On June 3, Amber Book hosted AI in Architectural Practice: What Works, What Doesn't, and What's Next with Jacob Russo, AIA, senior computational design architect at Skidmore, Owings & Merrill (SOM), AI for Architects instructor at ELVTR, and School of Continuing and Professional Studies (SCPS) Lecturer at Pratt Institute. More than 1,000 architects registered.
That kind of turnout doesn't happen for box-checking continuing education (CE). It happens when practitioners feel that something in their profession is shifting and want to understand it on their own terms, not from a vendor pitch.
If you were there, this is your recap and a chance to revisit the content. If you missed it, here's what the room learned and how you can go deeper.
According to Bluebeam's Building the Future: AEC Technology Outlook 2026 report, only 27% of AEC firms are currently using AI in a meaningful way. But 94% of those firms plan to expand their use this year. 68% of early adopters have saved at least $50,000. Nearly half have reclaimed 500 to 1,000 hours.
Jacob opened with those four stats, setting the tone for everything that followed.
And then this: only 20% of firms feel highly prepared (ASCE 2026).
Jacob's read on that last number was direct: "The gap isn't awareness. Architects know AI is here, and more is coming. The gap is really implementation. And that's exactly what today is about."
That framing held for the rest of the session. The session wasn't pitched at skeptics trying to decide if AI is worth their attention. It was built for practitioners ready to move from awareness to action: which tools, in which phases, evaluated how, adopted by whom.
Before getting into specific tools, Jacob laid out a six-phase landscape: pre-design, schematic, design development, documentation, construction, and operations. AI has meaningful applications at every stage — but the tools differ, and so does the ROI.
In early phases, tools like Forma, Veris, and Crea are showing up before a firm has proposed anything concrete, analyzing site conditions and generating early massing options. In design development, parametric tools like Raven use natural language to write and modify Grasshopper scripts — including generating working parametric definitions from a reference image, with no text prompt required.
Documentation is where Jacob pointed to the clearest, most immediate ROI. A useful starting point is understanding what AI is doing in a BIM workflow: augmenting tasks that still require interpretation and judgement or automating tasks that just require knowing the rules and executing them consistently. Tasks that have no design judgement involved are exactly where AI delivers ROI fastest, because the time these tasks consume is disproportionate to the thinking they require.
Sketch Pro, a co-pilot that lives natively inside Revit, handles sheet creation, view setup, dimensioning, tagging, and QA review through natural language commands — and can cross-reference your firm's own drawing standards to do it consistently. The numbers Jacob cited come from Sketch Pro's own data: 20+ hours saved per designer each month, and 30% more project capacity with the same team without hiring.
His gloss on those figures: "That's a business model argument, not just a workflow argument."
That reframe matters. Most internal conversations about AI adoption start as operations questions: can we make this faster, can we reduce rework, can we cut rendering time? Those are legitimate questions. But 30% more project capacity with the same team is a different kind of answer. It's a growth conversation, and it belongs in a principal meeting. The firms moving fastest on AI are the ones where leadership understands the business case and drives adoption with intention.
The skills AI makes most valuable in practice are the ones architects have been developing since studio.
The role, he said, is shifting from producer to curator — and that's the reassuring part. The curatorial skills that matter most in that shift — recognizing quality and design intent in AI output, describing a building's material logic and lighting conditions precisely enough that AI can execute it, knowing when an output is leading you away from design intent rather than toward it — all come from design training, not tech fluency.
"A vague prompt gets a vague answer. An architect who knows how to give precise, contextually rich prompts is going to get dramatically better output than one who doesn't."
The translation from architectural vocabulary to prompt engineering, Jacob argued, is more natural for architects than for almost any other professional. The same training that teaches you to describe material logic, compositional intent, and lighting conditions is exactly the training that makes you good at AI prompting.
Jacob also broke down three tiers of sophistication in how architects are using LLMs –– and the gap between them matters. At the practical level: drafting emails, proposals, and project narratives. At the significant level: comparing systems, summarizing specs, reviewing documents. At the highest level: using the LLM as a genuine thinking partner –– stress-testing rationales, identifying scope gaps, evaluating project risk with actual project context loaded in. Most practitioners are operating at tier one. The firms moving fastest are pushing toward tier three.
Jacob spent time on data privacy because, as he put it, it's where he sees the most confusion in practice.
The short version: free consumer tiers of LLM tools train on your data by default, which means project descriptions, client names, and drawing content could be used to improve the model. Paid business tiers (ChatGPT Team and Enterprise, Claude Pro and above, Gemini for Workspace) don't. And if a vendor’s data privacy policy is hard to find on their website, that’s a red flag.
"The cost of a team subscription is negligible compared to the liability risk of a data exposure."
For firms working on government or healthcare projects, Jacob was even more direct: vetting data residency and privacy policies isn't optional, and it needs to happen before testing a tool, not after.
The section most attendees flagged in the Q&A was Jacob's four-step tool evaluation framework: define the problem first, check data and integration fit, assess timing honestly, then run a real pilot: something with real constraints and real team dynamics.
What makes a pilot a pilot, he said, is a measurable outcome. Reducing the time to generate a cartoon set from four hours to one hour is a testable claim. You either hit it or you don't.
And: "The decision to abandon a tool is just as valuable as the decision to scale it — if you record why."
The internal champion matters as much as the tool itself. One person who owns the pilot, documents what worked and what didn't, and advocates for or against expansion turns a two- to four-week experiment into a firm-wide decision based on real evidence.
New tools will keep coming, and many will have the same underlying limitations as the ones that didn’t work. A firm that has documented its evaluations doesn’t have to start from zero –– that institutional knowledge carries forward.
Jacob closed with a three-to-five-year view — some of which is already beginning, some of which is projection. Self-documenting BIM that monitors models in real time and checks code compliance as a background process. Multimodal design workflows where describing a building in words or sketches generates parametric options, not just static images. And firm-trained models built on captured project data, decisions, and outcomes.
That last one is the development Jacob called most strategically interesting for firms: "Firms that systematically capture project data, decisions, performance outcomes, [and] client feedback, will be able to train AI that encodes their institutional knowledge. That's your secret sauce. That's what is your IP. And to be able to train AI on that information, that becomes a competitive asset that compounds over time."
Want to experience it for yourself? The full session is now available as a free preview CE course on the Amber Book site, worth 1.5 LU elective credit.
Access the course here: https://www.amberbook.com/continuing-education-preview-course/
If you attended the live CE course, you've already earned your 1.5 LU for the session. The course gives you a way to revisit the content at your own pace, including the full tool walkthrough material, the pilot framework in detail, and Jacob's emerging roles and skills breakdown for architects navigating the next few years.
The course is free, self-paced, and built around exactly the implementation gap Jacob described: knowing AI matters is one thing; knowing how to actually use it in practice is another.
If you want to keep going beyond this session, an Amber Book subscription unlocks the full CE catalog, including dozens of practitioner-led courses built around real projects and real decisions.
Explore subscription options: https://www.amberbook.com/continuing-education/
NCARB updated its online proctoring requirements on June 1, 2026 — here's what changed and what to do about it...Read More
Getting licensed as an architect in any U.S. jurisdiction is no small feat. It takes years of education, thous...Read More
The AIA Conference on Architecture is like the Super Bowl for architects. It’s that perfect mix of connecting ...Read More
