Every practice has now seen the same demonstration. A line of text, forty seconds, a photorealistic interior. It is genuinely impressive, and it solves almost nothing we are actually short of.
The bottleneck in a small studio is not producing images. It is that a drawing set, a bill of quantities, three procurement conversations, a client review and two site queries are all live in the same week, and every one of them depends on a decision made in one of the others.
The work that quietly eats the week
Sit with a project for a few days and you can watch where the hours go. Reconciling a material schedule against what the drawings now say after the third revision. Taking off quantities to sanity-check a vendor's number before it reaches the client. Reading a specification back looking for the clause that contradicts a different clause. Writing the same weekly site report in the same format for the eleventh week running. Digging through a folder for the version of a detail that was approved rather than the version that was drawn.
None of that is design. All of it has to be right. And all of it is the kind of careful, repetitive comparison that people are measurably worse at on a Friday evening than a Tuesday morning.
This is where the tools earn their place. Pulling an answer out of a long specification. Summarising a meeting into actions while everyone still remembers what was agreed. Drafting the first version of a report that a person then corrects. Checking one structured list against another. Unglamorous work, and genuinely useful.
Where I would not hand over
The trouble starts when the output looks finished.
A material can be suggested that reads perfectly well on paper and is wrong for the room. It might be right for a residence and hopeless in a branch office where two hundred people touch the same edge every day. Nothing in the description tells you that. Knowing it comes from having watched a similar finish fail.
A programme can look efficient and ignore procurement. Sequence the trades beautifully, then discover the specified veneer is a six week lead time and nobody asked. The bar chart was never wrong. It was just answering a different question from the one that mattered.
A drawing can be internally consistent and still miss a beam that is actually there, because the model it came from was built from a drawing rather than from a site.
What these have in common is not that the tool made an arithmetic mistake. It is that construction decisions carry context and responsibility. Somebody signs the specification. Somebody is answerable when the joint opens up in year two. That accountability cannot be delegated to something that has never stood in the room.
The thing worth being careful about
There is also a question that a practice has to answer before the productivity one. Client drawings, budgets and tender documents are not ours to feed into somebody else's service. Working out where that data goes, and being able to tell a client the answer, is slower and duller than trying the tools. It also comes first.
On renders
The generated image is the least useful application in a studio that builds, and I would go further and say it is often harmful.
A visualisation that was never derived from a drawing sets an expectation the drawing cannot meet. On a design-build project we are the ones who then have to build the thing, so that gap lands on us rather than on somebody else. Every image we put in front of a client should be traceable to something that can actually be constructed. That rule has cost us a few pretty pictures and saved several difficult conversations.
Where we are
We are not using these tools in our design or delivery process today, and the honest reason is the confidentiality question above. We have not answered it to our own satisfaction yet.
The applications I am watching are the dull ones. Schedule reconciliation, quantity checking, a second read of a specification before it goes out. If those work, they buy back hours that currently come out of design time.
The aim is not an automated practice. It is a practice where the repetitive work takes less of the week, so there is more of it left for coordination and judgement. AI should give a professional more time to think, not a reason to stop thinking.
- AI
- Practice
- Specification
