How will AI change the way architects learn, work and get paid?
Yesterday, I wrote something on LinkedIn that travelled much further than I expected. I said that AI was about to kill the “CAD monkey” years of becoming an architect, and that I wasn't sure architects should celebrate that too quickly.
I was thinking about my own time in the profession. I spent nearly ten years in architecture, and like most people of my generation, some of those earlier years involved a lot of work I would have happily handed to AI if it had existed. Endless Photoshop changes before presentations, dimensions in Revit that needed updating after somebody changed the model, room schedules that somehow became sixty pages long, reports, minutes, drawing packages and all the other things that had to be finished before you went home.
There is already technology capable of removing a significant amount of that work, and more is coming. On the face of it, that should be brilliant for architects. Less time producing information and more time designing, thinking, developing ideas, speaking to clients and doing the parts of the job people actually entered the profession to do.
But there was something about that argument that bothered me.
When I think back to those years now, I realise that some of the work I considered repetitive wasn't simply production. Without really noticing it, I was learning how buildings worked. I understood drawings because I had spent endless hours inside them. I learnt to spot when something didn't make sense because I had probably already made that mistake myself. I began understanding how information connected across a project because I had repeatedly been responsible for producing and coordinating it.
Hidden inside some of the work we are now understandably excited to automate was part of the apprenticeship.
The discussion underneath that post became much more interesting than the original post itself. Architects from different generations began talking about drawing, repetition, technology, fees, training, construction and what practice might eventually look like if significantly fewer people are required to produce the same amount of work.
It changed the question for me. This is not simply about whether AI will make architects more productive. It is about what happens to the profession when producing architectural information becomes easier, and where the value of an architect moves as a result.
Production was doing two jobs
One of the most useful observations in the discussion came from Simon Maughan, who described apprenticeship as “learning what while figuring out why.”
I think that captures the issue particularly well.
When a graduate spends hours checking a room schedule against a model, the obvious output is a checked room schedule. On a timesheet it may look like an inefficient use of someone's time, particularly if software could complete the same task in minutes.
But something else is happening during those hours. The person checking sixty pages is repeatedly seeing how rooms, information, dimensions and requirements connect. They begin noticing inconsistencies. Patterns become familiar. Eventually something looks wrong before they can necessarily explain why.
That accumulated recognition is part of professional judgement.
Drawing has historically done something similar. David Armour raised the importance of drawing as a learning mechanism, not simply a means of producing an output. When you have repeatedly drawn junctions, details, plans and sections, you are forced to think about how things actually connect.
Technology has already changed that relationship considerably. Ian Layzell made the point that computers created a degree of separation between architects and the physical and material nature of buildings long before AI arrived. AI could potentially introduce another layer of separation if architects increasingly make decisions about information they have played very little role in producing.
None of this is an argument for preserving inefficient work simply because previous generations had to suffer through it. I have no nostalgia for spending half a night changing the same thing across a presentation because somebody changed their mind at 8pm.
The more useful question is what somebody was learning while doing that work, and whether we have another way of developing the same understanding once the work disappears.
Perhaps we need to separate producing from checking
Matt Perdeaux made a distinction in the discussion that developed my thinking considerably. Instead of treating production as one activity, he suggested separating producing from checking.
That could become important.
AI may increasingly produce the first version of a schedule, drawing, report, image or piece of coordination information. That does not mean a junior architect should simply accept it and move on to the next task. They could instead become responsible for interrogating it, checking it against the model or brief, understanding why decisions have been made and being able to explain the output before it goes anywhere near a client or construction team.
In some ways, that could create a better apprenticeship than the one many of us experienced. There was plenty of repetitive work in architecture that taught very little. If technology removes the genuinely administrative part while practices deliberately preserve the parts that develop understanding, younger architects could potentially learn faster rather than less.
But that will not happen automatically.
If the commercial response to AI is simply to remove the hours and reduce the number of junior people involved, then the learning disappears with the task.
Every time a practice automates something, I think there is therefore another question worth asking alongside the productivity calculation: what was somebody learning by doing this, and where will that learning happen now?
The next generation may understand the technology before they understand the profession
Another part of the discussion came from Dr Giuseppe Parita, who pointed out that younger people entering architecture are already digital natives. Working with AI may become completely natural to them, perhaps much more quickly than it does for people who have spent twenty or thirty years practising in a different way.
I suspect he is right.
A graduate joining a studio in five years may be considerably more fluent with some of these tools than the director reviewing their work.
But knowing how to use AI and knowing when AI is wrong are very different skills.
That creates an unusual inversion. The youngest person in the room may be the most capable operator of the technology while simultaneously being the person with the least professional experience to challenge what it produces.
