Engineering’s Spatial AI Moment - Campfire & Gravity Sketch

Michael Finocchiaro: we're live. ⁓ Hello, everybody. is welcome to the AI Across the Product Lifecycle podcast. Now sponsored by AWS.

There's a link to download their white paper on agentic AI. ⁓ I'm joined by Jay ⁓ founder of ⁓ and Oli say, ⁓ I can get the ⁓ amazing gravity sketch. I finally got I got to meet him finally. ⁓ O Sosanya: Thank Thank much.

Michael Finocchiaro: at the CDFAM conference in Barcelona, which is pretty awesome. I have not had the pleasure of meeting Jay yet. Why don't you gentlemen introduce yourselves to everybody? Jay Wright: Sure, my name is Jay Wright.

I'm co-founder and CEO of Campfire. I've spent a lot of years in this immersive tech space and got a lot of learning. Looking forward to our discussion this morning. Michael Finocchiaro: Awesome, thank you.

O Sosanya: Yeah. Hey everyone. I'm Oluwase Sosanya. I go by Shay.

Co-founder and CEO of Gravity Sketch. I've spent about a decade in product development, physical product development, from furniture, consumer electronics, to car interiors before leaving and starting out Gravity Sketch, also about a decade ago. ⁓ I just find that the technology itself is not the thing that I'm focused on, but actually what it provides and enables the individuals to do is like... It's just like really groundbreaking and we've always been trying to help everyone understand the potential of this technology given its maturity curve ⁓ and how it can offset some of the complicated and unnecessary back and forths that happened in the product development process.

Michael Finocchiaro: I think there be a little bit of bandwidth over there. Well, it's such a pleasure to have you both. I ⁓ also excited because I have two people that are both experts and leaders in ⁓ this whole field of virtual reality, which a of people haven't even tried on Google Glasses or the Vision Pro. ⁓ Jay Wright: ⁓ don't say that.

Michael Finocchiaro: Well, they were expensive, right? I wore them once at the Google headquarters like 10 years ago. And I remember they burned in a hole in the side of my head because the battery was in the right, wrong place, you know? And there's Vision Pro.

mean, it's not everybody's got three grand to drop on a pair of dark glasses. ⁓ I was just, so I usually start this interview asking about, ⁓ to ask you guys to think back to November, 2022. You know, there's a before and an after to. Jay Wright: Yeah.

Yeah. That's for sure. Yeah. Michael Finocchiaro: chat GPT, right?

I mean, it changed the way we think about everything basically. ⁓ I'm wondering, when you first saw you're both software people, you saw that happen, were you guys initially bullish or skeptical in terms of how that was going to change ⁓ engineering? Did you think, okay, well, it's going to change the consumer side, but engineering is engineering, it's going to stay the same? Were you guys bullish or skeptical on how that was going to be transformative?

Jay Wright: I was pretty bullish. mean, LLMs were sort of not my first AI rodeo. ⁓ Building the Vuforia business at Qualcomm and then at PTC, I'd seen what AI, then we were calling it deep learning, had done for computer vision. It was just completely transformative and changed the algorithms and the way we were doing everything.

So I kind of looked at LLMs as the next wave of what was happening, this time with language. And I think for me, the thing I was most taken with is, OK, we got this natural language interface thing down now. That was the big takeaway, is we can actually talk to a computer with natural language. And was clear also that it was going to make it accessible to a lot more people.

So I think where my head went first was, wow, working with 3D and engineering, we got a lot of mouse and keyboard gymnastics we do to move 3D around. ⁓ If we can at least start by just bringing a natural language interface to 3D, we can make this stuff a lot easier to work with and accessible for more people, which is just really critical to what we're doing with Campfire. So I think when those first ⁓ LLMs came out, think ChatGPT had an API within a couple ⁓ of months. and we started running our first experiments.

So bullish, bullish then, more bullish now. Michael Finocchiaro: We'll get back to that. How about you, Shay? Were you also ⁓ super enthusiastic or maybe a little bit more reserved?

Jay Wright: Yeah. O Sosanya: Definitely curious, but also cautious. mean, you know, it's what are we actually dealing with? What's the kind of, ⁓ I guess what's driving it was the logic that's behind all these systems that are running.

And I don't come from that background, so I don't have a whole lot of predisposed information around it. ⁓ But also we've been promised this future of AI for a very long time. And I think that's where like the caution came from. It's like, okay, this is what, you know, the videos and the movies are telling us.

And so I, you know, I really was hoping to wait until, you know, we actually see something more substantial with what we have. And I think the thing that I wasn't really prepared for was the pace. You know, it was actually like so fast that we're seeing iterations and improvements. And also the whole world now is exposed to this, right?

Like, Michael Finocchiaro: Hmm, this is insane. Jay Wright: it. O Sosanya: My pension is exposed to this, whether I like it or not. And so I think there's just like this momentum that has come along with it that ⁓ has really driven a lot of awareness.

And I would say whether you like it or not, you're part of this journey. And so for me, like now I'm thinking a lot about how do we offset some of the tedious tasks, similar to what Jay said, to the LLMs, but also this world of 3D. It's a very unique world. And most of our customers don't have data floating around that can be trained on.

They actually don't even have their data accurately logged inside of their organization to train against. And this data that is generated in 3D right now, it is really depth. It's not berep information. It's not nervous-based information.

It's not mathematically driven. And so I think we still have ways before we can get to the point where the models are actually going to drive mathematically resolved models that can then be natively adjusted using you know, our current geometry kernel or, you know, a Simmons kernel or something of that sorts. So we're seeing a lot of adoption in our customers. It's happening a lot at the front end.

