AI for Engineering Is Leaving the Demo Phase

Fino: And we're live. This is ⁓ Michael Finacero of the ⁓ AI Across the Product Lifecycle Podcast. ⁓ I'm joined by my friends Pradu and Martin, who we've met actually in person a couple of times now over conferences over the past year, which is a lot of fun. so Pradu, why don't you introduce yourself and and build a little bit?

Pradyut: Yeah, so my name's Purdue. I'm one of the two co-founders of Build. ⁓ Build is a CAD data management platform. We're essentially connecting all things engineering into manufacturing.

Our focus is really around data management and automating the workloads built on top of that that data. Fino: Awesome. ⁓ and Martin Dielecki of Bench. Do you wanna explain what what Bench is doing?

Martin Bielicki: Absolutely. Great to be here. ⁓ my name is Martin. I'm the CEO and one of the co-founders here at Bench.

And at Bench we're building ⁓ an orchestration layer for engineering. So essentially we're looking to apply AI to ⁓ automating AI end-to-end workflows in engineering, so spanning CAD, simulations, ⁓ PLM and the like. ⁓ and yeah, starting ⁓ starting with with CAD and simulations mainly. Fino: Awesome.

so I usually like to open the podcast with a question, like looking back a little bit, like you know, the open AI moment that we all had in November twenty twenty two, and we were like, ⁓ my god, AI I can do that now. ⁓ I I've talked to ⁓ you know over a hundred and sixty startups the last ⁓ year and a half. ⁓ do you guys did you guys feel ⁓ really bullish about AI ⁓ at that moment? When you first saw OpenAI and Chat GPT, or were you maybe you'd already worked on the earlier ones, ChatGPT two?

I mean, where where did you guys say? Were you super bullish or were you a bit skeptical? Martin Bielicki: So for us, we basically started a company because of it. So we we didn't exist beforehand.

So so yeah, we didn't exist beforehand. So so in our case it was more so seeing what I can do for ⁓ in digital coding. It was very early results, but still ⁓ we could definitely see where it's going. And we knew that it will transform essentially any industry out there.

⁓ engineering is the one we obviously had expertise on, so we wanted to to facilitate that change and wanted to bring that that into reality. Fino: ⁓ so so bullish. Awesome. How about you, Predict?

Pradyut: Yeah, we we're the we're the we're the opposite. So we started our company out in twenty twenty one, but like personally I had an early access to Dolly. So this was even before Chat GPT had Dolly and you could like prompt it and it would make these like images and it was really cool. I mean like I kind of knew like wow, this is it was the first real time where you could prompt something and it would create, you know, something brand new that was ⁓ even before Chat GPT happened and it was not text based, very visual.

Fino: ⁓ wow. Okay. Pradyut: ⁓ so I I kind of knew that this it was, you know, it would really kind of change the world. Chat GPT was the first application of it, obviously.

And since then we've seen such an amazing proliferation of like AI and these different use cases. I mean, now it's it's everywhere, right? From like our own tech stack to some of the services and products that we're offering to our customers. ⁓ and then even just like helping accelerate ⁓ our go to market.

So n forget the engineering side, right? Like we're using it across the entire organization. And I think that any company that is not leveraging AI in these different facets is really just going to be like ten times slower than you'll you'll honestly not win. Fino: Mm.

⁓ so in terms of ⁓ I mean we just talked on the origin of two companies at roughly a a year apart, but how has ⁓ AI changed the way you guys code and the way you guys manage developers and manage the development process? ⁓ I suppose it's a big difference, right? I mean, nobody's really sitting in front of an IDE actually writing lines of code anymore, which is a huge change from even as late as twenty twenty two, right? Pradyut: Yeah.

I mean our hiring strategy has like really shifted. ⁓ we actually find ⁓ like kind of like testing and like QA and like validation to now be our bottleneck. So like we're hiring a lot of ⁓ kind of people in the back end, right? So like have ⁓ Claud code or cursor, right?

