The Exchanges

Q on like AI for materials discovery, which is an interesting sub niche of this, this whole universe. Um, less, less on the engineering side where I know you spent a lot of time and I find pretty interesting as well. Um, but yeah, I mean, how, How bullish are you on this, like, using AI to do better discovery, or I guess design is maybe the other component of this?

A Yeah. I'm quite bullish. I mean, you just look at what's happening in software engineering, right? I mean, I think the way Cursor and tons of other software engineering tools are taking off, um, is pretty wild. And I think it actually speaks to, like, we may even see multiple big companies in a given category. Like in software engineering, you're seeing folks who are helping you with your existing code base, folks who are helping you with new projects, folks who are helping you translate from one language to another, that those are, might all be different sort of companies doing that. Um, so I think this is like really similar to agentic AI of how can we do work better, faster, cheaper, and actually can we take the parts of the engineering work that engineers don't enjoy doing off their plate. Um, so I have an early investment in a company called Catstrom. They're building a co-pilot for electrical engineers on the PCB design side. They're starting on the verification and validation end of things, like checking downstream as opposed to sort of design upstream. What's good about that is the engineers often don't enjoy doing that and have to spend a lot of time sort of error checking their stuff. Um, so it does like delight the user in that way. And I think the layer cake they're building here is actually quite replicatable to sort of other engineering domains as well, which is t…

AI assessment note: “Yeah. I'm quite bullish. I mean, you just look at what's happening”

Q is, is that a reason for there to be something specialized? Or could you take all that unstructured data and all those PDFs and just upload them to whatever generic LLM, upload them to Anthropic or, or ChatGPT or whatever it's going to be, and it'll synthesize them just as well. Like, is there a reason why unstructured data is better suited to a particular, like a specialized model? Yeah.

A We think this relates maybe actually to the third point, which is most of this data is private. So most of the data is private and proprietary and lives on these customers' cloud in a best case scenario, maybe on-prem, um, but certainly not in sort of like the public domain where OpenAI has trained on it. And so, um, I think the reality is these models are just not that good today at sort of understanding, um, you know, a very technical diagram, a blueprint, because they haven't seen enough of them. And so you can imagine, like, AEC, like, architecture, engineering, construction is a particularly good example, um, because there's a lot of geometric and spatial representations. You need to understand that design intent, um, proprietary training data, and most of that doesn't look like, you know, the sort of normal tech space or video, um, data that, that's on the internet sort of pretty broadly.

AI assessment note: “these models are just not that good today at sort of understanding”

Q as a better knowledge base. It's to solve the problem of all the data and all the information being siloed in a million different places or held by people who are retiring or whatever. So it's, I think of this as just being like a really, really smart encyclopedia that is specialized to the individual needs of the individual sector, the individual customer. Am I thinking about that one right?

A I think it's more than encyclopedia. I think it's, like, making sense of complexity. So, um, maybe one way to think about it is, like, I think people are just wasting a lot of time today, searching for information, manually cross-referencing sources, or making decisions based on incomplete data. And I think a really important part of that is, like, again, data across systems and marrying that. Um, so, like, maybe one very simple example that I'll give in sort of construction land, um, Is, say you're on a construction project and your question is, you know, when are the light fixtures that are going on the second floor coming in? It's a really actually quite easy question conceptually, but today quite hard to answer because we need to do a few things. We need to process that natural language of understanding what light fixtures on the second floor means and referencing that against likely blueprints or schematics or whatever that show exactly what fixture we're referring to, and then taking that understanding of which skew we're referring to and looking that up in your supply chain system. We're talking about probably a few different sort of types of documents, a few different systems, and marrying that data to answer what's a relatively simple question, actually, but today requires someone to, probably requires multiple people to look into multiple systems to, like, figure out …

AI assessment note: “I think it's more than encyclopedia. I think it's, like, making sense of complexity.”

Q you stack up your ARR numbers and you've, you've got all your metrics. In AI world, there's at least some discussion of kind of turning the pricing model, at least on its head. I'm curious what you've been seeing from all these companies that are out there doing AI in the physical world. Is it, is it a standard software as a service type offering or is there innovation there?

A Yeah, so I think the, the, the choices in a very black and white sense are sort of sell the work or sell the technology, right? So selling technology is sort of a traditional SaaS model you're talking about of, um, offering your AI product to the user to use on their own. Um, selling the work is saying, hey, I'm just gonna actually do this whole task for you. I'm gonna consume the AI myself and sort of obfuscate whatever human in the loop or services are needed on top of that as part of that service. Um, I think where we're finding that to be most interesting to customers is where they are already outsourcing. So if you're already outsourcing your compliance work, for example, to a third party consultant, you're probably sort of willing to outsource that to a different company instead. And especially if that company is going to staff up, right? So that's always the question is like, what kind of services are you actually providing on top of the technology? But if you're actually going to provide me an account manager or case manager, And someone who's experienced in the space and is going to effectively be my new consultant, but by the way, they have technology behind them now, then, like, maybe that could work. Um, I think this, this will have interesting implications to margin profile, actually, like, as will the technology broadly, and I think this is something we should all…

AI assessment note: “the choices in a very black and white sense are sort of sell the work”

Q industries and in industrial sectors, I don't know, whatever we want to call it. Um, why don't you start with just like the high level thesis? You've been spending a lot of time in this space. Walk me through at the, at the highest level. How are you thinking about where there is value in the applications of, of this new wave of AI to these big, heavy industrial industries?

