Featured image: A digital rendering of a speculative city created for Google DeepMind’s Visualising AI project. Credit: Photo by Linus Zoll / Google DeepMind / Unsplash, free under the Unsplash License.
A building used to move through a familiar sequence: someone sketched an idea, worked it into drawings, tested a model, and eventually handed a detailed plan to engineers, contractors, and the city. Computers have been part of that process for years, but generative design is changing how many possibilities a team can consider before choosing a direction.
Software can help compare daylight, wind, circulation, energy use, materials, and cost across different proposals. That's useful, especially when a site has competing demands, but an efficient building isn't automatically a place people want to live around. The harder question is whether the design is solving the right problems in the first place.
The algorithm can draw. The harder question is what it should value.
Generative tools can explore combinations that would take a design team far longer to draw and compare by hand. Given a site and a set of constraints, a model might identify ways to admit more daylight, reduce certain material needs, or arrange homes more efficiently. Those options can expand a designer's thinking, provided they're treated as possibilities rather than instructions.
Before producing hundreds of versions, somebody still has to decide what counts as a good result. Does the project prioritize lower construction costs, more homes, shade, access, privacy, or room for neighbors to gather? The software doesn't make those choices disappear; it carries them into every option it produces.
Efficiency is not the same as a good city
Some of the things people love about cities are difficult to put on a performance chart. A slightly awkward corner becomes a familiar meeting place, a small storefront interrupts a long block, and an older building gives a street some of its character. Those details don't always look efficient on paper, but people build their lives around them.
Optimization tends to reward what can be measured, which makes it tempting to design every project around the same few targets. If developers use similar briefs and software to solve similar financial problems, the results may be technically stronger while feeling increasingly alike. A better skyline shouldn't have to lose the irregularities that make a city recognizable.
AI will change the economics of iteration
Trying more design ideas usually costs more time and money. Each alternative needs people to draw, analyze, coordinate, and present it, which helps explain why teams often narrow their options early. Generative tools could reduce some of that effort and make it easier to explore different approaches before major decisions are locked in.
Smaller practices and public agencies may benefit from that flexibility, but cheaper options aren't the same as better decisions. If residents have no influence over the brief, they'll simply be shown more versions of someone else's priorities. Ten thousand variations won't help when the underlying question is wrong.
Public participation needs better tools, not more renderings
Communities are often invited to respond after the costly decisions have already been made. By then, a public meeting can feel like a choice between approving the proposed design or watching the process move ahead without much change. Using design tools earlier could make those conversations more useful.
Residents might compare what happens when a project provides more trees instead of parking, or larger homes instead of more units. They could see the trade-offs between a taller building with open space around it and a shorter one that covers more of the block. The value wouldn't be another convincing picture; it would be a clearer explanation of what different choices mean.
Architecture is not only form
AI in architecture is often illustrated with strange towers because unusual shapes make good headlines. The less visible work may be just as important. Design teams could use software to compare building performance, coordinate maintenance information, or understand how a space behaves after people begin using it.
Imagine a school tracking where classrooms overheat or an apartment building recognizing patterns that suggest a repair is needed. Those possibilities still depend on sound data, responsible oversight, and systems that actually work, but they shift attention from opening-day photographs toward the everyday life of a building. The skyline is only one part of the story.
The danger is invisible authorship
When a design tool recommends an option, its output can look objective simply because it arrives as a calculation. It isn't free of assumptions. The data used to develop the system, the constraints written into the brief, and the priorities set by its operators all shape what the software makes easy to see.
A model built around expensive commercial projects may suggest familiar commercial answers, while a brief that rewards cost savings above everything else will express that choice in the finished design. Architects and planners therefore need to understand not only how to use these tools, but how to question the patterns they reinforce.
Good cities require human disagreement
Cities hold competing needs, and no software can make them vanish. Residents want homes they can afford, builders need projects that can be financed, local businesses want customers, and governments have to account for infrastructure and environmental limits. Those interests overlap, but they won't always agree.
AI may help people compare the consequences of different decisions, which could make a difficult conversation more informed. It can't decide whose needs deserve priority, though, and it shouldn't be treated as an excuse to avoid that argument. Choosing how a city changes remains a public responsibility.
The architecture of tomorrow will reveal what we asked technology to care about
I don't see AI simply replacing architects and filling cities with machine-designed buildings. It's more likely to become another part of the design process, much as digital modeling has become ordinary over time. That makes the questions surrounding judgment and responsibility even more important.
When a team can generate hundreds of options, taste also means knowing which ones to reject. When software can optimize many variables, judgment means deciding which variables shouldn't dominate. We can use new tools to imagine more possibilities, but the choice still has to make sense to the people who will live, work, and walk beneath the finished buildings.
What to Take Away
The next time a proposal celebrates an AI-generated building, examine the criteria that shaped its options. Ask about affordability, public space, climate, maintenance and who was able to participate in the design conversation. Generative tools can help compare possibilities, but a city becomes better only when the people who inhabit it have a meaningful voice in what gets built.