This matters because AI does not necessarily look uncertain when it is wrong. A plausible drawing, answer, report or recommendation can appear extremely convincing. The ability to interrogate it requires knowledge that cannot simply be generated by the same system.
The question is therefore not whether the next generation will learn how to use AI. I have very little doubt that they will. It is how they acquire enough experience of architecture, construction, clients, materials and consequences to know when they should disagree with it.
What happens to the bottom of the practice?
Yann Leroy took the discussion in another direction by thinking about the organisational structure of architecture practices.
He described the early jobs that formed part of his own route into the profession: print rooms, tracing paper, schedules, reports and AutoCAD. Those tasks were not glamorous, but they were entry points. They got younger people inside practices, where they could watch experienced architects work, make mistakes, gradually take on responsibility and become useful in increasingly sophisticated ways.
If much of that production layer disappears, the traditional pyramid of an architecture practice could become considerably narrower.
A studio that previously required ten junior people to produce a certain amount of work may eventually need far fewer. That could be commercially attractive to the practice, but it raises an obvious question about where the next generation enters the profession.
There is a darker version of that future in which established practices become increasingly concentrated around rainmakers, experienced technical people and a much smaller number of junior positions, while architecture schools continue producing graduates for roles that no longer exist in anything like the same numbers.
I think that risk is real.
But there is another possibility that appears contradictory at first.
AI could remove junior jobs while making it easier for younger architects to start studios
If technology allows a very small number of people to produce work that previously required a much larger team, the economics of starting an architecture practice change as well.
Pedro Suescun pointed out during the discussion that this process has already begun. Technology has enabled very small studios to compete for and deliver work that would once have required considerably more production resource.
AI could accelerate that.
A capable young architect with the right experience, judgement and relationships may eventually be able to establish a lean studio much earlier because they no longer need a large team simply to create the volume of information required to operate.
That does not solve the employment problem. Demand for buildings is not infinite, and greater productivity does not magically create enough work for everybody. Pedro rightly challenged that assumption. If ten architects can eventually produce what previously required one hundred, society does not suddenly need ten times as many buildings.
The outcome could therefore be uncomfortable and uneven. There may be fewer conventional positions inside established practices while more small practices compete for the available work. Large studios may become leaner. Some services that were previously uneconomic may become viable. The barrier to creating a practice may fall at exactly the same time as competition between practices increases.
We do not yet know where that settles.
What seems increasingly likely to me is that AI will affect more than the tasks architects perform. It could alter the organisational model through which architectural services are delivered.
This is not simply another CAD transition
Architecture has been through technological transitions before, so CAD is an obvious comparison.
There is something useful in that comparison. CAD, BIM, rendering software and increasingly sophisticated digital workflows all changed the amount of work practices could produce and the people required to produce it. Architects did not disappear.
But Pedro made an important point about why AI may be different.
CAD adoption required significant infrastructure, hardware, software and training. The transition took time, and traditional methods could remain commercially competitive for years.
Much of the infrastructure required for AI already exists. The computer is already on the desk. The phone is already in the pocket. Many of the tools can be accessed immediately and cheaply enough for experimentation to begin without a major investment decision by the practice.
AI is also moving beyond drawing production. It can increasingly participate in research, writing, analysis, visualisation, coordination and other activities that would previously have been considered cognitive work rather than simply drafting.
That does not mean every prediction being made about AI will happen. Alan Holstein quite reasonably challenged another assumption underlying much of this discussion: that access to increasingly capable AI will remain as inexpensive as it currently appears. The infrastructure required to provide these systems is enormously expensive, and today's pricing does not necessarily tell us what the economics will look like indefinitely.
The direction may be relatively clear while the speed, scale and eventual commercial model remain much less certain.
Architecture still has to survive contact with a building site
Michael Gildea brought the conversation back to something that can get lost very quickly when discussing digital technology.
Architecture eventually becomes physical.
Buildings have to be procured, coordinated, constructed, occupied and maintained. Materials behave in particular ways. Contractors need usable information. Regulations have consequences. Details that look convincing on a screen still need to work when somebody tries to build them.
As producing plausible information becomes easier, understanding whether that information is actually correct may become more important.
The same applies to responsibility. An AI system can produce an answer, but somebody still has to decide whether that answer should become part of a building.
Marina Kozul extended this into questions of ethics and professional principles. As technology becomes capable of generating more options and recommendations, human responsibility does not disappear. Architects still have to decide what should be automated, what compromises are acceptable, what constitutes good architecture and who is accountable when something goes wrong.
Technology can increase the number of possible answers. It cannot remove the need to choose between them.