And what we're seeing is that there's a lot more optionality in the front end. And so taste makers are really important, you know, really deciding which models you want to move forward with and actually put into the more ⁓ formal 3D development process. Michael Finocchiaro: Because it's okay. O Sosanya: But I would say I'm cautiously bullish.

would say I'm cautiously bullish. And not because of the hype. I think that there's a long-term potential here. We just need a lot of people to start aligning on what are we going to feed this thing, how we can actually get some real meaningful information out of it for our engineering workflows.

Michael Finocchiaro: I guess maybe for the people that are not as AI savvy, maybe we to mention that AIs are a language model. They're dealing with language and 3D has its own formats, its own unique way of working. So that's sort of where the rub is for engineering. do you get that?

And particularly when you guys are using virtual reality, which is another level of 3D is not just on a screen. I need glasses or some kind of augmented device. Which makes it even harder to make AI because you've got to somehow, well, you've got to use other techniques than just a pure large language model because geometry today is still not code. It's not a language, right?

Jay Wright: It's not. And I think in general, when the first large language models came out, thought, it's going to be a while before these things are going to be useful in 3D because large language models were trained on language, right? Like not 3D. It's not like there's a bunch of CAD files out there in the corpus of training data that the labs are using.

But I'd say I've been extremely surprised what people are pulling off, Campfire included. without having true understanding of 3D, but using current models to manage essentially scene graphs. ⁓ There's a of power that can be derived. And you've seen a bunch of examples of this too.

I'd say my learning over the past six months is just sort of expected to go even faster than you thought. Capabilities I thought a year ago were gonna take two years happened in six months, and actually not the way I thought. ⁓ Michael Finocchiaro: Nice. Jay Wright: I love it.

so exciting. The tech hasn't been this exciting since I was 10 years old typing a basic program into a TRS-80 and seeing it go. It's just unbelievable. Michael Finocchiaro: All right.

O Sosanya: ⁓ I treated it akin to when ⁓ polygon modeling came out, like in full effect. And so it's another of ⁓ using ⁓ this technology maybe drive a result. ⁓ But you NURBS-based and you solid-based model much different you model. And I imagine there's going to a world where you are, you know...

discussing with the AI and getting some initial models or revising a model. I think that will be a slightly different way of operating and a different mental model. Like when I approach, I've been using CAD tools pretty much my whole professional career. When I approach a problem in Rhino, I approach it differently than I approach in SOLIDWORKS.

And I approach that differently than I approach in Alias. So, you know, having AI as a companion on all three of those skews, I would probably approach those things even differently at different stages. Potentially I wouldn't have to set up a scene anymore. I just get stuck into an existing template or something like that that I can generate through language.

So I think there's just a way of approaching it as another technology or another geometry type potentially in your tool set. Michael Finocchiaro: I wonder if implicit might be one of the ways we get there, right? Because that's already a mathematical expression of geometry. ⁓ I wanted to ask now, like, of course you guys are both developers and well, you've been in the software world.

AI is absolutely completely transformed the way we do software development, right? I mean, now it's just the insane velocity at all. I'm wondering how quickly did you guys add that to the stack of what your developers were using? O Sosanya: Thank Michael Finocchiaro: And how much has it changed?

it, probably you started maybe doing ⁓ the specification documents and moved on, but I'm just curious to know how you guys started adopting AI in the development process. You can start shit if you want. Go ahead. O Sosanya: Yeah, sure.

⁓ Straight away. think even before specification documents, our devs were at the forefront of this. so ⁓ we really quickly started to explore changing different things in the product really quickly to prototyping things, to figuring if we can kind bug fix or solve some really complicated parts of our ⁓ stroke algorithm. Things like this where you just...

Jay Wright: Yeah, go ahead. O Sosanya: You have someone to kind of ping ideas against that has an infinite amount of knowledge, or at least you could point it into like a researching area of deep knowledge. And now it's kind of spread across the whole, the whole company. Like we use it in every stage of the process.

There's still part of our code base that we do need to like architect so that it could work better with the LLMs. I think there's like that there's like this legacy code that we didn't build with the mindset that these LLMs will come into play and we'd be able to kind of use these as like a companion. in the process. we're working on that at the moment.

⁓ But also it's starting to kind of bleed into what we present to customers, right? Like we're now getting a much more rich understanding of some of the domain challenges for our customers and building out our presentations as well as how we might structure some of the settings in the product to better meet a company that's making excavators versus a company that's designing cars, right? Like there's, can kind of get to the granularity without the cost. the human cost that we had in the past.

Michael Finocchiaro: of a future. Jay Wright: ⁓ we're pretty, ⁓ we're pretty AI-pilled right now. ⁓ it was definitely, think, devs, devs jumped in real quick with the tools that were there. ⁓ we were, mean, experimenting just with it in product at the get-go, but I mean, we're, we're now at the point where the entire process is, is all automated and controlled with AI.

I I can actually get in now instead of reporting a bug to others. I can jump in. root cause, create an issue, fix it, and PR, right? And boom, I mean, so the entire process just getting compressed is, it's just unbelievable.

And I think the rate of progress has changed pretty dramatically. You know, it's moving so fast that you have to kind of spend dedicated time just to sort of share learnings within the team, make sure everyone's getting along, because everyone's got some new trick that they've figured out. ⁓ Michael Finocchiaro: Hmm. Jay Wright: But it's been a lot of fun.

It's just been a lot of fun, and seeing the results has been tremendous. Michael Finocchiaro: So you guys inviting your agents to the scrum meetings now? Jay Wright: ⁓ 100%. O Sosanya: meetings.

