Thinking AI-First: How Architects Are Adapting to the Future, Now

By Danielle DiLeo Kim, AIA, 2025 President of AIA Philadelphia

In the last six weeks, I’ve started a new habit. Each time I sit down to begin a work-related task, I ask myself: “How can I use AI to enhance, expedite, or advance this?”

Then I act on it—using the tools I’ve been testing so far: Perplexity, ChatGPT, Gemini and NotebookLM. It’s a small shift, but a significant one. This mindset is called “AI-first thinking,” and I adopted it after hearing Allie K. Miller, a leading AI expert, deliver a powerful keynote at AIA Conference on Architecture & Design 2025 in Boston. She offered a much-needed wake-up call: we must get comfortable with AI now to improve our people, processes, and products in our practices.

AI is here—and it’s already reshaping how we practice architecture.

To explore how our community is integrating AI into daily work, I turned to AIA Philadelphia’s Technology + Innovation committee and asked several members how their firms are approaching this fast-moving frontier. Four design leaders generously shared their insights: Christopher Connock and Brandon Cuffy of KieranTimberlake, Rolondo Lopez, AIA of DIGSAU, and Esra Abumounshar, Assoc. AIA of Stantec. 

How is AI assisting your practice?

DIGSAU
Over the past year, our office has explored a wide range of AI tools aimed at improving administrative tasks, enhancing design visualization, and standardizing documentation. While the potential is clear, many of these tools are still in the early stages of development. 

Stantec
The company is integrating AI to assist with various administrative tasks that are often time-consuming and can benefit from enhanced efficiency and quality assurance. AI technologies provide staff with additional input, enabling approaches to design services that may not have been considered previously due to budgetary or scheduling limitations. This allows the exploration of a wider range of design solutions, with AI serving as a collaborative tool for generating ideas and obtaining prompt feedback.

KieranTimberlake
AI and/or Machine Learning (ML) is already working at a variety of scales within our practice  – automating the mundane through transcription and summary of meetings, elevating  quality through clustering thousands of similar details into digestible groups, and allowing  us to see beyond our own personal histories at KieranTimberlake through a direct integration with our knowledge management system. 

What kinds of AI tools are you using?

KieranTimberlake
Our team is experimenting across categories:

• Generative Design: We have long used simulation and evolutionary solvers to study hundreds of approaches  to scoped design problems like façade shading strategies. The setup and simulation of  each study was labor, computation and time intensive. We have begun testing surrogate  models trained on the results of many past complex (CFD) simulations to reduce the time  from intuitive design move to understanding of impact from hours to seconds. Autodesk  Forma has already integrated surrogate model driven wind and noise analysis in a  relatively easy to use interface.  
• Visualization: We have typically shied away from start-to-finish text-to-image generators (Midjourney,  Stable Diffusion). As architects, we design spaces for communities, not images. Our sites  have constraints and communities have needs that are not captured in the generative  models of today. The act(s) of design are iterative and multivariate – an “alchemy of art, science, analysis, and intuition” that typical generative image making sidesteps. That said, we heavily rely on images and drawings to communicate our design intent along the way. 
  • For this reason, we have adopted ML image editing available in many of our day-to-day  tools ranging from the Adobe Suite to Chaos Enscape. We can use ‘in-painting’ in Photoshop to rearrange parking in site photos, replace planting and add entourage in an  interface the whole office is familiar with. Enscape’s one button ‘AI Enhancer’ jumps the  uncanny valley by making their out-of-the-box entourage look more photo-realistic.  
  • Until recently, generative models could not achieve scene consistency through each  iteration without many additional tweaks in the prompt or model chain (Masks,  ControlNets). The release of Flux.1 Kontext enables editing an image with natural language  while still retaining all other aspects of said image – changing the season, sky or entourage  without accidentally adding an unintentional bit of architecture along the way. We are currently testing Flux as well as smaller local models trained on our own images (LoRA) to give us both flexibility and consistency of style in our image editing.  
• Documentation: We are currently using Pirros to harvest and index the details across all our projects. While  the main value add of Pirros is the extraction of our many details, it does have ML influenced suggestions and groupings of similar details that help us sift through the  haystack.  
  • As with visualization, some of our tools have already started to leverage AI. We are testing BlueBeam’s ‘Visual Search’ to identify and count instances of elements on sheets, quality  assuring takeoff and estimation.  
• Research: Our knowledge management platform, Knowledge Architecture Synthesis, is actively  developing AI integration in stages, building from general searches to employee queries to  project requests. We can now ask, in natural language, about our approaches to energy  modeling, life cycle assessment or stakeholder engagement across multiple projects.  
  • AI has become integral to the research and discovery process. Whether product research  or summarizing and categorizing extensive notes from stakeholder engagement sessions, Large Language Models (LLM) in particular, have helped speed up the data distillation  process.  
  • Our team is also actively testing our own modified (Retrieval Augmented Generation or  RAG) system focused on the documents (submittals) that may not live on our intranet but have a wealth of project history.  