When generation becomes cheap, selection becomes valuable
This may be one of the more fundamental shifts.
There is a tendency to demonstrate AI capability through volume. Generate twenty options in seconds. Produce hundreds of façade studies. Explore a thousand configurations.
That sounds extraordinary compared with a process in which each alternative required considerable human time.
But one thousand options are only more useful than ten if somebody can understand which one is worth pursuing.
The scarcity moves.
For much of architectural history, producing alternatives was expensive because each required labour. If generation becomes extremely cheap, the valuable capability increasingly becomes knowing what to ask for, recognising what matters, understanding the consequences and selecting the right direction.
Experience may therefore become more valuable rather than less.
The professional question gradually shifts from “Can you produce this?” towards “Can you understand, interrogate and choose the right thing?”
And that leads directly into the commercial problem I think architecture practices need to start thinking about now.
If 40 hours becomes 10, who gets the other 30?
James Ogston raised what I think was one of the most important questions in the entire discussion.
Imagine a piece of work that currently takes an architect forty hours. AI and automation eventually allow the same person to produce an equivalent or better outcome in ten.
The optimistic version is that the architect has gained thirty hours. They can spend more time thinking, designing, exploring alternatives, speaking to the client or simply doing better work.
But that assumes the client continues paying for forty hours.
Once everybody knows the work can be completed in ten, ten hours can become the expectation. Then another practice offers to do it in eight. Programmes shorten, competition adjusts and fees begin following the new production cost.
The thirty hours did not become creative freedom. They simply disappeared from the programme and eventually from the fee.
Architecture already has a difficult relationship with how it prices expertise. If an architect with thirty years of accumulated experience can use technology to reach the correct answer in ten hours rather than forty, it seems strange to conclude that their expertise has suddenly become worth 75 per cent less.
That makes AI a business-model question as much as a technology question.
Are clients paying architects for the hours required to produce information, or for the judgement required to arrive at the right answer?
If practices continue tying their value too closely to production time, the productivity gains created by AI could flow disproportionately towards clients rather than architects.
Everyone becomes faster. Deadlines become shorter. Clients expect more. Fees become tighter.
The profession runs faster while standing in roughly the same place.
That is not inevitable, but I think it is a possibility worth taking seriously.
Perhaps AI forces architecture to become clearer about where its value really sits
I began the original post wondering what would happen when the “CAD monkey” years disappeared.
The discussion underneath it left me thinking about something much broader.
Some junior roles may disappear. Training may need to become much more deliberate. Experienced architects could become dramatically more productive. Small practices may become capable of work that once required much larger teams. Competition could increase. Traditional fee structures could come under pressure. Construction knowledge and professional judgement may become more valuable precisely because producing convincing information becomes easier.
Several of those things can be true at the same time.
I don't think there is much value in pretending we know exactly where this ends. Architecture has always changed alongside the technology used to produce it, but AI is moving quickly enough that practices may need to make decisions before the eventual shape of the market is obvious.
What I do think is becoming clearer is where the interesting question lies.
If drawings, schedules, reports, images, research, options and information become dramatically easier to produce, what are clients ultimately appointing an architect for?
I suspect the answer moves further towards judgement, responsibility, interpretation, understanding construction, understanding people, choosing between possibilities and knowing when something is wrong.
If that is where the value moves, the profession will need to think carefully about how it teaches those capabilities to the next generation.
Architecture practices will also need to think about how they organise themselves around them.
And perhaps most importantly, they will need to become considerably better at explaining and charging for them.
What happened after I wrote about this
The original LinkedIn post that led to this article reached more than 18,000 impressions and generated over 70 comments, with the conversation travelling far beyond the people who already followed me. More interestingly, it brought new people into my network, onto the Grand House website and, in the days that followed, three people booked conversations into my calendar.
It became a useful real-world example of something else I've been thinking about: digital visibility. I hadn't written the post to promote Grand House or generate leads. I wrote about something I knew from nearly ten years inside architecture and had a genuine point of view on. That thinking travelled, people discovered the person and business behind it, and some chose to continue the conversation.
I've written more about how architecture and design studios can build that kind of visibility intentionally, without turning their founders into content machines or trying to game search and AI:
→ How can architecture and design studios become more visible in search and AI [ https://www.grandhouseldn.com/gh-knowledge-hub/architecture-design-studios-digital-visibility-search-ai ]
→ Also, link to be original LinkedIN post can be found here: [ Also link to the my original LinkedIn post can be found here:
https://www.linkedin.com/feed/update/urn:li:activity:7505146351588995073/ ]
By Richard Hall, Founder & CEO of Grand House