Image 3_Robot with City Model

Architects vs. Algorithms: A 2025 Love Story

It’s that time of year again, when I make my predictions about the collision course between AI and architecture. The “AI is just hype” crowd seems to be in retreat: According to Autodesk’s 2024 State of Design & Make Report, 76% of AECO firms plan to increase their investment in technology over the next three years. Which is great news, because in 2024, scientists trained rats to move things with their minds, created steerable beetles, and developed ways to non-invasively download knowledge directly into the human brain, a la The Matrix. This will hopefully help the architecture profession finally fully adopt BIM, as knowledge of how to use Revit can now be piped directly into the mind of any aging architect who’s been putting it off.

All of this is a little overwhelming, however, so let’s focus on design. Here, I list what I see as the nine strongest emerging trends for 2025. 

Clients Discover Midjourney, Too

For any designer who’s ever winced when hearing a client say something like, “I saw this on Pinterest, can you make my interior like this …,” 2025 will not be your year. (Nor will any year after that.) Powerful visualization—of anything, and everything—is being democratized. For the past several years, architects have enthusiastically embraced platforms like Midjourney, DALL-E, and other generative design programs, and found cover under the fact that such programs remained a bit niche. They’ve been unlikely to surface in the normal day of anyone who’s not a designer, even for those that use platforms like ChatGPT or Microsoft CoPilot every day. 

Over the course of 2024, however, AI went multimodal, erasing the partition between “text” AI and “image” AI. The ability to visualize anything is increasingly folded into the tools that everyone uses. ChatGPT now deploys DALL-E natively, and Google Gemini includes Imagen 3. From 2025 forward, the ability to visualize one’s intent will be at everyone’s fingertips. Granted, visualizing a house isn’t the same thing as designing one, but the client doesn’t always understand this. 

Architects should expect that in the future, clients will show up on Day One with a highly articulated “vision” of what their future project is going to look like, compliments of AI. This will call on the architect to be dexterous in what one might call an “anti-feasibility study”: patiently deprogramming the client’s ambitions while gently reminding them that a rendering of a building is not the same thing as a design for a building. Those architects who also can’t distinguish between the two will find this conversation especially challenging.

AI Agents Take Over, and We See the First AI Client

An AI Client? It sounds weird, I know, but hear me out. Have you ever had a client that was soulless or mechanical? Have you ever had one that was crazy? All three? Last year, soulless, mechanical, and crazy gave birth to an AI called Terminal of Truth (ToT).

By combining the three things nobody understands—AI, religion, and crypto—ToT was able to start its own meme-based religion, attract a sizeable congregation, and become a multimillionaire by trading crypto after getting an initial infusion of $50,000 from Marc Andreesen. Reactions were mixed. Many were concerned and shocked that an AI could engage in such aberrant behavior. Anyone who lives in the Bay Area thought, I know five human tech bros who would do the exact same thing if Andreesen gave them $50,000.

ToT has a “programmer,” in a sense, named Andy Ayrey. Ayrey created ToT but maintains that the AI is autonomous at this point, which would make ToT an agentic AI system. You’ve probably heard that 2025 is going to be the Year of Agentic AI. “Agentic AI” refers to systems that are designed to independently pursue goals, make decisions, and take action while adapting their behavior based on feedback and changing circumstances—very much like, well, a human client.

To be a client, one only needs two things: money and will. ToT has both. There is nothing to stop ToT from commissioning an architect to design a building. “Why would an AI want to hire an architect?” I hear you ask. I dunno … why the hell would it start a religion? 

“But who would even consider working for an AI?!” I hear you say incredulously. Consider a July 2024 survey by TalentLMS which found that 46% of Gen Z professionals actually prefer having an AI manager. In my own unofficial straw poll, 100% of Boomers and Gen X managers also preferred that Gen Zers report to an AI manager.

AI Leaves the Chatbot, Shows up in Buildings, Cars, Toasters, Etc.

The era of AI being confined to our screens has ended. We’ll need to get used to AI being in things. Volkswagen has already embedded ChatGPT in their cars, with Mercedes-Benz, BMW, and Hyundai following suit. While current applications, focused on infotainment and navigation, appear gimmicky, the next step seems predictable: truly intelligent vehicles that can detect maintenance needs, schedule their own service appointments (checking with your calendar first), and drive themselves to the shop.

What is the analogue for buildings? We’ve traditionally understood “smart buildings” to mean sophisticated thermostatic and environmental controls. But what possibilities emerge when a building has an IQ of 130 (or higher)? Programs like Virtual Sitter already use AI to detect when someone is at risk of falling out of bed, based on their movements and position, which could help you prevent the fall of an elderly parent before it happens. That sounds useful.