Like write up all of these ⁓ new features. ⁓ it's to a point where we have cursor hookup to Slack. And so when a customer reports ⁓ a new feature, I can basically forward it to cursor and say, ⁓ cursor, go do this. It'll build it in the back end and it's all interacting through Slack.

And then we essentially have Graphite, which you guys probably know about, which will do some of the testing. And then we'll actually have human in the loop do the validation, verification at the end, just you know making sure that it actually integrates properly to like through our workflows and in our application, and then we're able to push it out maybe even the same day, right? ⁓ so we're able to go from concept to delivery in a day, which was obviously never possible before.

⁓ I will say though, like our our the accelerant of the catalyst of actually using AI in our engineering database probably happened sometime around like mid to end of last year, like it before that, right? Before like Opus four point seven or you know, that model kind of came out, it was cool and like we you still had to do a lot of like checks, right? It was still like, ⁓ you know, AI, not I don't want to say slot, but it was like instead of writing, you know, five lines of code, it would write two hundred lines of code or something like that.

I think today it's an extremely efficient tool when it comes to coding and we use it almost every day, if not multiple times a day. Martin Bielicki: Yeah, we probably had a very similar experience. Yeah, probably very similar experience. So once again, we started the company basically after LMs came into the place.

So we kind of always used them. ⁓ but really the acceleration happened, I would say the first kind of bit in like mid 2024. And then another another kind of ⁓ push, maybe probably a year later. And then recently it was more so like this year and kind of late last year, it was more so like multi agent they kind of systems or kind of Fino: Same with you over.

Martin Bielicki: Our engineers would be kind of managing multiple ages in parallel, ⁓ which also brings its its challenges. Sometimes it the the limit becomes the human mind's capacity and kind of what how many threads can you manage at the same time. ⁓ but yeah, on my side, on the more commercial side, we we also just use AI for for all commercial stuff, kind of ⁓ customizing decks, sales and all the ⁓ the boring but important stuff that you have to do as well. Fino: Yeah, I know.

Pradyut: Yeah, actually like to to kind of Martin's point, I think like the commoditization of like what looks good. Like if your product doesn't look good today, right? Like it's just not an excuse. I think you really have to think through a lot you know, maybe like product manager is the wrong phrase here, but like someone that's really thinking through the product and then let the engineering aspect of it be handed off to to AI and these agents.

I think that's really what's gonna differentiate product to product, right? It's like the application and the workflow and how we think about like delivering value, that's gonna be the differentiation, not the actual product. It's just the vehicle of that value. Martin Bielicki: Yeah.

Agreed. Agreed. I think there's still space to be to be smart about how what the workflows and what the broader UI is, not just a design. The design yeah it can be plus I think I think there's a difference between having good okay design, which AI can do, and a amazing design, which is quite tough to get still.

Pradyut: Yeah. Yeah. And and I think like you look at we'll probably talk about this later, but you look at like the the incumbent players, right? The question also is like at what stack does AI help with coding, right?

Like if you're on like a nineteen nineties, right, software stack, it as much as you try to feed it, you're you're gonna hit a ceiling pretty quickly, right? Like if you're not on, you know, I I don't know what the latest like, you know, software stack is today because I'm not an engineer. Martin Bielicki: Mm. Pradyut: But like, you know, I'm sure if you're not on anything post twenty twenty, right?

Like React and JS and all those kind of like scripts, like I just I don't think that you're gonna be able to succeed. And I I look at some of these old architectures that I'm not gonna say their names, but all you know, like the vendors are on, it's it's challenging, right? You can only innovate so much. Fino: Mm.

I I've got a good question in here from on the chat from Morton Wil Weiberg who says, ⁓ to what degree are you depending on prompting skills as opposed to letting AI and fill in all the gaps? To reframe it slightly, how do you see the friction curve trending for AI in terms of autonomy? I use a lot of LMs every day and they all have reliability problems and with the output quality. Martin Bielicki: In in in my work and like the more kind of commercial side, I think the prompt isn't as important.

