Line of Site

By Michael Smith Architect / SRHD Founder

A few years ago, I was fairly confident that architects would not be replaced by ai.

Architecture was creative. It was human. It involved judgement, taste, memory, culture, compromise and care. It was not just production. It was not just information. It was not just drawing. It was the patient work of understanding people and place, then turning that understanding into something that could be built.

Surely, I thought, this was not the kind of work that artificial intelligence would reach.

I no longer think that.

In fact, many of the things architects assumed would protect us are already proving to be weak elsewhere. Creativity has not been a moat. Responsibility has not been a moat. Complexity has not been a moat. And professional pride, while understandable, is not a strategy.

This is not an argument that architects are about to disappear. It is not an argument that machines can replace the full depth of architectural judgement, or the human responsibility that sits behind a building. But it is an argument that the architecture profession is badly underestimating the speed, breadth and direction of change.

From the conversations I have been part of, much of the profession seems to be either blissfully unaware, quietly in denial, or simply disengaged from what is happening in artificial intelligence. Architects often see our work as unique. General rules somehow do not apply to architecture because we see things differently, or because our job is different, or because buildings are too complex for a machine to understand.

Others are in the brace-for-impact position. They can see something coming, but the conclusion seems to be that there is little point doing much about it. If AI comes for architecture, then presumably tens of thousands of other white-collar jobs will already have been transformed before ours. So we can just wait. We can see what happens. We can deal with it later.

That worries me.

Waiting is not neutral. Waiting means other people define the tools, the workflows, the business models and the rules. Waiting means the profession only notices the change once the market has already moved.

The future is not fully visible. But it is visible enough to act.

Creativity was not protection

One of the first assumptions to fall was the idea that creative work was safe.

A few years ago, it felt reasonable to believe that creativity would protect us. Architecture is not a spreadsheet. It is not repetitive clerical work. It deals with atmosphere, proportion, light, material, street, memory and human life. It requires judgement and imagination.

But AI did not begin by replacing only dull administrative tasks. It hit creative production early.

Not all creative production. Not the highest cultural work. Not the work of artists with a singular voice. But the commercial middle has already been shaken. Corporate illustration, website graphics, marketing images, presentation artwork, mood imagery and generic visual content can now often be produced by AI quickly, cheaply and at a level that is good enough for many clients.

That matters because most creative economies are not built only on masterpieces. They are built on commercial work. If that work evaporates, the impact on graphic designers, illustrators and similar artists is enormous.

Architects should be careful about saying, “AI cannot do great architecture.”

That may be true, and still not protect architectural labour.

The better question is: how much ordinary, commercially useful architectural work can AI do well enough?

Life safety was not protection either

Architecture also assumes it is protected because buildings affect life safety.

That is true. Buildings can hurt people. Bad decisions can have serious consequences. Drawings matter. Specifications matter. Details matter. There are real risks in the work.

But cars affect life safety too, and AI is already driving cars on public roads in the United States.

Life safety does not stop AI. It changes how AI enters. It enters more slowly, more controversially and with more need for supervision, but it still enters.

The architectural equivalent is probably not an AI independently signing off a building tomorrow. It is more likely to be AI drafting a detail, checking a clause, comparing products, reviewing a drawing, writing a specification, flagging a risk or producing a first-pass documentation set that a human then reviews.

The architect remains responsible. But the work around that responsibility changes.

The profession that borrowed our name

There is another warning sitting in plain sight.

Writing code was a top priority for the major AI companies. If AI could write code, it could help develop software. If it could help develop software, it could help build the next generation of AI systems. Code was not just another professional target. It was strategic.

We are now at the point where AI can write, test, debug and improve code in ways that are already changing software development.

Software engineers once looked like some of the safest workers in the economy. Their work was abstract, technical, difficult and commercially valuable. It was not merely typing. It required systems thinking, judgement, dependencies, interfaces, risk and long-term structure.

It was so analogous to architecture that the software industry borrowed our language. They created the role of the software architect.

The profession that borrowed our name may be showing us our future.

If AI can transform software architecture, it is hard to believe building architecture is protected simply because it is complex.