All the meetings. Michael Finocchiaro: How are you guys putting in guardrails though so it doesn't start putting pink elephants in the middle of your scene or whatever? O Sosanya: I mean, we have a review process. We have a review process.

All the engineers do kind of like a round robin when everyone reviews each other's code. have a QA team that's really thorough and comes through everything in the product. ⁓ It's not like we just kind of set it loose. I think that day will come, but I think it needs enough of like working with us to understand, you know, where it can push the boundaries and where it can't.

So I think we're again, we're in the very early stages in my mind of like what this technology will mean for the development process. And with things like Codec, I think they are also learning as they're developing the next tool sets and the next models. so I think anyone that's just letting it loose on the code base right now will most likely need to spend a bit of time doing some hygiene in the future. So for an early stage startup who doesn't have the legacy that we have, it's probably amazing.

It's just you could just crank out product left, right, and center. But I think over time, you might want to put a little bit of structure because maintenance is a really big thing. If you don't have enough of context of what needs to be maintained, what tends to break, what the user behaviors are, I think there could be like a little bit of ⁓ a kneel in a haystack kind of approach. So yeah, guess the way that we can maintain it right now is by just being a bit human in terms of the curation process of what actually goes to production.

Michael Finocchiaro: And Jay has ⁓ changed the way you look at organize the developers in terms of agile versus waterfall. Because those were the two ways we used to do it. Is there a new model emerging with the agentic stuff in the middle? Jay Wright: ⁓ I think Waterfall has gone the way of the dodo.

⁓ Everything's agile. Certainly roles have changed. Certainly there's fewer folks that are just sort of building software, right? We're converging more and more to the same people that can define the requirements, can make the changes in code.

And then having the just architecture and governance being in the hands of a few to make sure that, you know, the There's no pink elephants showing up. But it's super important. I've been doing software for a long time. And definitely bad things can happen.

And it's incumbent, I think, on everybody using this technology to make sure that they've got safeguards in place. Michael Finocchiaro: There is a question in the chat. My friend James, hey James, is asking, many designs start with existing assets, current design, bombs, legacy bombs, supply chain, sourced content, et cetera. How will AI be able to leverage existing diverse content and make new stuff?

O Sosanya: So yeah, existing 3D content. Because it already references existing 2D in language, right? Yeah, think right, so there is approach right now, exploring and playing with, which is just putting a bounding box around whatever geometry is there, and you get six images out. could then, the AI could then read that and say, okay, like, this is a vehicle.

This is a vehicle. This is a, you know, a vehicle with a stance of this X, Y, or Z, and then. Michael Finocchiaro: Yeah. Alright.

O Sosanya: We can crank out a couple of different constructions of that. So I think for iteration, like quick iteration, there's like already well-established tools for that. We just feed in what you currently have in the scene. ⁓ In terms of that being exactly the same topology and the same geometry set that, and even the same scale, to be honest, is like that's, there's another hurdle into that domain.

you know, we're still trying to figure out how to isolate things that you want to change. Like maybe you don't want to change whole thing. Maybe you just want to change the face or you want different wing mirrors. Like there's a lot of different things that we hope, but I think hope right now is not a matter of years.

It's a matter of months. So, ⁓ just being ready for that kind of really quick iteration cycle. Once models can help distinguish parts. And I already see a lot of that happening, but like parts or, or, ⁓ maybe more of the anatomy of the product that you're actually designing and developing.

I think then we'll see like an explosion of. how these things could just feed in tons of information and get more context. Because right now, if you prompt something and you say, want to get an SUV for seven seater that has the similarities of a Range Rover, ⁓ it's kind of going to give you something okay, but it's not going to give you the right proportion and stance. And that's tough, but I think that will come.

Michael Finocchiaro: And how about for you, Jake, as you're dealing more with collaboration than design, right? Jay Wright: Yeah, we're dealing with a wide range of what I describe as spatial workflows. So some of those are going to be in engineering design, some of those are training, some of those are sales. And generally those workflows depend on all different types of assets.

For example, if I'm trying to create guided instructions for an assembly procedure, I'm going to have some 3D assets that define where I need to move things, maybe some animations. I might have some documents or specifications that have the rules by which this is done. I might have a reference manual. And I think the amazing thing in immersive land when you're spatial is it brings all that context together.

I've got 3D context, 2D context, and even some context of the world around me, depending on the device I'm using. So now in Campfire, you're using AI to bring in your 3D assets. And we treat different types of 3D assets just like you treat different types of images in PowerPoint. You can just bring them in, move them around just by talking essentially to the AI.

And then you can also bring in external documents. So if you want, for example, your trainer, your virtual agent, we call it a spatial agent in Campfire, if you want it to be able to take questions from people going through the training, it'll do it using the documents you provide. So it's just tremendously powerful in being able to use 2D and 3D context, bring all that data together. and an experience and a workflow that's just incredibly easy for people to use.

You don't have to be an engineer to use this thing or even build this thing. It's incredibly simple. Michael Finocchiaro: Awesome. Just wanted to change gears.

like, my next question is always about where does AI sit in the stack of your software? Does the user encounter it directly in the UI because there's a chat bot or something? Is it just in the DNA or using foundational models that are AI based? I'm just wondering how AI native the applications are.

Jay Wright: Yeah, I can kick off on this one. For us, AI is a first-class user. And in fact, AI presents itself just like another human user does in Campfire. It actually appears as if another remote user were in there with you.

So we call it a spatial agent. You might talk about it as an embodied agent, too. It's like another kind of head that's floating around. It's aware where you are.