It's more so getting the right context in. So if if we have the right if if if AI has the right understanding of, for example, what customer I'm approaching, why, what's the use case, then it can kind of work out what I'm trying to do or d definitely kind of go in the right direction. The prompt itself I haven't seen mattering that much. Pradyut: Yeah.

And I think like these foundational models have also gotten really good about like, hey, if I don't have enough like ⁓ information from you in the prompt, I'm gonna follow up with questions, right? Like if you work in in cursor or cloud code, like it'll be like, ⁓ did you also want to look at this? And do you also want to consider XYZ? So like it's getting kind of better at asking you to be better at prompts as well, right?

If it doesn't have all the information inside that prompt. ⁓ I can't again, I can't speak and now there's like more than just prompt. It's like a visual. You can send it a image and like a product screenshot and say, go do this.

And it has all the context inside there. So I think we are getting to a point where it is multimodal. It doesn't just have to be a prompt. It could be so much more, as Martin said, right?

Like context matters so much more than prompts these days. Fino: I was reading f it was you know, some of these thousands of posts you get every day on LinkedIn about how they're gonna, you know, here here's the seventeen thousand prompts of cloud to make you better at mar at marketing, whatever. But someone was saying that we were moving from prompting to loops. So which is I think what GSD was trying to solve before the guy walked off with all the money, you know.

But ⁓ I think that that's sort of the idea. You have well, you basically build your ⁓ skills on top of Git and then you You have a ⁓ your scaffolding so you can ask questions, it remembers, it can go look and get to find out what it knew, you know, ⁓ ten minutes ago because of course it forgets instantly ⁓ when it writes something. ⁓ Yeah, I think that it's it's evolving a lot. I ⁓ it it must makes me wonder too, how do you and I I'm sorry, this is wasn't on the script, but it still ⁓ interests me because you mentioned Frontier L L ⁓ Well, Frontier LM is not cheap, right?

I mean, I don't know how much I'm paying like way too much to anthropic these days. How do you guys deal with that because as a founder before you just had the headcount, right? And then a couple of s SAS licenses. Now you've got this variable cost of anthropic or open AI or whatever.

And it and I don't know about you guys, but like a ma even Cloud Max is gone within like the first five days of the month. I've already burned through the entire subscription. And then suddenly I'm on paper use and damn it, the bill is just so high. How do you how do you guys deal with that?

I because that's gotta be an extra burn on your on on your resources, right? Pradyut: Yeah, but we've seen s probably something similar, right, in the past of like with AWS as an example, right? People would say, ⁓ you had a you had a you had a server, right? And like you just paid it once and you just had to keep it in your closet and like you didn't have to think much about it.

And now like you've got S three buckets and EC twos and EC twos aren't, you know, cheap. So it's like how do you think about that as you scale your business and you know, cloud computing? And so I think we're going through that same like mental shift when it comes to cost, right? You're going Fino: Well, S three and things like that.

Pradyut: from, you know, headcount costs to to maybe tokens, right? You know, when we did AWS, you're going from physical real estate costs and hardware costs to cloud computing. And and that you could also say is kind of like a token, right? Where an EC two, you're you're really getting charged an E C two per minute or per hour, whatever, you know, like denomination they use, but kind of this the same thing at at the end of the day.

Fino: Okay. Martin Bielicki: Curious how it's gonna play out because ⁓ we were seeing like companies like Uber, for example, capping ⁓ I think AI spend like $1,500 per month for their coders. I do want to know how it's gonna play out in the long term, and now like the cost conversation now is coming into like back into the scene where it wasn't a conversation like a year ago. And also, I'm by extension, I'm curious how it's gonna work in our accounts and like engineering companies.

I think many of them are still not using Gen AI for like their development. Most of them, in my eyes, are using Fino: Yeah, of course. Martin Bielicki: AI experience simulations or the co pilots. I don't think for many of them it's a topic yet, but I think it will be one in like a year or two.