Architectural thinking has become executable

This was the point that made me start pulling at threads.

What is now possible that previously was not?

Oddly enough, once AI can write code, a one-way bridge is built between architectural thinking and software production. Previously, an architect could have a good idea for a tool, a workflow or a system, but the barrier was code. You needed developers, budgets, specifications, project management and months of translation between professional intent and software execution.

That barrier is now much lower.

Architects are trained to think in systems. We ask what information matters. We ask what the workflow should be. We identify constraints, risks, exceptions and edge cases. We understand that an output has to be legible to a user. We know that a process has to produce something that can be trusted.

Those are also software questions.

The architect no longer needs to become a software engineer to build software. They need to think clearly enough for AI to build from their intent.

AI has made architectural thinking executable.

That changes the question from “What will AI do to architecture?” to “What can architects now build that we could never have built before?”

We noticed the image first

As architects began to notice AI, many naturally looked at the visual side.

That makes sense. We are image-literate. We understand atmosphere, materiality, light, form and seduction. So we have seen an explosion of AI imagery: cinematic houses, impossible interiors, speculative towers, fantasy masterplans and enchanting scenes pretending they could be buildings.

Some of it is impressive. Some of it is useful. Some of it is just slop.

But the deeper opportunity may not be the image. It may be the system behind the work.

Can AI help us throw a net over architectural knowledge? Can it sit above the messy world of codes, products, planning controls, specifications, warranties, practice notes, emails, meeting notes, drawing sets and supplier information? Can it become a chat interface, an email interface, a research assistant, a checking system, a practice memory?

That is less glamorous than cinematic imagery, but it may be far more disruptive.

Meanwhile, the image problem is already arriving at the front door of practice. Clients are beginning to come to architects with AI images. Not sketches. Not magazine clippings. Not Pinterest boards. Fully formed, highly persuasive images of what their project could be, or what they want it to be.

How should an architect respond to that?

Is the image a brief, a fantasy or a design direction? Who owns it? How do you explain feasibility without seeming defensive? How do you protect authorship? How do you charge for translating an AI image into buildable architecture? How do you manage expectations when the image has no structure, budget, planning pathway, construction logic or relationship to place?

There is no practice note for this yet.

It has not been written because the practice problem has only just arrived.

The AI draftsperson is now in sight

As I kept pulling at the thread, one possibility became harder to dismiss: the AI draftsperson, or perhaps the AI draftscomputer.

By that I do not mean that a perfect product already exists. I mean we are no longer waiting on a completely undiscovered technology. We can see how the pieces fit together.

AI can read images and drawings. AI can write code. AI can operate software. AI can search documents. AI can interpret instructions. AI can generate structured outputs. AI can compare information. AI can find inconsistencies. AI can use tools.

The AI draftsperson is not magic. It is integration.

So we have to ask the uncomfortable question now.

What happens when the cost of producing documentation drops off a cliff?

What happens if a project can move from a town planning set to a full working drawing package in a couple of days?

Architecture has always assumed that documentation is slow because buildings are complex. Some of that is true. But some of it is slow because humans are manually translating decisions into drawings, schedules, notes, references, details and specifications.

If AI can perform large parts of that translation, the bottleneck moves.

The scarce skill may no longer be drawing production. It may become knowing what should be drawn, what should not be accepted, what matters, what is risky, what is buildable, what is wrong and what the AI has misunderstood.

When documentation becomes cheap, judgement becomes more valuable. But only if architects can clearly define and price that judgement.

From laughable to negligent

We have already seen a pattern in other areas of work.

First, people say: “Imagine letting AI do that.”

Then they say: “This practitioner is incompetent for allowing AI to do that.”

Then: “AI can do that, but only if you are careful and check everything.”

Then: “AI can do it well. You mainly need to spot the errors.”

And eventually: “AI can do this better than humans. It may be negligent not to use it.”

That final stage is the one architects needs to think about carefully.

At first, AI use looks unprofessional. Later, not using AI may look unprofessional.

If AI tools can identify missing fire-stopping notes, inconsistent door schedules, contradictory specifications, unresolved access issues, weak waterproofing details or non-compliant assumptions, then the standard of care may shift.

