Serving as Judges for an AI×Architecture Competition—and Exploring Our Own Entry

Introduction

I recently had the honor of serving as a judge for the U25 AI×3D PERSPECTIVE AWARD 2026 (organized by the Real Estate Research Institute), an architecture competition built around the use of AI.

It all started when I received a direct message on Messenger one day from Dai Murayama, a student at Keio University. Murayama wants to start his own company eventually, and as training for that, he wanted to organize an architecture competition. He was taking on a completely new challenge with no existing foothold, and since I too had once been at Keio, I accepted, hoping to help in whatever way I could.

Competition results can have a real impact on participants' careers, so the role demands a sense of responsibility. Before the competition began, I did some preparation, working out from the theme what criteria the entries should be judged against (see the "Premises" section below).

Premises

Let me first share the perspective from which I read the proposals.

Theme

The competition website set out the brief as follows.

In an age when AI is transforming work, what is an office for? Use AI and 3D perspective renderings to depict a space no one has ever seen. It falls to the next generation to redefine what the office should be—and how its value is communicated. Show us the office you would want to work in.

THEME: "The Office" — Beyond convention, a place you want to work. Envisioning the next-generation workplace with AI.

Evaluation Criteria

Given the theme, I felt the following perspectives were the important ones.

• The proposal should point the way for the world and the industry: I see a competition less as a contest of which solution is best today and more as a place to set out a vision that guides the industry. I wanted proposals that look years or decades ahead. Historically, competition and exhibition entries have had an enormous impact on the industry even when they were never built. So rather than solid schemes that could be realized right away, I gave high marks to ideas that take the kind of risk only a competition allows, yet would still hold up in the future (5 to 10 years out, say)—what I call recklessness and boldness.

• The proposal should anticipate the future of the office: As AI rapidly streamlines and automates work, the value we ask of an office changes. Routine tasks shrink, people focus more on creative work, and the relative value of physical contact and chance encounters rises. I hoped to see visions of space that anticipate that future.

• The relationship between AI and people should be appropriate: Rather than portraying AI as a mere tool, I looked at how finely each entry rendered the relationship—AI running alongside people like a digital twin, or AI agents taking over human tasks. I favored proposals grounded in the premise that people's role will shift from routine work toward judgment.

Judging Criteria: General Criteria and My Own

Based on the above, I scored each entry on a five-point scale across the following six criteria. Group A covers the criteria typical of any competition; Group B reflects my personal perspective.

A: General competition criteria
A1: Response to the theme
A2: Use of AI within the proposal
A3: Expression and presentation

B: My own criteria
B1: Guidance for the world and the industry (time horizon)
B2: Recklessness and boldness (core idea and novelty)
B3: Appropriateness of the relationship between AI and people

The Entries and My Assessment

My comments follow, in order of presentation.

1: Working Equilibrium (The Persona-Prediction Office)

Overview: "What makes a workplace comfortable differs from person to person, yet office space remains uniform." The proposal starts from the problem that designers can only imagine users in broad strokes, so a gap opens up between the space and actual needs. It brings AI-driven persona prediction into the design process, generating an initial zoning plan from a diverse cast of user profiles. Even after occupancy, the space keeps flexibly changing in response to who gathers there and how the movable furniture is arranged. From static to active: a community-rooted office that changes every day, driven by prediction.

What worked: The idea of a dynamic office that becomes progressively optimized for its individual users (similar to MIT's Ori Living). Since every kind of service tends to evolve from one-size-fits-all toward individual optimization (adaptive education, for instance), this approach feels well judged.

What could be improved: Is movable furniture alone enough to make an office space truly dynamic? It would be better if the other elements that make up the space were dynamic too. A purely bottom-up approach like this won't necessarily lead to a state that is right for the company as an organization, so a top-down perspective on how employees relate to one another would also be welcome. And with people's thinking and organizational structures themselves constantly shifting, the presentation didn't make clear how that information would be fed back into the office space to keep improving it.

2: Be a Next-gen Nomad

Overview: "Organizations and people alike become freer." The proposal predicts that in the AI era, compact organizations with a limited number of human employees will become the norm, and it assumes a "hyper-nomadic" way of working in which people move from office to office on contracts lasting a few months. Using AI to analyze past offices, it identifies "open × emotive" as an unoccupied territory in design. The result is a resort-like space with no walls, a floor that rises and falls like terrain, and forest, river, rock and water engaging all five senses. AI is positioned as "the starting point for concept development, free of human bias," in an attempt to create a space no one has seen before.

What worked: The proposal is grounded in a prediction that organizations will become smaller. Its realistic workflow makes it highly feasible.

What could be improved: Co-working spaces for nomads already exist, so it was hard to see where the novelty lies. The design workflow itself is unchanged—people using AI at each step, from ideation to research to rendering—with human checks still required at every stage, so the designer's way of working hasn't fundamentally changed.

