Only 6 percent of American architects use AI regularly, according to a March 2025 AIA report. The studios that do are compressing a week of work into an hour.
In the spring of 2024, Kelly Wearstler described a working session on a legacy hotel project in Lake Tahoe. Her team prompted an image model for millwork details, brutalist in style, 1970s in period, and received fifty options in ten minutes. Her estimate of the same study done the traditional way: a junior or intermediate designer working for up to a week (The Grand Tourist). Her Los Angeles studio, roughly 55 people by her own count, now has a dedicated digital and AI innovation team, runs ChatGPT, Midjourney, DALL-E, and Claude inside its workflow, and was building its own language model. "Every day we integrate new and better A.I. tools into our process," she wrote in 2025, "and our studio is becoming more human, not less" (WEARSTLERWORLD).
That is the glamorous version of the story, and it is real. But the more consequential version is quieter, and the survey data gives it away. When Houzz studied 722 American construction and design firms in 2025, it found that among interior designers using AI, 70 percent applied it to administrative work like emails and document summaries, 59 percent to sales and marketing, and only 34 percent to design work itself (Business of Home). The famous experiments are in the renderings. The recovered hours are in the inbox.
The short version: Architecture and interior design firms now run AI in two distinct layers. The visible layer is design ideation, where firms like Gensler, GRAFT, and Kelly Wearstler's studio compress weeks of concept iteration into days or hours using Midjourney and similar tools. The invisible layer is the business of the studio, proposals, client updates, briefs, and case studies, where surveys show most of the real adoption is happening. This briefing covers ten sourced cases across the United States, Canada, Mexico, Brazil, Germany, and Singapore, separates audited numbers from vendor claims, and closes with a 90-day starting sequence for a principal.
The pattern in one line: the studios winning with AI run it on iteration speed and on the communication layer, under a written policy and a senior review standard, and never delegate taste. The model drafts and iterates; the designer decides.
The American engine room
The largest architecture firm in the world treats AI as a presentation and iteration engine. Gensler co-CEO Jordan Goldstein, writing in Fortune in September 2025, described generative video, scenario modeling, and real-time rendering used to prototype how people will feel in a hospital or airport before it is built, and put a number on the change: "What once took weeks of iteration, we now explore in days." His framing is deliberate: "We're not just using AI, we're shaping it to reflect the integrity of our craft. Rather than pushing a button, we're building advanced, customizable workflows that honor the design process" (Fortune).
Dallas-based Corgan, one of the ten largest American firms by revenue, did something rarer than adopting a tool: it ran a controlled experiment and published the results. In March 2023, its in-house research team, Hugo, put groups of its own designers in front of DALL-E 2 with photographs of Corgan's Dallas office and a sequence of progressively detailed prompts, then surveyed them. The published findings read like a field manual: designers are open to the technology, it is most useful at concept design and visualization, visual prompting is a trainable skill, and the designers themselves were split on whether clients should ever touch the tools directly (Corgan). Corgan designers also fold Midjourney and Autodesk's Spacemaker, now Forma, into iterative design work (Corgan).
Perkins&Will, the Chicago-headquartered interdisciplinary practice, is the strongest American example of AI moving past experiments into infrastructure. By late 2024 the firm had run more than 30 internal AI experiments and built two tools worth copying. The first is a custom iterative rendering application that turns simple massing models into stylized or photorealistic renders and, critically, allows edits to specific parts of an image rather than regenerating the whole frame. The second is an internal language-model tool named Chato, Portuguese for "boring," which does exactly the boring thing every large studio needs: it compares and corrects room, department, and space names across the space program, the BIM model, and the cost, material, and carbon data, replacing manual variance reports (Perkins&Will). The firm published AI ethics guidelines early, requires designers to disclose AI use to clients, and in December 2024 won Interior Design magazine's Best of Year award for Outstanding Integration of AI in the Design Process for its Cognitive Care Center project, one of the few independent validations in this entire field (Interior Design). "AI will not replace what gifted designers can create," says Nick Cameron, the firm's director of digital practice. "Instead, it will amplify their talents."
