Summary:
In the final Chaos Talks AI session of 2025, experts from Chaos, SHoP Architects, and Creative Lighting discussed how AI is transforming visualization and design workflows in AEC. While AI accelerates ideation and content creation, the panel emphasized that it works best as a co-pilot. Human judgment and expertise are essential for guiding AI, spotting issues, and delivering meaningful design outcomes as more advanced, agentic AI tools begin to emerge.
AI is moving fast in the AEC industry. Faster than many workflows, faster than education, and sometimes faster than our ability to decide how and where it should be used.
In the last Chaos Talks AI session for 2025, host Phil Read was joined by Bill Allen (Chaos), Mengyi Yan (SHoP Architects), and Nikos Nikolopoulos (Creative Lighting) to explore what’s really changing with AI, what still requires human expertise, and why judgment remains the most critical skill in an AI-powered workflow.
Here are the key takeaways from the conversation 🍿💡.
AI gets you close, but rarely all the way there
Across visualization, planning, and early-stage design, AI has made major leaps in speed and quality. Image-to-video generation, prompt-based layouts, and automated ideation are now part of everyday experimentation in many firms.
Phil Read pointed out that while AI makes it easy to generate outputs quickly, turning those outputs into something meaningful still requires judgment and intent.
The group agreed that AI often produces results that look convincing at first glance, but don’t always hold up under closer inspection. Mengyi Yan, Director of Visualization, shared examples from SHoP Architects’ internal testing. “If you zoom in really closely, that’s when you start seeing all the hallucinations.”
For now, AI works best as an accelerator, not a replacement for design thinking.
The last mile still belongs to humans
A recurring theme was knowing when to stop prompting and start making decisions. “I use AI until it starts frustrating me, and then I stop and just do the rest manually,” Phil Read said.
That frustration is often a signal that judgment is needed. Without understanding how to do something manually, it becomes difficult to fix AI output or even recognize when it’s wrong.
From a visualization perspective, Creative Lighting's Founder and Creative Director, Nikos Nikolopoulos, emphasized the same point:
“I've seen some huge improvements this year in quality, in speed, the AI tools, they're getting better. But you know, without traditional knowledge, how can you judge? How can you complete your story or your idea or your vision? So basically we need that traditional knowledge.
If you want to judge AI, you need to have knowledge about lighting and composition and all this kind of stuff.”
Faster doesn’t always mean better, but it can mean more
Rather than reducing workloads, AI is often increasing output. Mengyi compared AI to earlier technologies like 3D printing. “When 3D printers came out, you didn’t suddenly stop needing model makers. You still needed these people to bring in their expertise to tell you how to do things the perfect way and to put it together and make sure it doesn't fall, and to understand materiality and tell you what's the best material to use for this kind of printer. I think AI is sort of a similar tool.”
At SHoP Architects, AI hasn’t replaced existing workflows. It has expanded them.
“We’re not producing the same amount of work. We’re producing more work. And hopefully through that work we can get to that one gem of an idea faster, like that one perfect image faster.”
AI works best as a co-pilot, with us in the driver's seat
One of the key distinctions discussed was how AI differs from previous technologies - how AI introduces a form of reasoning that changes what’s possible in design workflows.
As Bill Allen (Founder of EvolveLab, now part of Chaos, which created the AI visualization tool, Veras) explained:
“… the one thing that is different about this technology is that no other technology before it was able to reason. And so that's where these tools are very, very different… it can take information and reason about it similar to how we think. So it's a very, very different technology and it opens up the door for different and new possibilities.”
That shift doesn’t remove the need for human involvement. Instead, it changes the nature of it. Designers are no longer just executing tasks, but actively guiding how these systems are used.
As Bill put it later in the discussion:
“You’re the orchestrator… there’s a reason we call it copilot… we’re supposed to be in the driver’s seat, we’re supposed to be the pilot of these things.”
Example of an AI-generated image produced by Veras
Stick with the tools that work, and go deeper
With new AI tools launching constantly, the panel warned against chasing every update. Nikos advised that people should find the tools that work for them and stick with them.
“I think it's time to stop investigating and experimenting with every new tool. I think it's time to invest and go deeper in specific tools that do the job.
So let's say you're doing video, find the tool that you like and stick with, because otherwise you're gonna be always fighting and waiting for the next update and the next big thing. Whatever works for every person, find the tools, stick to them because they're gonna get better, all of them.”
Instead of experimenting endlessly, teams benefit more from committing to a small number of tools and building real workflows around them.
Judgment, accountability, and human connection still win
Despite the fast evolution of AI, the panel was clear on one point: tools don’t replace trust.
Mengyi Yan made the point that while AI can support the design process, it doesn’t replace the need for judgment or tailored solutions.
Clients don’t hire architects or visualization teams because of software. They hire them for their decision-making abilities, accountability, and the experience of working together 🤝.
Agentic AI is the next phase
The discussion also touched on the idea of agentic AI.
Most AI tools today are still transactional. You ask for an image, a video, or a layout, and you get a result. Agentic AI points to a different model. Instead of executing single tasks, these systems begin to coordinate multiple steps toward a goal. In practice, that feels less like using a tool and more like working with a team, with a workforce.
“I think... we're gonna have what we're experiencing right now, but then I think this next phase... is the agentic AI piece.
So your orchestrating your project and saying I want 10 different perspectives of my project. I want them animated. I want you to also create 20 animations of those. I want you to go through and document all of the project. It's going to start to connect all these different pieces of the project lifecycle that we all do as architects, and then going to be the orchestrator of all these different pieces.” explained Bill.
Final thoughts
AI is changing how quickly we can explore ideas and communicate design intent. But it hasn’t changed what matters most. Judgment, craft, and human responsibility still define good outcomes. AI simply makes those qualities more visible and, when used well, more powerful.
In 2026, we’ll be diving into new topics as part of our ‘Chaos Talks’ series. Stay tuned for more information coming soon!
Catch up on the first two sessions of Chaos Talks AI:
> Discussing AI's role in architecture, design, & visualization
> Navigating the state of AI in AEC & the future