It's aware of where the models are. understands all the tools that are in Campfire and can use them. But when it comes to our stack and our development process, that means now every feature that we implement, we think about exposing on multiple surfaces. On one, that new feature has got to be exposed through the UI for a human user, but it also has to be exposed via tools for our own agent as well as third party agents.

But the reality is that means that the AI, and I'll be more specific and say the harness, the harness is woven into every aspect of the stack all the way through. So, ⁓ and in fact, I'd say most of what you spend time on now with in the world of AI and agents is sort of development around that harness ⁓ and less around kind of the 3D stuff. The harness is where the action is. O Sosanya: Yeah, in a space.

Yeah, in a spatial environment, the interface is so critical, I can't stress this enough. And so there's, there's things that are inherently physical, because we are almost like teleporting our physical self into a digital space. And there's like AI, which is our AR, which is overlaying. And so I think there's things that natively feel intuitive to the human, and we want to always make those a priority.

Michael Finocchiaro: How about you, share? O Sosanya: And then there's things that don't feel intuitive because it's more of a computational thing and you're sitting down in front of the computer with a keyboard or something like this. so we try to offset those things to kind of natural language. And so there's a kind of a delicate dance.

And we find that a lot of our customers, our workflows are really deep into, we need to deliver this product in a certain timeframe. We need this type of buy-in. so there sometimes are human feedback, which needs to be driven primarily through the human. And then there's just really about capturing.

that feedback and then turning into bullet points of executables. And so there's that element that overlays as well. So for us, we're still kind of finding that right balance, but we prioritize human first interaction. And then AI is supplemental to help enhance the communication.

Just like we use AI now for all of our meetings, we use AI to summarize things. This is really the way that we're thinking about it in its current form. As we move forward into the future, we think iterations, we think simulations. There's a lot of white space there still to explore.

yeah, dealing with geometry, dealing with a physical product that needs to reach the market, you're always dealing with a human being. And we want to make sure that the human being is represented and the AI is helping the human being rather than ⁓ offsetting the need for you to validate something visually or physically, if that makes sense. Michael Finocchiaro: It really does. Are you guys actually using your own foundational models at all?

And I didn't ask before, I think, you built your own graphics kernel. You're not ⁓ dependent on somebody else's kernel, I believe. O Sosanya: geometry engine in our own kernel. Yeah, I mean, we're using just a lot of the off-the-shelf stuff, if I'm honest, for like 3D gen.

It's the Trellis and Huanyuan, all the different kind of models that are readily available to anyone. And then on all the natural language stuff, again, off-the-shelf stuff. But this stuff is quite smart. You can kind of get a feedback loop.

You could pre-prompt it with stuff that you really want it to take into consideration so it behaves in the way that a GravitySketch user needs to experience it. Michael Finocchiaro: Nice. How about for you, Jay? Jay Wright: Yeah, we're not we're not training anything.

⁓ We're based on foundation models that are out there in and use today. I'd say in general now when you talk about AI, it's maybe not just one model, right? It's an entire stack or harness of agents and tools that are using different models at different places. So depending on what you're doing, there's different stuff going on under the hood.

But I think that's where the magic is, is sort of. ⁓ Composing them appropriately and creating AI that gives accurate results. Before you talked about sort of the pink elephants and making those things show up. ⁓ In engineering, there's no room for hallucination.

⁓ We like to take pride in the fact that we make AI that knows how to say, don't know. And that's critical. It's really important. yeah, use the models that are there, compose them appropriately.

Give grounded results. Michael Finocchiaro: ⁓ That makes you think of another question, which is the cost of AI. I use cloud code all the time and I'm always a bit shocked at the end of the month at the anthropic bill. There's also the fear, particularly when you're developing software of cloud training and stuff on your data without actually telling you.

How are you guys doing your users? Can they bring their own ⁓ LLM and their own keys so that they don't... Jay Wright: Yeah. Michael Finocchiaro: Maybe ⁓ spend yours or how are you gating the data so that the IP isn't leaking out to cloud or open AI?

Does that make sense as a question? Jay Wright: Absolutely. Yeah, I think in our case, ⁓ there is an option where the customer can bring their own. So we have two different deployment models for the cloud, one that we call like a managed cloud, where a customer's data lives in a multi-tenant architecture.

And then there's another hybrid cloud, where the customer's data is managed within their own sort cloud account. And in that hybrid cloud architecture, the customer brings their own. So they are getting their own bill. ⁓ for AI but they're using the same campfire on top of it.

O Sosanya: Yeah, on our side, ⁓ we have just one kind of model for our customers. And so we go through the extensive security checks and so forth so we can get into these big OEMs. You know, it's really, it was really a lot of work in the very beginning, like 2018 to kind of get into the doors of Ford and GM and the likes. And so we've kind of gone through all that regulation and certification.

And if they want to bring in their own stuff, we are just making, ⁓ we're marrying like what you're currently using and paying for with our software. So it can almost be used in conjunction with one another. That way we don't have to have any exposure to some of those things. So if you've already signed up for service X or Y, ⁓ you can use that in tandem with gravity sketch.

and you're buying into the security of that particular provider, knowing that we have our own security layer as well. Michael Finocchiaro: so ⁓ Jay, you are super enthusiastic ⁓ and bullish. ⁓ Shay, you were ⁓ also enthusiastic, but a little bit more cautious. ⁓ We're four years into this AI revolution, right?

⁓ So one part of the question is, has your opinion changed? think Jay, you've already said you're actually more bullish than you were before, ⁓ which is fine. So the question is like, what's next? And particularly, have we seen or, well, the answer is obviously no.

If there was an open AI moment in engineering, we'd all know, right? Because there would be a for and after. And I don't think we're there yet. I think that...