To in to in my eyes, our industry is like a year or two behind coding. Pradyut: Yeah, but ⁓ cost is also just like one f like ⁓ dimension that we look at. It's like it time, right? Like even if the the the the number of tokens or the cost of tokens we end up spending in a year is like hundred, two hundred thousand dollars, right?

Like that is a replacement of sure like one engineer, but the the amount of like that we would have had in those two hundred thousand dollars of tokens would be much higher than had count and engineering and it would be done way faster, right? So now you're able to not just Fino: Mm. Pradyut: It's not just cost. You're able to ship products faster.

Maybe you have to do a little bit more validation there, but like there is a given trade, even if costs creep up to the cost of like human labor. Martin Bielicki: Yeah. Agreed, agreed. But I think it's gonna be hard for companies to measure ⁓ that output of times.

And I that's also why Uber's capping camping theirs probably. yeah, it's a tough one. Fino: What about ⁓ I mean in terms of foundation ⁓ sorry, Frontier LLMs, what about homegrown ones? Are you guys looking at ⁓ training your own models or internally using LM Studio or Olama or whatever so that you avoid, you know, going all every single time out to sonnet or fable whenever the US government allows us to use it, right?

It is and I used it. I used it this afternoon. Very briefly, because it's so expensive. Pradyut: Back.

It's back as of yesterday. Martin Bielicki: For us for us, we're not training our models yet. It's on the kind of longer term roadmap. But what we are doing is kind of ⁓ essentially ensuring that we can use the cheaper models as often as possible.

So essentially building tooling around ⁓ around the the ⁓ the models and our kind of what we have built internally, deterministic tooling that essentially allows us to use lower intelligence models and that that's how we how we manage it. Pradyut: Yeah, we we look at it a little bit different way. It's kind of like a stitch between what you said, Michael, on like the loops and the what you said on like context. Yeah, nowadays people are calling it like the orchestration layer, right?

Like if you can create a well kind of like orchestrated understanding of when do we have to use right like Fable Five versus when can we use like a really cheap model? and you know, are we gonna use a VLM? Are we gonna use text? Are we gonna put in the raw step?

⁓ you know, that's also really important to that orchestration layer, I think, has really been our focus. And behind the scenes, there could be a variety of different models, right? To to your point, Martin. We could, you know, we have to make the smart decisions on knowing not to give the PhD right basic arithmetics.

⁓ understand kind of like what level of model do we need to use for complex tasks, and then like what are the the simple tasks ⁓ that much cheaper models are going to be. Great ad, and those they're gonna be faster at it too as well. Fino: ⁓ and and has ⁓ last question and we'll move on to ⁓ more AI inside your stack, but what how has AI changed the way you organize your programmers in the organization? Otherwise, has it changed Agile and ⁓ Waterfall?

I mean, everybody was already on Agile, but I thought that Waterfall was sort of making a little bit of a comeback. And now I wonder how agents have changed all that. I mean, it seems like soon you'll be inviting agents to your scrum meetings rather than humans, right? Because they're the ones doing all the The heavy lifting.

How how is that changing the way you guys cause you're guys both managing these enterprises with lots of developers? How are you how does that change the way you manage developers, basically? Pradyut: I mean we're startup. I don't think we have scrum meetings.

Like right. Like we kind of we're we have to be a lot faster than that. I I think I I talked about this, right? Like our our focus is really on like the bottleneck is just now more on like QA validation, human loop before we get in like market.

Like what questions like what is an engineer? Like I'm I am now just you know forwarding a customer ticket on a Slack channel to another Slack channel and saying add cursor. Fino: Mm. ⁓ Right.

Pradyut: Do this and it's just getting done in the background, right? So, like the a lot of people in our organization that didn't have the ability to make changes to, you know, with limits, right? And and with constraints, ⁓ that didn't have the ability to to make changes to our product. ⁓ something as simple as, hey, we just want, you know, there's there's a drop-down list and there's 300 items in that list.