Seattle-born NBBJ gave the discipline its best phrase: "AI on the shoulder." The firm uses language models to summarize design research and develop narratives, machine learning trained on past projects to predict timelines, and surrogate models that stand in for compute-heavy daylight and wind simulations (NBBJ). And HOK supplies the necessary caution. "Design is primarily driven by reflection," says Paul Harrison, a design principal at the firm. "The dangerous part of working with AI is when you use it to minimize the amount of reflection" (HOK).
Now hold those firms against the profession's own numbers. The American Institute of Architects reported in March 2025 that only 6 percent of architects regularly use AI in their work, and only 8 percent of firms have implemented AI solutions, with another 20 percent implementing, led overwhelmingly by firms of fifty people or more (AIA). The AIA's broader 2024 firm survey put about a third of firms at some day-to-day AI use (AIA). In the UK, for comparison, RIBA found 59 percent of practices using AI in 2025, up from 41 percent a year earlier (RIBA). The gap between the firms in this section and the median firm is not talent. It is the decision to treat AI for architects as an operating discipline instead of a toy.
The image generators get the press. The hours come back somewhere else.
The interior design frontier
AI for interior designers is the same race run with a smaller engine and, in places, more nerve. The 2024 baseline was 16 percent of interior designers using AI, per a 1stDibs survey. By 2025, Houzz had it at 31 percent, and the firms using it claimed productivity gains worth about $74,400 a year, a figure Business of Home rightly flags as soft, built from roughly three hours saved per week and extrapolated (Business of Home).
At the top of the market sits the Wearstler studio, covered above. At the volume end sits Denver-based Havenly, which spent a decade having human designers produce roughly 2.4 million room renderings and then, in October 2025, trained a consumer design engine on that proprietary archive. A user uploads a room photo, chats with the bot, edits elements with a press-and-hold gesture, and gets handed to a human designer when the request outgrows the machine. Founder Lee Mayer's logic is the entire industry's dilemma in one sentence: "Either someone is going to do this and disrupt us, or, hey, we're actually pretty well positioned to do it ourselves. We've got the data, we've got the designer perspective, we've got humans in the loop" (Business of Home). In its first week of testing, free AI users were already converting into paid human-designer engagements.
Between those poles, the honest picture of the working interior designer is the Houzz breakdown: AI on the inbox, the proposal, and the Instagram caption first, and on the mood board second.
Canada: the pragmatists
The best window into Canadian practice is a November 2025 Globe and Mail investigation of how AI is redrawing the country's firms (The Globe and Mail). Toronto's Core Architects, the firm behind One Bloor West, has spent two years working AI into practice. It leans on Midjourney for concept inspiration, produces AI fly-through videos of unbuilt towers, and has its BIM specialist building custom internal tools to automate repetitive work, with one director dedicating part of each week to AI research. Partner Brian Laye is both convert and skeptic. The tools inspire "like any firm looking for images in books or magazines," but ask one for a 100-square-foot bedroom and it may hand you a room two feet wide and fifty feet long, technically correct and functionally useless. His conclusion: "No inventions have replaced the need for critical thinking."
The same article catches the governance instinct that separates large firms. At 350-person Diamond Schmitt, public tools like Midjourney are banned outright; the firm visualizes early ideas with proprietary in-house AI for privacy and quality control. "It's saving us quite a bit of time and effort," principal Victor Lima told the Globe. Vancouver's Henriquez Partners used AI to composite its Venice Biennale tower studies into Canaletto landscapes, with founder Gregory Henriquez calling the technology "a wild horse right now" that "requires rigour and experimentation." And in Montreal, the startup Maket, which raised a $3.4 million seed round in October 2025 and says it has passed a million registered users, sells instant floor-plan generation to builders and developers, a reminder that some of the disruption arrives from outside the profession (Tech Funding News).