When people see what you guys are doing with Campfire and Gravity Sketch, they can actually start looking at what the future looks like. But, you know, it's not super wide adoption. So we're still kind of on the verge. So do you guys think like the open AI moment for engineers is months away, years away, or maybe we'll never have one?

What do you guys think about the future of this whole AI thing? Who wants to go first? O Sosanya: Yeah, I can take it. ⁓ With engineering, at least with us, physical products ⁓ primarily are products being used to develop and design.

And it's not as fast as hitting a publish button and it all going to every single user on your platform. ⁓ You have a lot of different considerations. It's materiality, really. So you're going to make something out of a sheet metal.

⁓ You're to have ejection molded components. And so what we're seeing, I think, in this kind of AI moment is less of an AI moment and more of a gradual rollout across different stages of the workflow. We've obviously seen it in design. It's rolled out quite aggressively.

But now how do we take that AI generated content and push that through the pipeline and use AI where we're needed, whether it be simulation or versions or whatever the case may be, to the final product that reaches the shelves? And I can say with confidence, we've seen a couple of footwear designs. come from the generated AI asset through the process to the shelves of the foot lockers, JD Sports of the world. ⁓ Where AI had an influence on each one of those stages is still a little bit up in the air, definitely in the front end.

It was a huge help, but how are the injection tooling manufacturers, or the injection tooling modelers using AI? I'm not sure yet. I don't know if there's enough data that's been trained upon that. So I don't think there's gonna be a single moment.

I think it's a series of shifts that lead to a more robust workflow. And what we probably will be seeing and thinking about is time to market for product development will collapse quite, quite aggressively. And so we'll see that products are hitting the shelves a little bit faster, more well-resolved, potentially customization and variety. So that shift I think is going to be this kind of ripple effect as AI has this kind of injection points at every stage of the process.

But you could think about. bring a physical product to market, takes a lot of people. It's not just a single design team that's able to publish and push on their own autonomously. It really does take a lot of different moving components.

And so each one of those components leveraging it in their own way to make a more efficient pipeline. Michael Finocchiaro: I like the point you bring up about the manufacturing because that's one of the major issues I'm seeing is just this gap between engineering and manufacturing. I did a couple of podcasts recently about that. ⁓ When I was in the PLM components conference up in Cambridge last week, ⁓ Materialize was talking and that was interesting because he was, ⁓ at least they're starting to fusion additive subtractive manufacturing together with the molding.

So, you know, when you're doing additive, you're always going to have too much material. So rather than go into another tool, the same tool can say, okay, I need to come in with a Miller and take this piece out and do a chamfer over here. So at least some of that starts coming together, but there's still a big gap between engineering and manufacturing. I guess, Jay, maybe Campfire is there to help bridge that gap, right?

In terms of collaboration. ⁓ Jay Wright: Yeah, we definitely bridge that gap. think generally when products are done, they get handed over to manufacturing to try and figure out if they can build the thing. ⁓ And being to validate the build process is pretty tough.

So ⁓ if you're able to take both the cap models from your products and the cap models from your manufacturing cells, throw them all together ⁓ and simulate workflow, you are going to save yourself millions and millions of dollars and months and months of. of the headache. Maybe I just wanted to go back to your previous question because I think we might be closer than you think on just the open AI moment for engineering. I mean, to me, the open AI or the chat GPT moment wasn't when it started replacing all of our workflows and was totally deterministic and did everything well.

It was kind of the holy shit moment of like, ⁓ my God, this changes the game. This is all about the change. And if you define it with that metric, I think I think we're pretty close in that we're starting to see mainstream tools, and I'll just point to, I think, Blender, Fusion 360, get like ⁓ MCP and like clock connectors. And so I think once that happens, we've all seen the ton of videos on LinkedIn now of what people can do once they start connecting those tools.

But I think that's the analogous moment, the one where we're using AI to just control the tools we have. Is that replacing the modeling or the engineering process with AI and expect it to be deterministic? No, no, it's just, it's really just about, think using that natural language interface to be able to do things faster and automate things that are happening today. Michael Finocchiaro: ⁓ I wanted to...

⁓ O Sosanya: ⁓ with Jay on that point because ⁓ I saw the connector thing, I was kind of blown away as well. started playing with it. was like, it's almost there. It's almost the point.

I think ⁓ a whole, because I ⁓ so many of my years in manufacturing, ⁓ a whole, still have this longing ⁓ desire to have the expertise of an injection molding engineer, like the guy that's making the tool. I want that in the AI as well, if that makes sense. I want to be able to Jay Wright: ⁓ O Sosanya: confidently know that this thing could be injection molded and like it could kick out a tool that has all the moving components so I can have this, you know, injection molded ABS without the tooling overhead and costs that I would normally incur.

I guess my my holy shit moment is probably still at the physical. I want to hold the thing in my hand knowing that I was able to kind of like compress that timeline and the dependency I have on the tooling engineer and not trying to put them out of a job. I think they're still valuable because they do like Michael Finocchiaro: Nice. O Sosanya: 12 tools at once rather than focusing on doing one injection tool at a time, I think.

Michael Finocchiaro: Well, okay, so that's another question I didn't want to forget is a lot of my audience is a pretty big segment of rather young engineers that are probably having a bit of anxiety over the say I think like, my God, it will I actually have a job. So from your perspectives as, as leaders and people that are hiring engineers and developers, what kinds of things do the, does this younger generation of graduates needs to focus on where they're going to be essential to the process and not replaceable?

