249: 'Level Two Autonomy for Architects', with Julian Geiger
A conversation with Julian Geiger about bringing practical AI into AEC workflows, automating repetitive tasks while preserving design judgment, and building shared project context across tools.
Julian Geiger joins the podcast to talk about what it takes to bring AI into the software architects and builders use every day. As Chief AI Officer at the Nemetschek Group, Julian works on AI inside products like Archicad, Solibri, and Bluebeam, and on how the company uses AI internally. The conversation covers why the hardest problems in AEC are about process, collaboration, and information, why AI should take on the code lookups and documentation grind, how a shared intelligence layer across tools could keep project context intact, and why design judgment may end up worth even more.
This episode is especially relevant for firm leaders, BIM managers, and design technologists trying to separate AI demos from AI they can rely on. Julian's advice is practical. Use the tools yourself, watch how fast they change, and get your project data organized well enough that a new hire could make sense of it.
Original episode page: https://trxl.co/249
Watch This on YouTube:
Jump to a section
0:00 Introduction
2:33 Julian's path from Google to Nemetschek
6:38 Why AEC's problem is process, not technology
13:14 Why change the industry, and what AI should take on
16:38 From idea to BIM in seconds, with live verification
21:36 The GPT-6 villa demo and the constructability gap
23:39 Level two autonomy for design AI
28:00 Why today's AI is the worst you'll ever use
31:16 CAD benchmarks vs. design judgment
37:34 Why architects stay needed as taste earns a premium
40:19 One AI strategy across Nemetschek's brands
51:34 AI in Archicad, Solibri, and Bluebeam today
59:27 Telling users what's new when nobody reads release notes
1:15:27 Where to start: use the tools yourself
1:18:34 The new-hire test for your project data
1:22:53 Let the AI interview you about a folder of your work
Episode Links:
Connect with the Guest
- Julian Geiger
Nemetschek and Its AI Strategy
- Nemetschek Group
- Why it's relevant: Julian is Chief AI Officer here. Nemetschek is the parent company of Graphisoft, Bluebeam, Allplan, Solibri, dRofus, and more, and he says AI is the reason the group is integrating more than ever after years as a financial holding company.
- Nemetschek AI Layer announcement (October 2024)
- Why it's relevant: Julian refers to the intelligence layer Nemetschek announced in October 2024, the shared AI foundation meant to carry project context across the group's tools.
- Open Collaboration Platform
- Why it's relevant: Julian talks about Nemetschek using this cross-tool platform internally first, and about openness as the reason customers should choose the tools, not lock-in.
AI in the Tools Architects Use Today
- Graphisoft Archicad and the AI Assistant
- Graphisoft AI Solutions
- Why it's relevant: Julian calls the first Archicad AI Assistant a beta experiment and describes what it could do from the start, like resizing every window in a model or a first-pass code compliance check in some jurisdictions, with a deeper version on the way.
- Solibri AI-powered Assistant
- Why it's relevant: Julian's example of AI taking the pain out of rule setup for model checking.
- Bluebeam
- Smart Overlay drawing comparison
- Why it's relevant: Julian points to Bluebeam's AI-powered review and drawing comparison, which shows up as a regular button in the interface rather than a chatbot. Smart Overlay looks like the feature he means.
- Allplan
- Why it's relevant: One of the Nemetschek tools where Julian wants AI to have the full project context, whether that context lives in Archicad, Allplan, Solibri, dRofus, Bluebeam, or third-party software.
- dRofus
- Why it's relevant: Another Nemetschek brand on Julian's list of places project context lives today.
AI Models and Tools Mentioned
- GPT-6 Astra (OpenAI)
- Why it's relevant: The model that shipped the week before this conversation was recorded. Julian cites its big jump on a CAD benchmark next to a much flatter result on design judgment, and its slick villa demo that still says nothing about constructability.
- BenchCAD
- Why it's relevant: The CAD reconstruction benchmark behind the scores Julian mentions.
- Claude Code and Claude Cowork
- Claude Cowork
- Why it's relevant: Julian's story about his father, an accordion, and a driver Claude Code rewrote, plus my suggestion to point these tools at a copied folder of your own work and have them interview you.
- ChatGPT
- Why it's relevant: One of the tools I suggest connecting to a copied folder of your own work so it can interview you, and the kind of foundation model that already understands what you mean when a site search doesn't.
- Perplexity
- Why it's relevant: Where Julian's father started before moving to Claude Code.
- Obsidian and Markdown
- Markdown Guide
- Why it's relevant: Julian runs his personal second brain on Markdown notes in Obsidian, with his agents feeding it and using it for context.
- Reverb
- Why it's relevant: My example of a site search that still can't handle a typo, in a conversation about AI interpreting what you actually mean.
Ideas and References
- Levels of driving automation (SAE J3016)
- Why it's relevant: Julian uses level two driver assistance, where the car keeps the lane but you're still responsible, as the level AI in design still has to reach.
- Clippy (Microsoft Office Assistant)
- Why it's relevant: I bring up Clippy, and Julian explains why old in-app suggestions failed. "Business letter" was far too broad a category, and there wasn't enough intelligence behind the suggestion to beat what you'd do yourself.
- Dogfooding
- Why it's relevant: How Nemetschek is testing its own collaboration platform internally before customers get it.
Previous TRXL Episode Mentioned
- 238: Half Your Day Isn't Design, with Holger Kreienbrink of Graphisoft
- Why it's relevant: Julian points back to Holger's point on that episode about how manual and step-by-step design work still is.
About Julian Geiger:
Julian Geiger is Chief AI Officer at the Nemetschek Group, the parent company of Graphisoft, Bluebeam, Allplan, Solibri, and other AEC software brands. He leads the group's AI and data strategy, covering both AI inside its products and how the company uses AI internally. Before joining Nemetschek in 2024, he was a product manager at Google, where he led manufacturing product globally, and earlier worked at BMW and started his career at Boston Consulting Group. He is not an architect by training and focuses on turning fast-moving AI progress into tools AEC professionals can rely on.