Professional negligence is partly defined by what a reasonable practitioner could have known or done at the time. AI may change that standard faster than the profession is ready for.

A different workflow

Architecture needs to adapt, and adaptation takes time. We need to start now.

One possible future workflow looks very different from the current model.

The first stage is enhanced briefing.

Architects already do this well, but in an AI-enabled process the brief becomes more important, not less. If parts of the process are automated, the quality of the input becomes critical. A vague brief will produce fast mediocrity. A strong brief becomes the architect’s first act of authorship.

There may also need to be a second brief: the brief of place.

Our highest value may be in protecting the built environment from AI slop. Not just bad images or bad drawings, but the homogenisation of place. If AI systems are fed generic precedent, generic imagery and generic market logic, they will tend toward sameness.

If we do not brief place, AI will average it.

The second stage is inputs.

This is about collecting and structuring the data needed to feed the system: site information, planning controls, survey, client requirements, sketches, design intent, precedents, cost targets, material preferences, sustainability goals and risk constraints.

For some typologies, the form may be optimised against planning rules and site constraints to develop an efficient floor plate. For others, where the project is not purely about maximising yield, the architect’s authorship will be much more direct.

The third stage is generation to a decision point.

At first, the system may only run to sketch design, producing options that can be evaluated before next steps are taken. Further down the track, it may generate substantially documented options with cost estimates and risk checks attached.

The fourth stage is evaluation.

This may become one of the most important parts of practice. Rather than drawing one building slowly, architects may evaluate many possible buildings quickly. Which one best satisfies the brief? Which one best protects place? Which one is buildable? Which one carries unacceptable risk? Which one looks convincing but fails as architecture?

The architect’s role shifts from producing the drawing to deciding which building deserves to exist.

The fifth stage is authentication.

The AI may draft. The AI may coordinate. The AI may check. But the architect must authenticate. This is the final professional act of reviewing, accepting and standing behind the drawings.

The future architect may not draw every line, but they will need to stand behind every line.

The sixth stage is construction.

Construction does not disappear. Buildings still happen in the real world, with real builders, consultants, clients, budgets, substitutions, defects, delays and disputes. AI may assist with RFIs, contract administration, site reports and variation assessments, but construction remains physical, legal and human.

AI may compress documentation. It does not remove responsibility.

The fee reckoning

This matters because architecture has been in a perpetual state of crisis for decades.

Architects are not paid enough. Graduates are not paid enough. Project architects and associates are often not paid enough. Much has been written about why, but one reason is that architects do not charge enough for the risks they take.

When the tide goes out on documentation services, this has to change.

For every major technology leap, the profession has tended to hand the efficiency savings back to clients while also paying a vendor for the new technology, often through the nose.

CAD made drawing more efficient. BIM made coordination more powerful. Cloud software improved collaboration. Rendering tools improved presentation. Project management platforms improved communication.

Yet too often the result was not higher margins or better pay. Clients expected more. Programs compressed. Documentation expectations increased. Architects paid more for software. Staff worked under more pressure. Risk stayed with the architect.

AI cannot be allowed to repeat this pattern.

The most obvious thing that needs to change is hourly billing.

Hourly rates may not be the profession’s primary pricing model, but they are common enough to shape how many architects think about value. They need to be urgently phased down as a measure of worth. If the hours required to deliver a project drop by 50 per cent, it is not sustainable to simply hand that saving back to the client.

The risk has not halved. The responsibility has not halved. The professional indemnity exposure has not halved. The consequences of error have not halved.

If anything, faster production may require more careful checking, stronger judgement and clearer authentication.

Speed is not the same as low value.

If a practice has refined a method to deliver an amazing project in only a handful of days, the value of that project is not a handful of days of labour. It may represent years of experience, systems, judgement, templates, risk controls and professional knowledge compressed into a shorter period.

When AI compresses time, it should not compress value.

Clients do not buy hours. They buy confidence that the right building can be designed, documented, approved and built.

The nightmare scenario

There is another possibility the profession needs to confront.

If we can build the AI draftsperson, there is no technical reason the technology has to be sold only to architects. It can be put directly into consumers’ hands.