Mexico and Brazil: the client walks in with a Midjourney image
In Mexico City, Michel Rojkind of Rojkind Arquitectos, the firm behind the Cineteca Nacional renovation, has been publicly documenting his AI experiments since early 2023. The studio tested DALL-E and Midjourney, settled on Midjourney for image quality, and uses ChatGPT to write better Midjourney prompts, in one documented exercise generating an image series from the stanzas of David Bowie's Space Oddity. More interesting than the technique is where he points it: at the client. The firm treats AI images as digital sketches for proposals and generates them together with clients in the room, a co-creation workflow. "It's interesting, the way we communicate with AI, how it learns from us and we from it," he told an industry forum (Centro Urbano).
SĂŁo Paulo's EstĂşdio Guto Requena, a UNESCO Prix Versailles winner that has kept computer scientists at the design table for years, states it plainly: artificial intelligence is part of the company's daily routine. Not just renderings. The studio uses AI for floor-plan composition, project visualization, office organization, and even floor-tile pagination (CASACOR). Requena has also named the shift every studio owner in this article will recognize: clients now arrive at the office with Midjourney images of what they want. "AIs bring a democratic way for clients to show what they like," he says. "Architecture offices will need to adapt to this."
The same CASACOR report documents the interiors angle. Designer Gabriel Rosa presented his CASACOR SĂŁo Paulo 2025 space, Adega Legado, to the show's curators using concept images that were entirely AI-generated, before the project was fully resolved. "I managed to position myself clearly before having a 100 percent finished project," he said. He also uses AI to clean up client briefings and to speed up his social posting, and draws the same line as everyone else in this briefing: the welcome a client receives must stay human.
Clients now arrive with Midjourney images of what they want. The studios that thrive treat that as a briefing gift, not a threat.
Germany: confidentiality first
German firms supply the governance blueprint. Berlin's GRAFT, the studio of Lars KrĂĽckeberg, Wolfram Putz, and Thomas Willemeit, started testing image generators in mid-2022 and built the most complete studio AI system documented in Europe. It standardized on Midjourney after testing DALL-E, Stable Diffusion, and the interior-specific tools, built an office prompt library encoding its own render styles, formed an internal AI working group open to any employee with the firm paying for memberships and training, and, tellingly, pays for Midjourney stealth accounts so client designs never enter the public gallery. It has gone further: an in-house assistant named EDDI, built on OpenAI and trained by the firm for its own programming and analysis tools, plus offline local models fed by the studio itself, explicitly to protect authorship and project confidentiality. AI-assisted concept studies appear in real pursuits, including the ZLB Berlin central library and the EXPO 2030 German Pavilion. Putz's summary deserves printing out: AI is "a devilishly good inspiration tool," and yet "we still need our entire professional knowledge as interior designers and architects to steer the process and get a usable, financially robust plan" (GRAFT; BauNetz ID).
Hamburg's gmp, von Gerkan, Marg und Partner, one of Germany's largest practices, runs the sober large-firm version: an isolated, on-premise image model installed on its own network, trainable on gmp's own project photos and renderings "without the software phoning home," used to turn hand sketches into renderings and to generate facade and interior material variants. The firm trains staff formally in prompting, uses Autodesk Forma for AI-accelerated early-stage analysis, and holds one rule above all: every AI output passes an additional quality-assurance review by experienced planners. "The biggest risk in using AI is the naivety of the users," Andreas Dieckmann of gmp told DETAIL in January 2025 (DETAIL).
The German chamber of architects' 2025 survey of 15,624 members shows how far ahead those two firms are: 29 percent of German architects use AI in daily work, rising from 19 percent at offices under five people to 67 percent at firms over fifty. And the top use, by a wide margin, is not rendering. It is text: 58 percent of AI-using architects apply it to writing work like PR copy and meeting minutes, while only 16 percent qualify as intensive design-process users (Deutsches Architektenblatt). Even in the country of the digital twin, the communication layer is where adoption actually lives.