DIGSAU
We are making use of:

• Digital Assistants: Among the most widely recognized tools are digital assistants like ChatGPT. I’ve found this tool to be helpful for tasks like paraphrasing, spell-checking, and refining text for emails, presentations, and reports. What I find especially powerful is its ability to process large amounts of information in seconds. I can upload documents, have the assistant read them, and then ask questions or discuss the content. This has been a game-changer for analyzing reports, product specifications, and technical literature. 
• Meeting Summaries: Our office utilizes Zoom for virtual calls. They have recently introduced an AI Companion that is available while taking calls. It listens during calls and automatically sends a summary afterward, highlighting key discussion points and next steps. This has made post-meeting follow-ups much more efficient. 
• Visualization Tools: Our office has also been experimenting with AI tools for design visualization and rendering. This area is evolving quickly, with many companies currently making similar software. So far, I’ve found these tools useful for enhancing rendering materials and lighting in renderings we produce in house. However, they’re not yet reliable enough to replace traditional modeling or rendering tools, as they often struggle with prompt accuracy and consistency across multiple iterations.

Stantec
We see value in:

• Administrative Tools: AI tools such as Microsoft Copilot have been incorporated into day-to-day activities, including checking drafts, auditing documents, reviewing compliance, summarizing documents, taking meeting notes, and identifying important tasks within email inboxes. These tools assist with organization and help users stay informed about ongoing tasks.
• Design Ideation: We are increasingly integrating AI into our design process to generate ideas, interpret client goals, and address functionality and sustainability through advanced tools.
• Visualization and Simulation: AI is central to our visualization efforts, enhancing animations and helping us communicate spatial experiences to clients. We now prioritize workflow tools with AI that support sustainable design from the conceptual stage, enabling real-time carbon emission and environmental analyses. This transformation allows our practitioners to deliver more sustainable, responsive designs and improve project outcomes.

What excites you most about AI’s potential?

KieranTimberlake
One immediate opportunity when employing AI is the chance to more holistically consider data quality and curation. The knowledge management stick has truly turned into a carrot –  there is real value in building and maintaining data conventions.  

The personalization of AI, to our own histories, contexts and tools is incredibly exciting. Smaller light weight models can run locally and address domain (Architecture, Engineering.  Construction) specific tasks. Subsequently, these lighter models can be more easily  integrated into native authoring environments used by designers, lowering barriers to entry, by having AI tools in the same place you’re working. Model Context Protocols (MCP)  are an increasingly popular way to leverage powerful state-of-the-art frontier models like  GPT4 or Claude AI into our everyday tools in an intuitive way.  

The collaboration with AI, through agents that are not limited to one question or authoring  environment, is also high on our radar. AI Agents can initiate and manage complex tasks with minimal human intervention that would ordinarily require many prompts and specific  directives for an LLM to complete in a satisfactory manner. This represents a shift from a reactive to a more proactive use of AI technology in design.  