A Yeah, so I think the first thing we've been thinking about is sort of, like, whether and why and where there will be sort of, like, specialized vertical applied AI products versus using sort of general tools here. So I think there's, like, no perfect answer. We're hearing, you know, anecdote on both sides. Like, I know there are some sort of general purpose, really high growth AI tools that I hear are losing some deals to, you know, vertical specific players because they are better suited to that Context. But on the other hand, like I know law firms that are using Anthropic instead of Harvey or legal AI tools, so there's definitely anecdote on both sides. I don't think we know exactly where this is going to net out. I think my theory is that there is room for and a real need for actually sort of like vertical applied AI tools in the industrial and physical economy categories. I think there are a few reasons why. I think most of them live on the sort of like supply side of this, of, you know, why technology And how technology will serve this sector well. And this is mostly about the customer's data. So like, if we think about the customers here across a bunch of different sort of industrial categories. So think manufacturing, natural resources, like mining and materials. Construction in the built environment, supply chain and logistics, all these categories, what we're talking a…

AI assessment note: “my theory is that there is room for and a real need for actually sort of like vertical”

Q That is especially true in these In physical industries? Like, is it more true that the data is siloed and in PDFs relative to other industries? Or is it just that that's like true universally? And so there's a, per that criterion, there is an opportunity for verticalized things kind of anywhere.

A I think there's probably some of it everywhere, but I think more of it here. And I think the tooling, again, is more legacy as opposed to like, I don't know, if you imagine you're a large tech company, you've probably built a lot of your own tooling or using like contemporary software, as opposed to software, like if you look at, I think we were looking at once, um, you know, companies that have built sort of multi-billion category defining products selling to hardware engineers, and the last one was started in like the nineties. Maybe even the eighties. Um, so there's just not a lot of contemporary products here compared to what you imagine the tech stack or sort of back office stack even looks like for a parallel technology company.

AI assessment note: “I think there's probably some of it everywhere, but I think more of it here.”

Q Um, ok, so, so legacy systems of record, all this unstructured documentation, And then the data is private. It's a walled garden. What else?

A So those are the main things on the, like, data side, which I think, again, lead me to believe that a specialized product, we can talk about that's the model itself or something else, will be more performant than, you know, the alternative, again, sort of general tool. I think there are a couple demand side factors, but I, I think really the main thrust is what we just talked about. On the demand side, from the customer piece, I think one thing we're seeing is, you know, the great crew change, which applies to a lot of these categories where, uh, Um, there's a bunch of sort of skilled, experienced workers, whether that's, um, field engineers, um, field technicians, all these kinds of things who are reaching sort of age of retirement, and there's not enough folks to fill the gap behind them, and that means two things. One, we actually just need to do more work with fewer people in the future, as we have sort of a potentially smaller labor force to do the same work, and or that knowledge is actually retiring, and so we need a way to capture that expertise and sort of make sure we have a lot of that locked in into a product, Given that today it lives in people's heads, and that's how these systems sort of operate over time. So that's one big piece.

AI assessment note: “On the demand side, from the customer piece, I think one thing we're seeing is, you know, the great crew change”

Q like, is more the question of, like, can you build it? Can you build it economically, I suppose, is the question. And when I say it, I mean something specialized and verticalized and good enough to compete with whatever, however the performance would be of a generalized solution trying to do the same thing. So how do you think about the sort of cost to build something bespoke like this?

A Yeah, so I think the question is, How you build it, and what needs to be true there. So, I think on the technical side, there's no perfect answer, but in talking to, you know, founders who are actually building, it seems like prompt engineering can get you quite far, right? So like, it's widely variable, but let's say you can get to like, 70, 90% of where you need to be from a performance perspective. I think the question is whether that's sufficient for the task at hand, and whether like, you're going to consume the AI Internally, or the customers are going to fill the gap with human in the loop. Um, because prompt engineering, obviously, is sort of cheap, scalable, all those kinds of things. I think the, like, layers beyond that would be, are you embedding, um, are you doing vector embedding, are you doing fine-tuning, or are you actually building your own model? Um, I think I'm hearing really widely variable things there. I'm hearing founders say, you know, embedding is important, we're doing it, but it doesn't work very well for non-textual data. I'm hearing some folks say, like, fine-tuning is sort of like a natural, um, Evolution of prompt engineering, and I've heard other folks saying, we're going to build our own model, and that's just part of what we need to do in this category. Um, I think the reality is, like, we're at the very early innings of this, and there are lo…

AI assessment note: “prompt engineering can get you quite far... prompt engineering, obviously, is sort of cheap”