This matters most for small practice.

Alterations and additions are not marginal work. They are the economic base of many small architectural practices. They allow practices to survive, train staff, build relationships, develop local knowledge and gradually climb toward larger or more complex projects.

But we already know that, in many situations, builders can be involved in producing drawings for a building permit. In practice this can mean very modest documentation being assembled and passed through the system without the independence or professional care that architects would expect.

That status quo needs to be challenged before AI makes it much worse.

The profession should be saying to regulators that permit documents must be produced, reviewed or authenticated by appropriately registered and independent professionals, such as architects, registered draftspeople or registered building designers. Not simply by the builder who will benefit from the construction contract.

If we do not do this, AI could hollow out a large part of the alterations and additions market.

Just as a large portion of commercial art can now be generated by AI, a large portion of ordinary residential alteration work could become the target of consumer-facing AI.

Imagine a phone app released by a major hardware store. Let us call it Hammer Barn.

A homeowner enters their address, uploads a few photos, describes the extra bedroom, deck, studio or kitchen extension they want, selects a style and nominates a budget. The app produces a design, a simple drawing set and a list of materials. It may even help prepare documentation for approval.

The app is free.

Why? Because the drawings are not the product. The materials are the product.

Every specified item comes from Hammer Barn.

In a world where construction is eye-wateringly expensive, competing with free is not viable for small practices.

Yes, there will still be wealthy clients who want the real deal. Yes, there will be highly sought-after practices that can proudly offer a crafted, fully human, hand-led model of architectural service. But that may become a small luxury tier.

The ordinary base of architectural practice could be badly exposed.

The danger is not just AI slop. It is commercially optimised slop: drawings that are good enough to sell a project, but not good enough to protect the client, the street, the builder, the neighbour or the built environment.

Regulation must be about the public, not us

So yes, we need to ask for better regulation.

But we cannot ask for better regulation simply to save our businesses or our profession.

No profession has an automatic right to exist. We have to earn that right by being valuable. And to be valuable, we have to act in the public interest.

The argument cannot be “protect architects because architects deserve to survive.”

The argument has to be “protect the public because the built environment matters.”

AI-generated building documentation will not just affect architects. It will affect streets, neighbours, safety, amenity, heritage, climate performance, construction quality, consumer protection and the long-term character of our cities and suburbs.

The regulatory question is not whether AI should be used. It will be used.

The question is who is allowed to use AI to produce permit documents, who checks the work, who is independent and who carries responsibility.

Architects are uniquely placed to help answer those questions. Not because we are entitled to protection, but because we understand the public consequences of bad buildings.

We should be going to politicians and regulators with a simple warning.

Slop is coming.

Not because AI is inherently bad, but because cheap automated production, weak regulation and commercial incentives will combine very quickly. If the rules are not written before the tools reach the mass market, the default setting will be speed, volume and sales conversion. Not quality. Not place. Not care.

We are not asking to be protected from AI.

We are asking for the public to be protected from unaccountable AI.

The choice is not AI or architects. The choice is accountable AI or commercially optimised slop.

Line of sight

Architects are trained to see what others miss.

We see the consequence of a bad detail before the leak arrives. We see the damage caused by a poor street interface before the public realm fails. We see the difference between a plan that fits on paper and a place that will be generous in life.

That is why the current disengagement is so troubling.

The profession does not need certainty to act. We rarely have certainty. We work with incomplete information all the time. We assess risk, test options, make judgements and move forward.

AI should be treated the same way.

We do not know exactly where this ends. We do not know how fast every piece will arrive. We do not know which tools will dominate, which business models will win or which parts of practice will resist automation longest.

But we have line of sight.

We can see enough to know that documentation may be transformed. We can see enough to know that clients will bring AI into the briefing process. We can see enough to know that consumer-facing tools may enter the residential market. We can see enough to know that hourly billing is becoming dangerous. We can see enough to know that professional responsibility will matter more, not less.

The future architect may spend less time producing every line and more time briefing, directing, evaluating, authenticating and defending the building.

That is not necessarily a smaller role.

It may be a more important one.

But only if we start adapting now.