Jay Wright: ⁓ I'll go ahead and start. think the best thing you can do is just fool around and experiment with everything you get your hands on as fast as you can. ⁓ hit YouTube, watch videos, download tools, experiment, the people that are going to move the needle the fastest are the ones that understand it the sonest. It's also very, very clear to me that the gap between high performers and low performers is just going to change dramatically with AI because the productivity gains from people that understand how to use it and apply it are just orders of magnitude ⁓ greater than.

than those that don't. just jump in with both feet. I don't know that it really matters where you start. Just jump in and play.

Learn. Michael Finocchiaro: Is that the same thing you'd say, Shay, or would you have a different twist on it? O Sosanya: I don't know. I'd echo that exactly.

I take it a little bit further as well. You know, a little known fact, I had a creative agency many years ago after leaving university. I did a few build outs for a couple of office buildings, ⁓ but it was just, it was so much work. was just a two man group, right?

And the process of actually making the interior, like we actually, we designed, manufactured, designed, presented and manufactured to our clients. And The work that did back then that would take me months can now take me a couple afternoons, right? Like after work, I could even fool around with this. The making process and the source and the material, the delivery of the material from one vendor to the next is another big component.

And so I feel... anyone that's like stepping into the space that has a bit of fear. think there's, there's so many moving parts. How do you extend yourself beyond being a single cog?

And I think that's like, that's what I learned about having my own creative agency is like, I had to be all these different cogs moving at different gear at different speeds. Right. And so if you're just breaking out of university and you're wanting to break into this space, I would say make stuff, use the tools to develop everything, you know, get the presentation, right. Get the deck, do some simulation.

and then try to get that thing physically made. Even if it is just a small gearbox assembly or something like that, you have 3D printing now, you have online resources where I never, I didn't have this stuff back in the day. I can just upload a file and I can have something CNC milled for me and shipped to my house, to my specifications, right? So that all builds a better understanding and a cohesive knowledge of how materiality works, of how assembly works.

And so eventually you're going to be talking to these tools at a much higher clock speed than Hey, I want to make this and I want to represent it in this way. It's like, Hey, I know that this material is going to be used. It's going to be, you know, a 720 or 660 aluminum. It's going to be, you know, these types of tools I want to be using against it, like this quarter quarter inch mill or whatever it may be.

And so you're actually having like a much more sophisticated conversation. So your results and your throughputs are going to be better. And so if you are joining an organization, they're bringing on someone that knows how to have a very deep conversation about the beginning and the delivery phase of a product. So I think it's almost like, I've been playing with this idea.

So this is the first time I'm talking about it out loud, but it's almost like going back in time where we would go to a cobbler and they would know exactly how to design a shoe for your foot because they were making shoes for your family the whole time, or they know exactly how to create the saddle for your horse. I think. Now the designer, the engineer can almost have like a very expansive understanding of the whole process for specific use cases. And that's how you become dangerous because you can then speak at a different level and level of customization.

I think that's going to come in is actually pretty cool. So pretty soon we're going to be able to just have custom product for individual. Michael Finocchiaro: Very cool. Yeah.

I think I think I was seeing some of that at the Nike demo at CD fan, wasn't it? That was pretty awesome. And by the way, ⁓ I did summaries of all the CD fan presentations. If that's a conference in Barcelona, just having a couple of weeks ago for those who hadn't heard about it.

⁓ Shays presentation I reviewed as well. So feel free to look on my LinkedIn for that. Also, AWS is our sponsor. So please click on the link and download the white paper for them.

⁓ So then the last section of this interview, I like to talk about digital transformation. I, I think that, know, AI is moving a lot faster than the fortune 500 is right. And AI adoption is well complicated by a lot of things. But I, when I think of digital maturity, cause obviously it has to reach a level of digital maturity before they can actually use AI, right?

Because you have all that, the legacy cost of having data to train on. Well, when I think of digital material, I think of a scale of one to five. Like one is basically using Excel and, you know, email to do collaboration, which probably makes Jay's skin crawl because he's got such a great collaboration system. It's so much better than email.

And then you have the companies that could be at level five, which would be like autonomous agentic digital twins. And I don't think anybody is at five. So first of all, first of all, question is your customers that you're selling a gravity sketch and. a campfire too.

Do you see them between one and two closer to one closer to three? Where do they sit today in 2026? Jay Wright: I could start. ⁓ We're dealing with mostly kind of industrial manufacturing, so automotive aerospace, industrial equipment.

as you know, this one skews pretty low. ⁓ Maybe a two-ish for most of what I see. ⁓ I would say that I probably see more customers that want to make sure we're not using any AI. ⁓ then I see customers that are at like five where they got the fully agentic digital twin.

I actually think there's probably not a lot of customers at five. Five might just be vendor territory. ⁓ But yes, it's generally low. And I think it's just because we're in a conservative industry that moves very, very slow and is regulated and it's just very, very cautious in their approach.

Michael Finocchiaro: It's not also due to the cost of, like you guys are both in VR, so there's a certain cost associated with the headsets and stuff. Is that also a bit of a drawback or that's not really a concern? Jay Wright: No, it's huge. think that's a, I would describe that maybe as a different axis of maturity.

You know, there's, digital maturity around, let's just call it like digital thread and agentic workflow. And, and I think that's really in like the thrill seeker, early adopter kind of space right now. Like they're kind of, you know, innovation folks are trying to understand what it is and what it means. But the, the bigger concern is, is really just, ⁓ InfoSec processes and the ability to adopt new technologies.

So. Bringing in something like an immersive device or a headset is a new headache for a lot of people. And I think that takes time before you can get that to be standard issue in a company with tens of thousands of people that they can just order and have supported by IT. And as crazy as it sounds, like the cloud, just being able to put proprietary data and store it in the cloud, there's still a lot of people that really want to demand that that's going to be on-prem, whatever the cloud solution is.

