Panel discussion (hosted/created by Christopher Parsons) exploring whether AI can help AEC firms build custom tools without negative trade-offs. A concrete anecdote describes a façade project where team members used AI-assisted coding to translate/expand Python to C and regenerate a façade pattern across 1,800 individually different panels after the original programmer left. Speakers debate ‘vibe coding’ (ad-hoc AI-assisted coding) as a pragmatic fix that risks reduced knowledge development and a "knowledge liability," while also noting AI’s positive roles: interrogating uncommented code, documenting intent, finding issues, and upskilling staff to reuse bespoke tools.
Questions this video answers
1,800 panels
October 19, 2025
Chapters
There was a great vignette, you guys. So for people listening and that haven't seen their K connect talk yet. I don't know. They're going to come out around the same time, so please go watch it. But there was a section around, I don't remember the project name, but you were doing these facades and you were writing custom programs, I think it was in Python to be able to even do all of the complex facade. But then the person that wrote the program left.
Yes.
Then. Then what happened?
This never happens. This doesn't happen.
Never happens. I know.
It's the.
It's the classic knowledge management conundrum, right? You have a very heavily knowledge based task or project and the person who had the knowledge leaves. Something happens and they leave. They're not there. And. And that happened here. And because the team had some facility with AI tools for coding, our IT team has facility with that and our different team members had been experimenting with it and so they were able to use AI assisted coding tools to bridge the gap between what had been written in Python to what needed to be written in C and expanded to recreate the facade pattern across 1800 panels that were all individually different. And so they used the AI tool to close that knowledge gap. And I respect that as a technique. I think it's a smart solution to the problem and it certainly got the job done. The question that I raised though is they were vibe coding and if everybody decided to vibe code every time we need to do computational design, I feel like that would be unsustainable and would lead to less knowledge development than more.
Less knowledge development. And also just liability. Like liability. Code is knowledge, right? You know?
Absolutely. Yes. Yeah. So, you know, it's one thing with a facade pattern, there may be a low liability there, but if, I guess.
I mean like a knowledge liability, not a. Not a.
No, no, no, you're right.
Safety and welfare. Liability isn't where I was going.
No, fair enough, Fair enough, Lia. A broad term. No, but you're right. Yes. It's a knowledge liability.
Yeah.
I think what's super interesting about this thread is that you can use the large language models to interrogate code and say, tell me what this is doing where that was never a thing before like that. And so to help bridge that gap can also become a learning process to upskill people who are taking on those new responsibilities. Because every firm out there has code laying around for some bespoke tool on a project. It happens, right? Like not every tool can be reused. A lot of stuff just is purpose built for that one thing. But there's a higher chance that it could be reused if there's something that can explain it to me, who has never looked at that spaghetti of whatever that is.
Uncommented code.
Uncommented code, Grasshopper scripts, Dynamo scripts, whatever those things are to then teach me. Oh, and then. And also, like, look for issues. Right. I think this is a pretty amazing development that I don't think we really saw coming.
Yeah. And that's actually where we. Where we went from that conversation in our presentation about the Vibe coding and the potential liability that it brings to the value that AI can actually add there by not only helping to bridge that knowledge gap, but also to document what they did and why. And as you say, you can also use it to go back and query the thing that you didn't know about. So there's actually, at the same time, there's a danger and a liability implied. There's an opportunity at the same time that AI can answer. And that's what's really interesting. And sometimes frustratingly.
Yes, I could imagine.
Yeah. It sort of creates its own conundrum where it's truly a blessing. Yeah.