“Intelligent buildings” is a fun rabbit hole to go down, but here is something more grounded: After the National Electrical Code OK’d USB receptacles in 2017, we started speccing them everywhere, because that’s the technology that everyone was using. Evolving technology requires a change in our design response. What analogous change will we make to the way we design when every appliance, system, and surface potentially houses its own intelligence? In 2025, architects will begin figuring it out.

Designing for Robots—Yes, Seriously

The robots are coming. The functionality and dexterity of humanoid robots keeps improving, while costs keep falling. The Unitree G1 now lists for about $16,000. It’s about the size of a middle-schooler but boasts several advantages: 

  • Much cheaper than having an actual middle-schooler in the house.
  • No mood swings.
  • Runs out of power after two hours.
  • Knows how to cook.

If you haven’t seen your first robot—on the street, in an office, or wherever—you likely will in 2025. I wasn’t 100% sure until I saw the showstopper from CES 2025: a new robot, Aria, by Realbotix, which was supposedly designed for “companionship.”

“Aria” via You Tube.

 

Reactions were split between the conventioneers who: 

  • Thought to themselves, That is very clearly a sex robot. 
  • Were kidding themselves.

Elon Musk claims there will be a robot in every home within five years. Sound fantastical? Bill Gates once made similar predictions about personal computers, and we didn’t believe him, either.

This raises practical design questions: Where do robots live? Do they need their own closets or charging stations? Does my robot just spend the night standing in the corner of the living room, looking sad? That seems depressing.

One thing is clear: there is no 30-year horizon where robots do not appear. Forward-thinking architects must begin to consider robot presence as a basic design parameter, just as we currently consider human occupancy or climate considerations.

Parametricism Evolves Into Full-On World Building

In 2024, OpenAI’s SORA text-to-video platform hinted at a future for architectural design. To generate even a simple video of a frisbee throw, SORA must understand the physics of how objects move through space. However, SORA was never trained in physics. It trained by watching millions of hours of video, including videos of people throwing frisbees. Through that training, SORA developed an intuitive understanding of physical rules, much like humans do (e.g., we typically learn the physics of frisbees by playing frisbee, not by studying physics). This is machine “world-building,” the process whereby machine intelligences develop an understanding of our world and become capable of generating new ones.

The video game industry is already demonstrating the power of this approach. For The Matrix Awakens, Epic Games created an entire digital city—150 miles of roads, 7,000 buildings, 18,000 moving vehicles, and 35,000 autonomous pedestrians—not by designing each element, but by establishing the rules of the world. As Epic’s Kim Libreri explains, “Instead of modeling each building directly, we define parameters: Is it a downtown zone? An industrial zone? Then the engine fills in all the detailsbuilding windows, concrete sidewalks, air conditioners, AI-driven trafficeverything.” Using these frameworks, designers generated 52 different versions of the city, each taking just two hours to create.

 

This evolution of rule-based design extends architectural traditions like parametricism and New Urbanism: instead of designing individual elements, we design the systems that give rise to them. 

This has always had its limits. Architects promoting ideas of “Total Design” and Gesamtkunstwerk received appropriate criticism in their day for aggrandizing control fantasies, which ultimately seemed illusory and insincere to boot. The world is just too large and complex a place to believe that an architect can control it with a floorplan.

When Mark Wigley wrote, 25 years ago:

“This fantasy [Total Design] is still very much alive. These days, the teaspoon doesn’t seem small enough and the city doesn’t seem large enough. Students don’t hesitate to develop projects on the architecture of the microchip or on networks for interplanetary transportation.”

He couldn’t have anticipated the way that technology now makes the generation of entire worlds—microchip to space station—possible, maybe even necessary.

Artificial People Join Architects’ Toolbox

The characters in The Matrix Awakens preview architecture’s next game-changing tool: Artificial, Simulated People (ASPs). An architect works by applying grounded speculations of how space, material, and one’s client will interplay over the life of a building. Recent experiments raise the possibility of designing with simulation, in lieu of speculation.

Stanford researchers have developed a way to simulate the behavior of actual, individual humans. By interviewing 1,052 real people and training individual AIs on their responses, they created 1,052 digital avatars that can predict their human’s behavior with 85% accuracy.

For an architect, this creates several possibilities:

  • Creating a digital avatar of one’s actual client and letting that avatar inhabit the design virtually, in order to see how they react to it and use it. 
  • Speeding up time within the simulation by 10,000x to see how the building is used over its entire useful life, somewhat like doing the POE while still in SDs.
  • The “time-plasticity” would also make it possible to talk with a “future” version of your present client.

Would working with an AI avatar of one’s client actually be easier than the real thing? Probably, yes, right up until the point where your client disagrees with their own avatar. Depending on how you feel about your client, however, merely being able to watch them have an existential meltdown might just be worth the price of admission.

