A short interview-style conversation (YouTube transcript) in which Christopher Parsons argues that AEC firms overestimate what AI will do for them if they rely only on unstructured repositories (e.g., proposal text). He contrasts two approaches—dumping many proposals into an AI versus investing in structured, authoritative project and boilerplate databases—and recommends structured data as a long-term source of truth to avoid inconsistent or outdated outputs from AI. The transcript includes examples (project sheets, changing project timelines, mismatched numerical facts) illustrating risks of unstructured-only strategies.
What do you think is a major blind spot that AEC firms have today when it comes to leveraging their knowledge?
Overestimating that AI is just going to do it for them, being too, I think, too, too optimistic about what AI is going to be able to do and too optimistic about the time horizon at which it's going to improve without you doing. Like, its reasoning capabilities are increasing really, really fast. Like, it's able to handle images and multimodal. It's like there's all kinds of awesome stuff happening with AI. It doesn't know your context, it doesn't know your values, it doesn't know your project history. It does, you know. So, like, you can't just throw. Can I take longer than like a little quick answer for this one? All right, there is a. I'm going to call out two ends of a spectrum around a way to approach knowledge in AI that I've seen. I have strong, strong takes. One of them is we're going to take 800 of our proposals, we're going to throw them into some kind of folder, point AI at it, and we're going to ask IT questions, and that's how we're going to answer marketing questions. The other one is we take a little bit more time. We're going to build boilerplate libraries. So this is our, like, this is how we describe our approach to mass timber or sustainability or equity or whatever the thing is, or our firm. And we're going to get solid project descriptions and they're going to go into a database like an open asset or a Dell Tech or a uninet or something like that. So structured information versus completely unstructured. Now, of course, I'm calling out two ends of the spectrum. There will always be something in the middle. But the point of view there, I believe that the structured database is going to kick the unstructured database's ass every single time. So when you ask a question like I'll do a couple, like, what's our approach to match timber? You might have written that approach in four different, 10 different, 30 different ways across those 800 proposals. And you don't know you're pulling a slot machine on which one you're going to get, you know, or let's just even go more. More like, let's say that you kind of cited one of a project that's still in construction and you cited its final construction cost and square footage or whatever. And then you cite it again in another proposal six months later and those numbers have changed. And when you then go Ask what was the square footage of this project like again? It's a slot machine. Which one are you going to get? So I think this hoping that AI will figure it out. With unstructured data, it can do something. It's better than nothing, I will give it that. But I think if you really want to have great outcomes with AI over the long term, you're going to want structured data as well as unstructured data. Like the example I gave before the housing project with the freeway, that is captured through unstructured data. But your core data points and descriptions and point of views and boilerplate stuff like please write it down in one place and make it a source of truth. That's my point of view.
Well, yeah, and I mean, just thinking of the project example, as marketers we start including project sheets for jobs as soon as we get them awarded. Right. And so the narrative for those is, you know, we're current, we're actively working it, it will be completed in. And then.
Right.
The project progresses and that narrative shifts to reflect the current timeline of it and then finally it's done. But then if you just have your AI pointing to your repository of proposals, it's not smart enough to know the current sense of the job. And if you've got a new person there, how are they going to know?
Correct? Exactly. You just hit on one of the key trends and I can't believe you've gotten this far without saying it. When you optimize content for AI, you're auto optimizing for humans and vice versa. So AI thrives on high quality, accurate, well explained data with context. We share the why, whatever it is, that's what humans want too, you know, and so you said it really well. And so like how is it, how is a new marketing coordinator going to know of this wall of 800 proposals and the AI? If I ask it today, it gives me this answer. If I ask it 10 minutes later, it gives me that answer like how do they know which answer is right? Like that just seems you're not setting them up for success. AI or a human, you're not setting them up for success.