3: Stranding

Overview: "Creativity does not come from a tidy environment; it lives in the moment when disparate things unexpectedly meet and collide." This is an experiment in implementing, as literal architectural space, the "coexistence of the heterogeneous" that generative AI achieves in its latent space. Workers drift through the space, physically encountering overlapping layers of real materials, different time scales, sizes and environmental conditions. The office is no longer an apparatus for production but "an ecosystem that provokes creativity." Under the keywords coexistence, serendipity and landscape, the entry bills itself as "an office of discovery, where you find the seeds of ideas."

What worked: The emphasis on creativity sparked by chance encounters captures what people's role will be in the future. Although the presenter said it hadn't fully succeeded, the attempt to connect AI with design software via MCP and shortcut the design process struck me as the right approach to using AI going forward. The balance between the AI-driven process and the design of the output was good.

What could be improved: I wanted a bit more explanation of how working here would make designers more creative than a conventional office does. I understand that AI cuts down the design process, but the entry would have been even stronger with a logic for how that simultaneously stimulates designers' creativity—how does saving time translate into greater creativity?

4: Henkansha, "The Crust of Knowledge"

Overview: An office for Henkansha, a next-generation publisher whose philosophy rests on "editing" and "circulation." A book never springs from one person's idea alone: a seed planted in conversation grows in someone's mind, is shared again, and is edited into a single volume. The proposal expresses that circulation and accumulation of knowledge in architecture. A meeting space enclosed by a central bookshelf forms the core of knowledge, and the books piling up there are called "strata of knowledge." The lineage of materials—wood to paper to book—is translated into space, under the credo that "knowledge is not private property but a shared resource nurtured through dialogue and editing."

What worked: Its focus on human creativity, and on space designed to foster it, captures people's future role. It thinks about space that promotes not just individual work but team communication and creativity. And by creating a space where people handle paper, it generates value that only humans can touch.

What could be improved: The use of AI was not especially novel. A clearer separation between ideas specific to a publisher and broadly applicable, cross-industry ideas would have made the proposal even more instructive for the industry.

5: Vibe Office Design (The Well-Side Office)

Overview: "From an age of 'writing' code to an age of bringing it to life in dialogue with AI—design stands on the same horizon." The entry champions "vibe" design, designing by feel. At the center of the floor sits neither the CEO's office nor a large conference room but a circular water-and-coffee counter: the "village well." Within a 3-meter radius there are no desks, just an open void for standing conversation that invites a word in passing. "Small talk is not the margin of work; incidental conversation is the very source of value." Here, the proposal declares, chatting is the work.

What worked: The concept modeled on vibe coding is strong and captures where the design process is heading. Of all the entries, this one designed—and actually carried out—the design process that most appropriately reflects the relationship between people and AI.

What could be improved: It was hard to see where the forward-looking vision for what an office should be lay. Beyond aggregating opinions bottom-up and translating them into space, reflecting a top-down vision for the office would likely produce a better space (collecting kindergarteners' opinions bottom-up, for example, doesn't guarantee a good space). The process was novel, but the resulting space didn't strike me as beautiful or new.

6: Ducting Office

Overview: An "open R&D office" for a comfort-tech company developing localized air conditioning, humidity control and air-quality sensing. Ducts, pipes and sensors normally hidden above the ceiling are exposed and brought down to floor and eye level, turning HVAC from "invisible equipment" into "environmental infrastructure you can experience." Instead of comfort delivered uniformly, the space embodies "comfort that people feel, choose and update." Office, showroom and living lab overlap, turning the workplace itself into a testing ground for environmental control.

What worked: Its focus on space for creative roles such as R&D points to the growing importance of R&D work. It creates opportunities for people to engage with the real thing—experiences you can only have by coming to the office.

What could be improved: It wasn't clear what impact using AI had on either the occupants or the designers (the content felt like it could have been proposed without AI). I also wanted a clearer expression or explanation of how this space would positively affect people in creative roles.

7: An Office for Exchanging "Thought" and "Sensation"

Proposal overview: This entry starts from the premise that, in an age when AI handles information processing, the value of the office shifts away from "sharing information" and toward two things that are hard to achieve without being in the same room: "exchanging thinking" and "exchanging sensation," the latter something only humans can do. The setting is an office beneath an elevated railway for product-based companies (apparel, curtains, cushions). A workshop sits at its heart, and the boundary between inside and outside is softened so that the activity within spills out into the streetscape. In short, once information processing has been handed off to AI, the proposal tries to give spatial form to "the value that only face-to-face encounters can create."

Strengths: The concept of letting people "physically engage" with the space zeroes in on something only humans can do, and I found it compelling. The team had thought through, and clearly articulated, how AI and people would relate to each other. They also deployed a wide range of presentation formats, including physical models and video, to communicate their message clearly and in detail.