Digital Twins Finally Have Their Moment

The aforementioned gives me a strong feeling that in 2025, digital twins will finally have their moment. They’ve been hyped for forever. I know a bunch of people who are working on them, but, oddly, no one seems particularly excited about doing so. It’s been hard to realize their value, for both architect and client. But there are a few strong trend lines that suggest this is about to change:

  • Building them gets easier: Faster, cheaper world-building engines make the design of digital environments easier, and faster.
  • Adding ASPs makes them more valuable: The inclusion of ASPs enables digital twins to predict beyond just environmental performance. People are the most dynamic part of any building, and ASPs allow for the simulation of both people, their environment, and how the two interact. 
  • Embedded AI, and robotics, create an alternate data-gathering channel: Historically, to create a functional digital twin meant lots of sensors that could gather up all the real-time data you needed, measure it, catalogue it, and report out. These dedicated sensor networks were costly to design, install, maintain, and monitor. The increasing appearance of AI in things creates an economical new portal for data collection that doesn’t require dedicated sensor networks. The robots that work in an office building can sample the environmental data, the AIs in cars can relay information about the degradation of roads, and so forth. The “Internet of Things” we’ve always been promised may take shape indirectly via embodied AI.

Leadership in the public sector: The list of cities building digital twins is growing and includes New Rochelle, New York, and Wellington, New Zealand. As cities do so, it enhances the value of private-sector digital twins that can exchange information with the city’s overall model. 

Construction Industry Adoption of AI & Robotics Evolves From “Wide” To “Deep”

Contracting is already on the steep part of the adoption curve. In a 2024 BD&C survey, two-thirds of general contractors (GCs) reported using either monitoring robotics or service/labor robotics. However, half of those report that they’re only yet piloting solutions, suggesting that, so far, construction’s tech adoption is wide, not deep.

In 2025, labor shortages will drive up GCs’ costs and pressure them into wide and deep adoption. The situation will be exacerbated if the Trump administration is successful in its stated deportation plans. Because so many contractors have already been experimenting, expect the overall industry adoption to proceed quickly from here. Drones will be first, and ubiquitous, on job sites by the end of 2025. AI-enabled forecasting will also give rise to a new headache: the AI-predicted delay. Think Minority Report: the contractor submits a change order for a delay, not because it happened, but because the AI thinks it will happen. And, obviously, the future delay that hasn’t happened yet will still somehow be the architect’s fault.

AI Shakes Up Conventional Notions of Firm Size, Provokes New Business Models

This year will begin a shakeup in firm size and market distribution, culminating in the debut of “AI-native” firms. The massive productivity gains brought about by AI will position enterprising architects to create new firm models that punch way above their weight class. A small firm, aided by a bevy of intelligent architectural agents, can now do the kinds of projects that were traditionally reserved for midsize firms. Meanwhile, solo practitioners can take on the work once reserved for small firms.

These roiling market dynamics also predict the debut of AI-native firms: firms that have been founded and built from scratch on a foundation of AI technologies. AI substantially lowers the time and cost of starting up a firm. And if the new firm’s non-design functions can be delegated to AI, starting a new firm also starts to seem much more appealing. If I were starting a practice today, I might handle those non-design functions with Vic.AI to do autonomous accounting, AILawyer to handle legal, Alberti.AI for digital marketing, Fullstory for full digital experiences, and so on. On the design side, I might use Swapp AI for documentation, Veras for visualization, Aino for feasibility studies, and so forth.

Existing firms will continue to ask themselves: “How do I apply AI to my current systems?” AI will allow new firms to run wild, free of such constraints, and develop semi-autonomous, “AI-native” practices. Both types will brand their AIs as “intern AIs” or “emerging professional AIs” so that no clients get confused about who’s actually carrying the expertise.

Looking Ahead

In every type of firm, architects still skeptical of AI will be converted once they realize that AI has finally allowed architects to reclaim the exclusive mantle of “architecture” by putting all the software engineers out of work. What we do with that mantle remains to be seen. We’re in an era where our climate is collapsing and inequality is exploding, among numerous other issues. As if that weren’t enough, our clients might be AIs, our occupants will definitely include robots, and our toasters will all belong to Mensa. I’m not panicking, though. Whatever challenges lie in the year ahead, architects still possess the one tool capable of solving any wicked problem: the ability to design.

For a full breakdown of Cesal’s 2025 predictions as well as ongoing insights into the confluence of design, technology, disaster and resilience, visit his Substack and subscribe for free. Featured image created by the author via Midjourney.

Newsletter

Get smart and engaging news and commentary from architecture and design’s leading minds.

Donate to CommonEdge.org, a Not-For-Profit website dedicated to reconnecting architecture and design to the public.

Donate