Michael Finocchiaro: Hmm. Jay Wright: You know, those hurdles, those hurdles are made. Michael Finocchiaro: Thanks, man. ⁓ What about you, Shay?

What's your take on it? O Sosanya: Jay covered it pretty well. I mean, like a lot of customers are still trying to fight with trying to get stuff on prem. And, know, I'll break it down into two different ways actually.

So the customer as a entity versus the individuals within the accounts. So the customer as an entity, they ha I think they have to be conservative. I mean, you've had, you know, the Firestone incident with the tires and so forth. Like you were putting human beings into these products and they have to protect the human being in the event of a fault.

And if the fault is introduced by the manufacturer, that could steal your company, right? So I think they are right in having this a conservative approach. ⁓ Headset cost is negligible if you think about it. A car from start to finish is $1.

5 billion. Like this is actually a cheap thing. But to get IT to say, yes, we're going to roll this out, as Jay said, you need to work with a supplier. And we don't really have a supplier in our industry, like a Dell or and HP that rolls out fleets and refreshes of hardware that include any kind of spatial device.

It just doesn't exist, right? So you're going through different vendors that are able to supply at that level and include the warranty and so forth. So I think that's another part of the industry that we need to continue to work together and try to unlock. But if it comes to the individual in the account, they're already using this stuff.

And they're using it against, I won't name and shame, but they're using it against the company's policies, right? So I walk in, people are already having headsets, they bought them, brought them on board. They're already using a slew of different AI tools on their personal computer and then just copying the text over or, you know, copying the imagery over. like the human being seeing the value in this technology in these organizations that move extremely slow is already there.

So I always try to encourage our customers, like at least get a task team. out in front of this stuff so you guys can actually start to see what the minimum viable integration is because the teams are already adopting it and they're adopting it against company policy purely because the pressures on them to deliver in the timeframe with the accuracy, it only increases. Like you didn't increase the timeline, you just increased the number of projects. You're not hiring more people, you're just throwing more projects at the same group of individuals.

And so of course it's human nature to kind of look at what is the best tool for the job and if my company doesn't have it. I'm gonna figure it out and bring it in house. And in our industry, I wonder if this kind of also parallels with Jay, ⁓ people are choosing to go into these creative disciplines. And I include engineering as a creative discipline.

This is a lifestyle for them. Like a lot of these folks at home are tinkering with stuff or designing stuff. They're continuing to creatively output anything. Maybe they're renovating parts of their home.

And so for them, finding the best tools for the job and the most exciting tools and the things that really help them stay creative and get locked into that flow, ⁓ that's just part of their lifestyle. So these organizations, think, ⁓ the technology is moving faster than they control it, but I think trying to find a halfway house where they can introduce things, maybe ring fence a couple of other things, but giving people a little bit of potential to explore beyond just innovation teams would be very valuable.

Michael Finocchiaro: That's a great answer, thanks a lot, man. Appreciate that. So my thesis is that ⁓ the big three, the ones that really dominate our industry, right? Das, Oseman, PTCR.

Well, they're really not ⁓ caught up to where AI is today, right? They're sort of in catch-up mode. You guys are on the vanguard of this. You guys are a lot more agile, you have a lot less technical debt, ⁓ and you're able to go at almost a velocity of ⁓ of all these tools.

⁓ It would seem to me that customers that wanted to move the needle from one to two or two to three would tend to do that better with an agile startup like a campfire or gravity sketch. I'm wondering, have you actually seen that when the customers start using your tools that are AI informed and AI native, is there a bit of an epiphany from people outside the project? They're like, ⁓ my God, if I had data governance right and I had connected my data silos and I got the departments of talking to each other rather than saying, those bastards, they don't know what they're doing.

Is there sort of an epiphany where like, ⁓ my God, I can actually achieve this. I can actually move the needle and become digitally more mature. ⁓ And that epiphany comes thanks to using your software. Jay Wright: I don't think any of our customers are using our software because they think our AI is going to be implemented better or faster than incumbent.

I think generally they're using it because they can get a job done, but they can't get done with that other software. We're not trying to do the same things or duplicate the same workflows. think generally, The wow moment for our customers versus incumbents is the fact that our software probably feels more like a video game than it does engineering software. And it feels more like just modern collaboration tools in sort of the 2D domain ⁓ than anything collaborative they've used before from incumbents.

And then generally, while everybody's trying to figure out what AI is for, I think I'm always cautious when people are looking at AI just for AI's sake. If there's not a specific problem they're solving. But for us, AI is about being able to deliver more workflows and make them easier to use full stop. And if it does that, great.

And at the end of the day, that's what resonates with the customer. Michael Finocchiaro: I agree, but I've had a couple of customers reach out to me and say, well, if know, management is saying we've got to do more AI and, and, ⁓ you know, the, the portfolio from the big three doesn't have that in there. ⁓ can you help me? Like which, sort of should I use?

So I get a little bit of that AI stuff and I can prove that the stuff works. So that, that's where that question was coming from. Jay Wright: Yeah, yeah. Yeah, I mean, look, when I get that, still, I do love that, but I want to get them anchored on the use case as soon as we can.

Just AI for AI's sake, right? So 100%, here's something that you can start, and it is AI today, and I don't want to leave that behind, because I see that too, but at the end of the day, if they're not looking for the use case and they're just looking for AI, then we're going to this one. Michael Finocchiaro: Great. O Sosanya: Yeah, sure.

