speak to us of the people we would like to become. AI has never met the people we would like to become.
There is a reason for this, AI is a prediction machine; it works by converging on the most expected answer to the words it is given. It is, by construction, an instrument of the predictable. Architecture is not, because architecture is made from people to people, and people are unpredictable. What make sense as a design choice for one person, does not resonate with the next.
Consider what happens to an architecture brief. Put a client in front of a prompt box and you receive a very basic list: I want four bedrooms, two bathrooms, open plan, something with warmth.
Sit the same client at a table with a human being (an architect) and something different unfolds. The conversation wanders; they relax; they contradict themselves. Like for one of our clients Christine, we learnt that one of their kids is sick and needs special care, so we openly discussed a treatment room, a salt pool for therapy. Or for Rod we discovered that he is a historian and he collects books and he likes to read with certain lighting, preferably sitting overlooking to his second passion, his garden. Or even Amy that likes Japanese culture and wants to see her sculpture collections become a visible part of her house – we designed that collection in the facade of hers home. None of this was in the brief. All of it came out during conversation and while getting to know our clients.
The real brief is subconscious, and it does not exist until a conversation draws it out. The prompt captures what you want. The conversation reveals who you are. A machine executing a dry brief and will hand you a perfect answer to the wrong question, and there is no quantity of computing power that can repair an input which was never typed.
Project homes and volume design are about to become faster, cheaper and more accurate: fewer errors on the drawings, fewer clashes on site, cleaner approvals. But we should be clear-eyed about what is being purchased. A model trained on everything produces the average of everything. Cookie-cutter design is about to become more cookie-cutter than it has ever been, only now without the mistakes. The suburbs of the near future will be better built and more identical at the same time. And when competence becomes free, competence stops being the product. What remains scarce, and therefore precious, is judgment. Nowhere is this clearer than in heritage.
Walk with me into an 1890s terrace and I will show you where AI stops.
The archive of a heritage building is not in a database; it is in the walls. The work begins by reading fabric: lime that has moved with the seasons, timber that warped a century ago and settled, the tool marks of a tradesman whose name nobody recorded. You read these things on site, with your eyes and your fingers and occasionally your nose, and there is no dataset for any of it, because every heritage building is a dataset of one.
Then comes the harder task, which is not reading but judging. Significance is not a pattern; it is a value. Heritage practice asks what matters enough to keep, what may be sacrificed, and what the original architect intended before decades of unsympathetic changes diluted the original idea. These are acts of interpretation and, finally, of custodianship, and someone must answer for them, because in heritage the mistakes are irreversible. A machine has no name on the drawings, no liability and no shame; answering is precisely the job it cannot hold.
Beneath both of these lies the reason closest to my heart.
The beauty of old buildings comes from imperfection: a brick set slightly askew, the gentle wave of a hand-floated wall, what Ruskin called the golden stain of time and what my professors in Turin called the building’s biography. These walls come from an era that placed craftsmanship at the centre of building, and what we respond to in them is the evidence of a person. The most advanced AI cannot replicate this, and not because it cannot draw irregularity; it can generate imperfection all day. What it cannot do is mean anything by it. Imperfection without history is merely error. Imperfection with history is beauty.
So the forecast is this: The mass market will get better and blander at once, and if a good, ordinary building is what you need, the future is bright and inexpensive. But if you own something singular, a terrace, a federation home, a building that exists once, made by hands that existed once, then a human architect is not nostalgia. It is the only instrument delicate enough for the work. Machines perfect the typical; people honour the specific.
AI can replace design. It will not replace architecture.
Your house is specific. Bring it to life with a human.
It will replace much of the design work architects currently sell: plausible plans, compliant facades, routine documentation. It will not replace the judgment, accountability and client understanding that define architecture, because these come from human conversation and human responsibility, which a prediction machine cannot supply.
No, and not for sentimental reasons. Heritage work requires reading original fabric on site, judging cultural significance, and answering for irreversible decisions. Every heritage building is a dataset of one; there is nothing for a model to generalise from, and no one inside it to take responsibility.
In good practices, as a labourer rather than a designer: schedules, checking, documentation, administration, option studies discarded by the dozen. The productive question is not whether architects use AI but where they draw its boundary. Ours sits at the point where judgment begins.
July 30, 2026
Created to help you clarify what matters most in your heritage home before making decisions that can’t be undone.
A calm guide to understanding your options and how to begin, with care.
Receive the Heritage Home Design Workbook
The Design
Workbook for
Heritage Homes
the free guide
DOWNLOAD HERE