Stantec
The implementation of AI presents the potential to deliver high-quality results without contributing to burnout or compromising the final product, while still meeting client and project requirements. When applied appropriately and responsibly, AI serves as a valuable resource for ensuring alignment with clients' objectives and maintaining project momentum.

DIGSAU
I’m excited about AI’s ability to work hand-in-hand with Architects to improve our ability to visualize and deliver high quality projects. Ideally, these tools would help to automate some of the more time-consuming administrative tasks, so that we can spend more time designing, collaborating, and documenting our projects.

What concerns you about AI in practice?

DIGSAU
Like any emerging technology, AI comes with its share of concerns. 

Environmental: One of the key issues is environmental. AI systems require substantial energy to support their computing power. As demand for this technology continues to expand, there are concerns about carbon emissions, power grid instability, and water resources. 

Authorship: Another concern relates to authorship. AI models are trained on large datasets, often compiled from publicly accessible content online. These datasets are curated and managed by the companies that develop the tools, but there’s often limited transparency about the sources of this information. This raises valid questions around authorship, attribution, and data ownership. As a general practice, I avoid uploading or sharing any content that could be considered private, sensitive, or confidential when using AI tools. 

Stantec
Over reliance: As with any emerging technology, the implementation of AI presents several challenges, particularly if staff are not adequately trained in its use. Employing AI without a solid understanding of best practices can be counterproductive, especially within design practice—since results generated by AI may lack relevance to clients if users are not well-versed in building or spatial design requirements. There is also the potential risk that overreliance on AI as the primary source of ideas may hinder creativity and critical thinking. It is essential to develop a cyclical relationship with AI-generated solutions, using them to enhance rather than replace our overall design approach. Users must establish a strong foundation for AI to provide informed and effective solutions. 

We have encountered similar obstacles during the transition to Building Information Modeling (BIM). While BIM was intended to streamline processes, improve coordination, and reduce costly changes, these benefits are only realized when the tool is used correctly. Our project experience has shown that designers may spend excessive time resolving issues caused by low-quality BIM models—often due to the automatic generation of 3D elements—which only seasoned practitioners can readily identify and address during review. This can compromise overall design outcomes and potentially lead to legal ramifications. The same considerations apply to the integration of AI in our workflows.

KieranTimberlake

Some of our concerns are about:

Authorship and Attribution: Most models are trained and finetuned on vast repositories of data, some of which are in  the public domain, and unfortunately, some of which are copyrighted and being used  without the creator’s consent. Attribution and authorship are some of the most ambiguous topics in the realm of AI use. In May, the United States Copyright Office released a  thoughtful and nuanced pre-publication report on the rights of materials used in generative training that gave arguments to all parties – content creators and model makers  alike. As we were crafting this very response, the White House released an ‘AI Action Plan’  that skews towards the development of generative models over content rights.  

There is a need to understand if a given AI platform processes data locally or on the cloud,  which raises questions about data sensitivity and how much data you are willing to share. This is why it becomes important to interrogate AI software Terms of Service, where we can  understand if the company will use our data to improve its model, if we retain the rights to  media we produced, or if they have an opt in/opt out policy for training on our data. For  this reason, we have begun to put specific protocols and policies in place when evaluating  AI-enabled software in our practice.  

Community / Environmental Impact: More broadly, AI requires a significant amount of electricity and water consumption for  training many of the well-known foundation models. This impacts communities far from  end users. Entering a simple prompt into an LLM may seem trivial, but somewhere the ecological footprint of that LLM’s training is felt. As designers, though the technology is exciting, we must consider the ramifications of its use, as we are tasked with crafting discourse around more responsible design and ecological stewardship in the built environment.  