Our company mission is to help people bring better products to life. And we're leveraging these technology tools to do that. ⁓ Michael Finocchiaro: So Shade, you wanna jump in? O Sosanya: I would wager that it's not too far off from some of the big guys in terms of their mission.

But you're right, we have this flexibility and we don't have this baggage and this legacy that the others have. And so I think when customers are seeking us, it's not so they can deploy the technology faster. It's more of like, okay, here's a problem that probably has persisted in some of these legacy technologies for like legacy CAD tools for a very long time or 3D environments for a very long time. You guys have seen to open up the hatch on that.

And whether it's just being to Jay's point, like just being spatial, being in that environment opens the whole door to like anyone that wants to get into 3D because you don't have to worry about navigating with a mouse, right? So I think it's those kinds of things that are bringing people closer to us. And it's our ability as a small startup to really pick a couple of workflows and work aggressively against, like we're not trying to boil the ocean. If you think about, you know, a DSO system.

I imagine most of the products on the shelves and stores have gone through one of their products, right? And that is a wide range of use cases. mean, holy cow. Like that's so many different use cases that they have to take into consideration and they have to work against it.

They to build a product that can service all of those. Whereas we are narrowing down into one part of the tool chain and we're really going to make that excellent. And, you know, over time we have big ambitions to go beyond that, but I think that's why we get some attention from customers. Now I won't, you know, mute the point around AI.

think a lot of these customers are getting AI budgets to go, even we've given a budget to our team, right? Like, Go and explore some these tools so you can get more efficient. I too, like Jay, I don't want a customer to come to us to say like, we have this AI budget, let's use it for Gravity Sketch, show us what you can do with AI. I think that can lead to a churned account.

I think we want to drive value, we don't want to drive novelty. And so it is hard to compete against those budgets because you do have a few players in the space that are just creating really eye candy type of AI experiences. ⁓ you know, they'll have to kind of true up to the value and the workflows at some point. But for us, I'd rather us grow with more confidence that this is going to be used in perpetuity than try to capture the wave right now, if that makes sense.

I'd rather capture the wave as part of workflow delivery rather than capture the wave as part of the hype cycle. Michael Finocchiaro: Awesome. Well, those are great answers. Thanks, guys.

I just wanted to close just before we say goodbye to everybody on the audience. Where can we see you guys? I ⁓ I got to see you, ⁓ in Barcelona, but where can people find you guys in trade shows before the summer or during the summer or in the early fall? If they want to come meet you and use Gravity Sketch and Campfire.

Jay Wright: so ⁓ October, mid-October Augmented Enterprise Summit, ⁓ probably the ⁓ best place to be the intersection of ⁓ enterprise engineering and immersive. Be there. Michael Finocchiaro: Nice. Awesome.

So mid October you said? Jay Wright: Yeah, I think it's October 13th through 15th in LA. Michael Finocchiaro: In Atlanta, nice. How about you, O Sosanya: ⁓ Yeah, for us, we come to you.

I was just at a customer onsite actually today. So if you are in Europe and the UK, we love to just kind of come onsite. What we find is like just walking through the halls, walking through projects together with clients is really helpful to give us context to like how we can actually help them. So when we come onsite, we usually come with someone from product, someone from our business development team, and then we have design consultants in house.

And so we're just always trying to evaluate the real challenges and see if we can match make that with the technology. We've done a few road shows. We'll probably do another one in June for, for June, for Munich area. So happy to kind of stop by any of you all out there in Munich, ⁓ who are working in the industrial design field and, kind of chat through that.

You can always ping us anytime directly through the site, ⁓ directly through LinkedIn for me. And we've done a lot of virtual demos. do virtual demos, probably like, ⁓ Michael Finocchiaro: Nice. O Sosanya: at least a dozen a month or so.

you know, we are, we're, constantly kind of hopping on. We have a webinar coming up relatively soon. Sorry. The date just slips me.

It's this next week, I think early this week. It's a, do webinars. Yeah, we do webinars very regularly. So catch one of those.

⁓ and then in terms of like the next thing, time that we'll be actually speaking at an event, it may be, we'll be sneaker week in August and in Portland, Oregon. ⁓ that's probably like publicly where we'll be next, but, ⁓ Michael Finocchiaro: You put in the chat, just. Nice. O Sosanya: Sometimes I just take opportunities as they come.

If there's something really interesting that we want to show up to, we'll rock up. Michael Finocchiaro: Well, you want to come to threaded Frankfurt when I announced that, O Sosanya: I mean, if you want to specifically check out what we're doing, just ping us and we'll make time for you. I think it's such an interesting space and it's really hard to kind of get a lot of enthusiastic people in an organization. So if you gather a few of your folks in your organization that are really enthusiastic, we'll give you the time for sure, absolutely.

Michael Finocchiaro: Awesome. Well, that's been great. I think I learned quite a lot and you guys gave some really original and excellent answers. I appreciate your time.

And once again, we're sponsored by Amazon Web Services Marketplace. There's a white paper, if you don't mind clicking on the download and checking that out about HNTK AI. Thanks once again to Shay and to Jay, Shay and Jay. That was awesome.

Thanks to the audience. We'll be back in a couple. in about two weeks, I'll ⁓ some AEC vendors talking about the... Also people that develop their own graphics kernel like you did, Koniq has their own graphics kernel as well.

⁓ So we'll be talking about that and ⁓ trying to get some machining ones and some more factory ⁓ MES kinds of ones as well. So ⁓ once again, thanks to my guests and thanks to the audience and we'll catch you in the next time on the AI Across the Product Bicycle. Jay Wright: Awesome. Thank you.

Michael Finocchiaro: Thank you.