And instead of just sorting it alphabetically, the search the customer wants to have a a search bar at the top. Of that list. Can we just add that? Right?

That that is quality of life. Like cursor is great at that, right? ⁓ that that is where we are finding the best use of AI right now inside of the non-engineering team, right? The obviously the engineering team is using AI in much, much different ways than non-engineers are.

So I think like the scope of what is an engineer and who can actually impact a product has gotten a lot wider in in our organization. Excuse me. But The you know, the other part of is ⁓ on the engineering side, what is the focus? And the focus right now is how do we enable those updates to actually enter ⁓ production?

And that just means testing validation, human the loop. Martin Bielicki: So yeah, for us, ⁓ we're a team of five, so we probably haven't changed our ways very much after after AI, a bit too early for that. ⁓ but what I have what I have been thinking a lot about is how AI will change how engineering companies organize themselves. And I think that has to come sooner or later or later, just how enterprise and software have changed how they organize their teams.

This has to come in engineering to actually harness the full value. And I think This will not come until ⁓ it this will be induced by a a software player coming in that kind of enables that clot code for hardware kind of curse for a hardware proposition that hasn't happened yet. ⁓ and that it'll still take time and it's gonna be huge ⁓ implementation experience probably for the for the consulting companies or whoever else ⁓ does it. ⁓ but yeah, but I'm curious about that and ⁓ and I've been thinking about that.

Fino: I guess maybe I should qualify the question. What I I guess what I was thinking is if you have multiple agents all hitting your code base at the same time, they could be r overriding each other and agile was supposed to help us not be destroying somebody else's code. So how do you how do you manage that? How do you if you're using multiple agents that are all hitting the same code base, how do make sure that agent number one didn't just overwrite the code that agent number two did because they're actually not even talking to each other 'cause they're all in their own little context universe that are are are not Talking to each other, right?

I mean that's why we used that child. So you could coordinate multiple people writing the same code with agents that are not really or or are you you think there's an is there an orchestration layer on top already? 'Cause I don't and Cloud Code. I mean I have seven Windows open side of warp and they don't know what the hell the other guy's doing.

That's for sure. You know, they they don't know. I don't know, maybe it's just a question. you're muted, I think, ⁓ Pradu.

I can't hear you. ⁓ Pradyut: That's a great question. That's a great question for RC too. I have no idea.

Martin Bielicki: No, I I got him. Pradyut: I was gonna say I great question for our CTO. I have no idea, I have no idea. Right.

⁓ I I couldn't tell you it, you know, too in detail, kind of like ⁓ you know, the engineering organization and like exactly how they're dealing with these merge conflicts or you know, whatever they're gonna be. But what I will say is like our it from like just not being a software engineer by background, like it the cape the capabilities and like the things that we interview for today, right? Fino: Ha ha ha. ⁓ Pradyut: ⁓ on a software engineering basis.

It's it's more around architecture. It's more about how do you think about like the ⁓ you know how do you think about the product. ⁓ maybe again going back to it, pro product management's maybe not the right, you know, ⁓ role or the the phrase here, but it's something really around like, okay, this thing has been built because again, as you can if you imagine it, it can be built today. It's really about how do we integrate it in a frictionless way.

to the product and to the workflow so that the end user can extract the most value from the thing that we just built. Right. ⁓ it's not just modules on top of each other where you have to kind of context switch and go to a different app or go to a different window. It all needs to be really seamless and the experience more than ever, right, matters today.

Martin Bielicki: Yeah, mm probably same same answer from my side. Yeah, but a bit of a more question to my CTO. Haven't heard that being a huge, huge issue though. So it's being managed somehow.