Room for improvement: The way AI was used felt conventional; it stayed within the realm of things such as environmental simulation that existing simulation software can already do. The underlying view of AI struck me as somewhat dated, and I did not see much novelty in the design workflow. Because the proposal was framed around the limits of what humans can do, it could have benefited from freer, bolder thinking.

Results

My evaluation

As noted above, I scored the entries against six criteria. My results were as follows.

1st: Stranding (23 points)
A1 Response to the theme: 4 / A2 Use of AI within the proposal: 4 / A3 Expression and presentation: 4
B1 Direction-setting for the world and the industry (time horizon): 4 / B2 Irresponsibility and boldness: 3 / B3 Appropriateness of the AI–human relationship: 4

2nd: Henkansha, "Crust of Knowledge" (20 points)
A1 Response to the theme: 4 / A2 Use of AI within the proposal: 3 / A3 Expression and presentation: 4
B1 Direction-setting for the world and the industry (time horizon): 3 / B2 Irresponsibility and boldness: 3 / B3 Appropriateness of the AI–human relationship: 3

3rd: Vibe Office (Idobata) (20 points) *Tied with the second-place entry on points, but since this is an architecture competition, I ranked Henkansha's "Crust of Knowledge" second for its superior aesthetic quality.
A1 Response to the theme: 2 / A2 Use of AI within the proposal: 4 / A3 Expression and presentation: 5
B1 Direction-setting for the world and the industry (time horizon): 4 / B2 Irresponsibility and boldness: 3 / B3 Appropriateness of the AI–human relationship: 4

Final evaluation by the full jury

1st: An office for exchanging "thinking" and "sensation"
2nd: Henkansha, "Crust of Knowledge"
3rd: Stranding

The other jurors all have long careers in practice and have been working at the front line of architecture. I, on the other hand, have been absorbed in AI development and have drifted away from day-to-day design work, so our perspectives naturally differed in many respects. Each juror gave thoughtful, engaged critiques, and I found it valuable to hear such a range of views.

My own proposals

Handing out critiques like the ones above without offering anything of my own would be rather presumptuous, so I would like to set down the proposals I came up with for the same theme.

Proposal 1: Returning the Surface

Concept: Give the ground surface back to nature and bury the AI infrastructure and data underground. People descend into a bedrock chamber below ground to make decisions, while the surface above becomes a forest. In contrast to the tense, tightly secured space underground, the surface is a space that maximizes human creativity and communication within teams.

Background: As AI advances, demand for data center capacity grows, and where to put those facilities becomes a major issue. Build them above ground and land runs short; push them out to the suburbs and power delivery becomes inefficient. This has led to ideas for distributing data centers and moving them underground in urban areas, an approach that turns them into an energy-saving device for the city as a whole. Servers continuously give off low-temperature waste heat at 30–45°C, which can be captured for district heating, while the cool temperatures underground assist with natural cooling.

Human work, meanwhile, is likely to shift toward high-stakes, highly confidential decision-making. The underground space becomes a place cut off from all communication, where decisions that cannot be exposed to any network are made. By deciding right beside the AI infrastructure, humans can physically inspect the actual machines and retain final authority (to halt or isolate them), making this the last line of human defense against AI that is becoming autonomous and cannot be assumed immune to attack. People recover in the forest above and descend into the quiet underground only when a truly critical decision must be made. Deep, uninterrupted concentration and the security and sovereignty of being disconnected from the network: this is the kind of place today's knowledge workers need.

Proposal 2: The Building as Organism

Concept: The office itself becomes an "environmental prompt" for AI. Employees' behavior and their dialogue with the office serve as continuous input, and the office space is generatively redrawn to suit the situation. There are no fixed plans; the AI reads how the space is used each day and regenerates it to fit the optimal conditions. In effect, the building undergoes its own metabolism, renewing itself without human hands, and becomes something like a living organism.

Background: Dialogue with AI will not stop at chat; it is expected to become linked to physical space (as in Physical AI). The office itself, through microphones and speakers, will be able to converse with its occupants (an AI office concierge). Through a variety of sensors, it will pick up on people's movements and needs, and continuously think about and update itself into the optimal space (via adjustable louvers, interior partitions, HVAC, lighting, and so on). When necessary, the building's own AI agent will reach out to a contractor and decide and carry out equipment repairs, extensions, and renovations on its own.

Those are the two proposals I sketched out. With only about 30 minutes to think them through, though, I suspect that when I look back at them later I will find plenty of loose ends.

Closing thoughts

The competition was a welcome chance to see how young designers are actually putting AI to use. It was also valuable to hear how the other jurors saw things and to take in such a range of opinions.

I look forward to continuing to think, together with all of you, about how AI will change architecture and cities.

My sincere thanks to Mr. Murayama, the Real Estate Research Institute, and everyone else involved for giving me this opportunity.