Singapore: the hard numbers
The single most concrete result in this briefing comes from Southeast Asia. DP Architects, Singapore's largest architecture firm, joined the government's AI Trailblazers initiative with Google Cloud and concluded that off-the-shelf models could not deliver what it needed for urban design work. So its digital transformation team trained a custom model on curated architectural data plus synthetic data. Within eight weeks, the firm reported productivity up more than 50 percent and ten times more design iterations in the same amount of time, with every output traceable back to its original template for audit and review. The model is now integrated into multiple internal tools and automates technical drawings (Singapore EDB). "We quickly realised that the output we want to achieve is unobtainable with the models accessible to us," says Chan Hui Min, who heads the firm's Smart Sustainability Unit. "Hence, we had to develop and train our own AI model." The figures are self-reported within a government-published case study, not independently audited, but the traceability requirement alone puts DPA ahead of most Western firms on governance.
The region also produced the most radical interiors story anywhere in this article. Singapore entrepreneur Hajar Ali, with no formal design training, built Reverse Orientalism, a registered architectural and interior design business, on AI generation. Her Instagram following went from about 10,000 at the start of 2023 to over 382,000 by November 2023, and the inbound clients included top developers in Qatar and the UAE. "AI is a democratising force in the same way that the invention of the printing press facilitated the dissemination of ideas," she told The Business Times. "It has been able to speed up the workflow tremendously, visualising spaces not within a day but in minutes. The speed at which I've been able to hand over projects using AI terrifies even me" (The Business Times). For an established studio owner, she is not a curiosity. She is the new competitive floor: if concept imagery alone can be produced this fast from outside the profession, concept imagery alone is no longer the moat.
The layer nobody photographs: where the hours actually return
Put the ten cases side by side and a pattern appears that none of the individual headlines show. The image-generation experiments are real, but the most senior people in this article keep pointing somewhere else. Corgan's tested takeaway was that AI lands first at concept communication and documentation. Perkins&Will's most valuable tool is named "boring" and reconciles spreadsheets. Germany's chamber survey found text work is the number one use. Houzz found 70 percent of AI-using interior designers apply it to admin and 59 percent to sales and marketing, against 34 percent for design. Gabriel Rosa won over CASACOR's curators with AI concept images, but his day-to-day uses are cleaning up briefings and feeding social channels.
That is the layer where a boutique studio actually recovers hours: the proposal and fee letter, the Monday client update, the post-meeting summary with a decision log, the scope-change response, the sourcing inquiry, the award submission, the project case study that never gets written because everyone is billing. None of it requires generating a single image, none of it touches the part of the work that is yours, and all of it currently eats the evenings of every principal we know. The studios above prove the discipline scales from a 55-person Los Angeles practice to Singapore's largest firm. The same discipline fits a studio of four, and the failure mode is identical at every size: AI output that reaches a client without a senior review pass. The model drafts. The designer decides. That is the entire operating system, and it is the system we teach.
How to read the numbers
Most figures in this field are self-reported, and the honest reading sorts them into three tiers. The most credible carry some independent check: Perkins&Will's Best of Year award from Interior Design magazine's jury, the Globe and Mail's independent reporting on Core Architects and Diamond Schmitt, and Business of Home's critical coverage of Havenly, which tested the tool and reported its flaws alongside the claims. The middle tier is government-published but firm-supplied: DP Architects' 50 percent productivity and 10x iteration figures sit inside a Singapore EDB case study. The third tier is the firm talking about itself: Gensler's "weeks to days," Wearstler's "week to an hour," GRAFT's and gmp's accounts, and every adoption survey that relies on what respondents say they do. Read the third tier as direction, not audit. The direction, across six countries and two professions, all points the same way.
And one line stays fixed in every credible case. Nothing that reaches a client, a contractor, or a permit authority goes out without a qualified human signing it. gmp formalized it as a quality-assurance rule, Perkins&Will as written ethics guidelines with client disclosure, Diamond Schmitt as a ban on public tools, GRAFT as stealth accounts and offline models. The firms moving fastest are the ones with the strictest rules, because the rules are what make speed safe.
How do you actually install this in a studio?