Fino: So in terms of your stacks, like where is ⁓ AI already ⁓ influencing the user experience? Is it ⁓ what's managing s the the underpinning the kind of the the yeah the the the plumbing? Or is AI and the entire you know from top to bottom it's already ⁓ all of stack or all of build and all of of bench are already leveraging AI just about everywhere. How how how have you guys integrated into your stack?

Martin Bielicki: Yeah, so AI is pretty central to our product. ⁓ we have built with AI in mind from from day one. ⁓ with that, it by design AI is not central to every part of the product. As I mentioned before, some parts of the product are more deterministic and we kind of lean on those more deterministic ⁓ methods, ⁓ like cat kernels or kind of different different tooling ⁓ to essentially ⁓ ease the load on on on the AI, ⁓ you know, ease the the dependence on AI and also just just to get determinism.

also AI is not that not made for determinism specifically. It's kind of how you ⁓ marry that ⁓ intelligence with determinism. ⁓ so currently we use AI to collect context, to plan tasks, to to kind of to make judgments on kind of some some decisions, ⁓ but we generally try to keep the execution and the actual completion of task ⁓ deterministic. And that's kind of how we how we get Really how we'll build trust with customers.

Pradyut: Yeah. We we've I mean, as we kind of started out this conversation saying like we started pre AI. and like just the nature of our business is we are storing all of our customers' CAD data, right? Millions of tens at now, probably like hundreds of millions of assets.

⁓ and so we have to be really careful in terms of like how we implement AI into our product and making sure that like the customer's very well aware that like their product data would touch AI if they choose to do it. Right. It has to be an opt-in for our customers, just kind of given the nature of ⁓ you know, how we implement build inside organizations. ⁓ now, over time, I think like the customer's mindset will shift and it'll be kind of like, ⁓ yeah, like, you know, why wouldn't I do it?

And you can kind of just throw it in your terms of conditions and that will be a general industry shift. But for right now, It is very much an opt-in like in your face. You understand that your CAD data is touching AI. ⁓ but also like we're not training any models on our customer data, right?

And we're very, very clear on that. So ⁓ I think trust is a big part of it, especially in this industry. and I think that again, just given the nature of ⁓ of build and and our product, just because we're we're hosting IP, it is really, really important for us to make sure that Our customers know if and when AI is touching their CAD and it has to be an opt in situation. Fino: Awesome.

⁓ so now we've been like four years into this whole AI thing and it's we've seen already the amount of change has just been mind boggling, right? ⁓ I wanna just ask like ⁓ in the last section of this part of just how how do you see things moving forward? Do you think that ⁓ we're gonna hit a sort of open AI moment where, you know, there was before AI and then now we've there's the after AI thing. Is there gonna be something where engineering just suddenly overnight just becomes a different thing?

Has it already happened and we missed it? ⁓ is it ten years from now? Is it six months from now? What what's your prediction on where AI is gonna take us and when it's actually gonna have a ⁓ immediate day to day irrevocable impact on how engineers build stuff?

Martin Bielicki: Yeah, so we're definitely building towards that moment. That's that's the thesis of our company really. ⁓ so the difficulty is kind of ⁓ in engineering, it's a bit harder to scale AI across every single use case. So in code, of course, ⁓ text is more native to kind of how we how we code.

⁓ so essentially ⁓ AI kind of impacted all of coding at the same time and got better at all the use cases at the same time. Whereas in engineering, it seems to be going more Use case by use case or kind of ⁓ software stack by software stack, maybe CAD first, then simulations, then then PLM. ⁓ so kind of the generalization is a bit tougher in engineering, but we definitely do see that coming. we do already see an irrevocable impact ⁓ of bench today.

The the question kind of so AI is impacting our customers' workflows. ⁓ it is just kind of the initial kind of workflows we we work with, not everything at the same time yet. So so yeah, ⁓ we are definitely pushing towards that. We believe that in sometime in the next 12 to 24 months we will have that open AI moment or that open AI AI moment where the the day to day of engineering changes.

