Skip to content 
A speculative design study, and a record of how we’re learning to use AI as a research instrument.
Research is half of our name. Long before this technology arrived, Plexus Research and Design was structured around a conviction: that the work gets better when the process itself is treated as a subject worth studying, not just a pipeline to run. AI is the most interesting thing to happen to that conviction in twenty years.
The SSP Museum is not a commission. It has no client, no site, and no budget. That is deliberate. It’s a research vehicle: a speculative museum that we pushed through every stage of our actual design process to learn what this technology is genuinely for. The answer surprised us, which turned out to be the answer.
The Genesis: We Asked for One Building and Got Another
The study began with a paragraph. The prompt, reproduced above exactly, asked for an ultra-detailed HDR photograph: a massive architectural landscape carved as if from one piece of material, ground rising into curving columns and flowing glass walls, every surface merging into the next.
That is not what came back.
What came back was a drawing: ink line and flat color, pale yellow facets against a matte blue sky, closer to a mid-century competition board than to the seamless carved monolith the words described. The moodboard’s lineage of retro-futurist collage had pulled the output somewhere the prompt never pointed. Faceted where we asked for flowing. Graphic where we asked for photographic. A drawing where we asked for a photograph.
This is the misunderstood truth about image generators: the slot-machine quality isn’t the flaw, it’s the value. These are not rendering tools. A rendering tool that ignored the brief this thoroughly would be broken. They are ideation machines. You get something like what you asked for, only different, and the difference is where the ideas live. No one on our team would have drawn this building. Now it existed, and it was too interesting to ignore.
So we did what architects do with an interesting drawing: we took it seriously.
Why We Invite the Surprise
Every practice develops a house style, and a house style is partly a set of solutions you’ve stopped questioning. The honest term is habit.
What AI offers, used deliberately, is heterogeneity. It proposes forms in languages we don’t default to, and it holds our own habits up where we can see them. The discipline isn’t in controlling the machine until it obeys; it’s in recognizing which surprises deserve development and which are just noise. That judgment is editorial, architectural, and human. It is the actual skill, and it’s the same one we’ve always practiced. The machine widened the search. It did not choose.
Read more: Why Is Architecture Important? Exploring Its Impact on Modern Society
From Accident to Architecture
A form that exists only in one image is a poster, so the drawing went dimensional: study models examined on a table, from angles the original image never showed, the way we’d interrogate any scheme in the studio. The composition held. The cantilevered prow, the folded tower, the plaza that climbs into the building. It wasn’t a lucky image. It was a parti.
Then the material question. The same massing ran as dusk-lit concrete, as white monolith, and as perforated steel: three arguments about what the building wants to be. The perforated skin won. It gave the form a second reading at close range that the concrete never earned, solid from the plaza and porous at arm’s length.
Stress-Testing the Fiction
We took it further than concept studies usually go: a speculative construction sequence showing frame, half-clad, and complete. Nothing here is being built; the label says so. But asking how this would actually assemble is how architects think, and forcing the fiction to answer construction’s questions is what separates a design study from an image collection. The scheme got more disciplined every time it had to explain itself.
By the detail pass, the perforation logic had become the project: dot-screens sized by orientation, apertures that read as texture from distance and as windows up close. The drawing’s graphic language, translated into a buildable skin.
Read more: What is Modern Vernacular Architecture?
What This Means for the Real Work
The SSP Museum will stay speculative. The process will not. The same loop now runs inside our commissioned work: reference, prompt, take the surprise seriously, model it, discipline it. It lets clients see more of their building, earlier, across a wider range of ideas than a traditional concept phase can afford. The judgment stays human. The search gets wider.
That, for us, is what AI in architecture is actually for: not faster pictures, but unexpected ones, and the practice of knowing which to keep.
Plexus Research and Design is an architecture practice in Atlanta, Georgia. The studio’s AI design research is documented at @ai.plexus; the built work is in the portfolio. If you’re considering a project, we’d like to hear about it.