The research above shows where the hours return: proposals, client updates, meeting memos, scope letters, and the case studies nobody has time to write. The installation layer is a reviewed workflow system: a clean room policy, a never-upload list, a 4-checkpoint senior review, and 10 named studio workflows you run on your own documents. We packaged that system as the AI for Architects and Interior Designers course, a step by step communication system for studios. It costs $395, and it never touches design generation; taste stays with you.
Frequently asked questions
How are architecture firms using AI in 2026?
In two layers: a design layer for concept ideation and visualization, and a business layer for communication and documentation, and the business layer is where most real adoption happens. On the design side, Gensler, Corgan, GRAFT, and Core Architects use tools like Midjourney and Autodesk Forma, while Perkins&Will and DP Architects built custom rendering and iteration models and NBBJ runs simulation surrogates. On the business side, Germany's BAK survey found 58 percent of AI-using architects apply it to text work, and Houzz found 70 percent of AI-using designers apply it to admin.
What percentage of architects and interior designers use AI?
As of 2025 reporting: AIA found 6 percent of US architects use AI regularly and about a third of firms report some day-to-day use; RIBA found 59 percent of UK practices using AI, up from 41 percent in 2024; Germany's BAK survey of 15,624 architects found 29 percent overall, rising to 67 percent at firms with more than 50 staff; and Houzz found 31 percent of US interior designers using AI, roughly double the 16 percent a 1stDibs survey found in 2024. Adoption is strongly correlated with firm size.
Which firms have the most credible AI results?
The ones with some independent check: Perkins&Will, which won Interior Design magazine's 2024 Best of Year award for AI integration; DP Architects in Singapore, whose 50 percent productivity gain and 10x iteration figures were published in a government case study with audit-traceable outputs; and the Canadian firms whose practices were independently reported by The Globe and Mail. Figures from Gensler, Kelly Wearstler, GRAFT, and gmp are self-reported and best read as directional.
Will AI replace interior designers and architects?
No credible case in this briefing points that way, including the most aggressive adopters. Kelly Wearstler: ideas are cheap, "80 percent is in the execution." HOK's Paul Harrison warns the danger is using AI "to minimize the amount of reflection." What is changing is the economics around the design act: clients arrive with Midjourney images, concept visuals are no longer a moat, and the studios that win run AI on iteration speed and on the communication layer while keeping taste, judgment, and liability with a human who signs the work.
What should a small design studio do first with AI?
Not image generation. Start at the communication layer, where the Houzz and BAK data show the hours actually return: a clean-room policy for what never enters an AI tool (drawings, fees, client identities), then one reviewed workflow such as the client project update or the proposal draft, then a prompt vault so the studio's voice compounds instead of resetting every Monday. That sequence, policy first, one workflow, then a vault and review standard, is exactly what our course AI for Architects and Interior Designers installs in 90 days.
The 90-day starting sequence
A principal who wants the results above without the headcount of a Gensler should run the adoption in this order.
- Weeks 1 and 2: write the clean-room policy. One page. What never enters an AI tool: drawings and BIM files, client names and addresses, fees, margins, contractor bids, unannounced projects. Diamond Schmitt banned public tools; GRAFT bought stealth accounts; gmp went on-premise. Your version can be a page of rules and placeholder conventions, but it must exist before the first prompt.
- Weeks 3 to 6: ship one communication workflow. Pick the highest-friction recurring document, usually the client project update or the proposal, and build a reviewed Claude workflow for it: raw notes in, drafted document out, senior review pass, send. One workflow, run weekly, beats ten experiments.
- Weeks 7 to 10: build the prompt vault. Save every prompt that produced client-ready work, with substitution variables, the way GRAFT built an office prompt library and Corgan treated prompting as a trainable skill. The vault is a firm asset; it is what makes month two faster than month one.
- Weeks 11 to 13: add the review standard, then the second workflow. Write the four-checkpoint review every AI draft must pass: accuracy, tone, technical precision, client readiness. Then, and only then, expand: meeting recaps, scope-change letters, the case studies your portfolio is missing. Document the hours recovered; that number is what justifies everything else.