Expectation Management: As we increasingly engage with AI tools, both social and technical concerns have arisen that  require careful consideration. Expectation management for users is critical to adoption, as  AI is often thought of as a single source solution for many design problems. However,  these tools are only as effective as the data and guidance they are given. Clearly  communicating what it can and cannot do is important to understanding how we  effectively integrate AI into our practice. In applications where it can meaningfully  contribute, results require a rigorous quality assurance process to validate output quality, to build trust among users. Due to the nature of architectural work, there are complexities  and nuances that current AI technology cannot adequately encode, requiring human  participation in any workflows involving AI. 
Staff Development: And finally, AI can complete tasks in the design process that may not require significant  experience but could be essential to building up the knowledge base of young emerging professionals. The use of AI must be balanced carefully with professional development, so as not to remove opportunities that would otherwise benefit the experience of younger designers.  

Who’s using AI in your office?

Stantec
Our roles vary: Project Managers use AI for organization and oversight, Designers for ideation and documentation, and Architects/Engineers for problem-solving and sustainability.

DIGSAU
Our adoption is mixed, with some team members using AI tools daily, and others still exploring.

KieranTimberlake
Everyone—though to varying degrees. Team members can be described as:

• Consumers (basic users),
• Curators (organizing knowledge), and
• Creators (building custom pipelines and models).

Are you collaborating with other disciplines on AI?

Stantec
Yes, our work extends beyond architecture and engineering—we closely collaborate with marketing and graphic design teams to shape a comprehensive AI strategy. By working across disciplines, we ensure our designs and AI-driven solutions are clearly communicated to clients, manage expectations, costs, and quality, and keep our business innovative and competitive.

KieranTimberlake
Soon! We want to, it is absolutely part of our ethos, and critical to the development of the  technology. We have a few opportunities in the works.  

Are there guidelines or best practices in place?

KieranTimberlake
Yes! When developing internal AI guidance for our office, we looked across the industry, to see  how others were addressing the topic. After an extensive survey, it was clear that many  were at the beginning of this process and figuring out how to responsibly align AI with their  practices. After synthesizing what we found, an AI Code of Ethics and Practices was established at KieranTimberlake, covering topics such as permitted use cases, approved AI  enabled software, quality assurance, disclosure, and data security. The landscape of AI  policy and best practices continue to rapidly change; therefore, our AI Code of Ethics and  Practices is a living document that will be amended over time to reflect the current  consensus on guidance for AI tools.  

An important corollary effort was developing forums for discussion like ‘Near Future Practices.’ This task team takes on new technologies and workflows, asking: Is it Possible?,  Is it Practical?, Is it worth pursuing? Through this task team, staff are brought together to discuss tools, outline evaluation methods, and test technologies within relevant projects across the office. The results of these evaluations, if the tool or workflow is deemed  beneficial, yield best practices that are shared among the office. This process builds  dialogue and a shared vocabulary around AI across all practice areas within our office,  daylighting opportunities and limitations of use.  

Stantec
As the AEC industry develops best practices, Stantec is creating a company-wide approach. We are currently focusing on early design concept image generation and integrating AI into our processes. Additionally, we've established a design automation team to support staff with AI tools, set usage guidelines, provide access and troubleshooting, and help teams achieve better results through effective prompts.  

DIGSAU
AI should not be seen as a substitute for design intent or design exploration. It should also not be used as a substitute for proper due diligence in research and code analysis. 

If there’s one thing that remains constant in the world of technology, it’s change. Much like the transformative impact of tools such as CAD and BIM in previous decades, AI has the potential to enhance productivity and open up new possibilities in design and project delivery. Through all these changes, our mission remains the same: to uplift the human experience through the power of architecture.

Final Thoughts

For anyone that is still holding out, Esra recommends that while “change may be uncomfortable, but it's the catalyst for growth. The tools we have today hold the potential to unlock success—but only if we commit to learning and using them wisely. When we invest in understanding and applying them with purpose, we don't just keep up with change—we lead it.”

From improving workflows to opening new possibilities in design and sustainability, AI is becoming a critical partner in architectural practice. The AIA Philadelphia community is not only adapting—but actively shaping—the future of this technology. I encourage you to try AI-first thinking if you are not already doing it. Start with something simple and see what you discover or uncover.

Thank you to the Philadelphia architectural community for sharing insights so we can move forward with curiosity, responsibility, and boldness—together.

In community,

Danielle