But out of to my previous point, even if the tech is here in 12 to 24 months, the implementation and the change management is gonna be a tough, tough one, especially in these companies that are by nature quite quite much slower than than a meta or an Amazon and their kind of development cultures. So I think both hurdles are gonna be quite significant ⁓ and both have to be taken to account. Pradyut: Yeah, and I I like ⁓ just on your last point, Martin, I think there'll ⁓ some industries that'll be faster to adopt ⁓ you know, AI and in their tooling.

⁓ consumers probably gonna, in my opinion, take kind of the lead there just because of lack of regulation and like you know, they're just trying to get products to market faster. Consumer, consumer. Yeah. ⁓ not consulting.

⁓ yeah, I mean I I don't I don't know when. Martin Bielicki: They say consulting, so you broke up for a second. ⁓ consumer, okay. Mm-hmm.

Fino: Yeah. Pradyut: I but like the if is not really a question, right? Like i I think it will happen. ⁓ i I just don't know the when.

But I I do agree with Martin that like there is like different aspects that people are focusing on and like the application. Today, like there's I probably see five videos a day on like ⁓ text to CAD, right? And then it's like, ⁓ just because you can design it doesn't mean you can manufacture it, right? And then like there's another like AI thing around, ⁓ DFM and the so there's like there's so much to it.

⁓ and it's not as simple, right, as as code. Not saying that code is simple, but like it to Martin's point, like it no one thought about back-end AI engineering and front-end AI engineering and software, right? But like in hardware, you have think about engineering, you have to think about manufacturing, you have to think about supply chain. There's a lot more facets to it.

⁓ And and so like for us, I think that moment really looks like the democratization of information and kind of like how we're feeling about that today. Like you don't need to be a back end expert to now manage a database and to have compute in the back end and to be able to code ⁓ and have an app live. You could do all of that even if you didn't know anything about software engineering yesterday. You could do that today.

And I think that is kind of like the moment that I'm really looking forward to in hardware where it's like, I don't really need to know all of the thousand rules of like GDNT and I don't really need to know how to use SOLIDWORKS or or NX or whatever that you know the tool is. I just need to know this is kind of like what I want it to look like. These are the components I want ⁓ to be part of it. And I need AI to rapidly create this this product and then we'll start thinking about the downstream implications of sourcing and an actual DFM, but like the actual LM or the AI tool will already start thinking about that, right?

So it's kind of really like how do we enable more people to build things without having to carry all this knowledge base with us? Fino: Interesting. did you guys see that there's gonna be that ⁓ demo of five or six startups in DC at the C D fam and D that's gonna be kind of crazy, right? 'Cause you know this whole workflow, ⁓ N Top is Starry Digital, Ciscoit, all these guys I've interviewed and and been really blown away by.

Wish I could go. I hope you guys have a good time out there, Duane and company. ⁓ but I think there'll be some there I'm I'm sort of expecting that to be a as Pradyut: Yeah, in a couple of weeks, right? Fino: Almost an open AI moment there.

I think there might be some amazing stuff happening. So I hope that somebody records it and shares that particular presentation with everybody. so I guess you guys are both ⁓ as bullish or more bullish than you were back in 2022, then. It sounds it from to me to you guys are both more bullish than before.

⁓ well before I move on to the ⁓ the last subject, ⁓ for the ⁓ in the demographics that people watch this podcast, there's actually quite a lot of Martin Bielicki: Yeah, definitely. Fino: Like entry level or very early engineers, some university kids, because I write a lot of educational stuff. ⁓ what what advice can you give its founders, ⁓ and of awesome software companies to these younger graduates that are maybe worried about being AI'd out of a job? That's sort of a big topic right now.

⁓ so ha how what kinds of skills, what kind of things well I mean, Prado, you you pointed to human testing and understanding how things work and and into intuition. But what other what other kinds of things can people do to prove that, hey, I'm actually better than an agent and so you should hire me? Pradyut: I think it's like different than being better than an agent. I think it's just being different than an agent, right?

Like don't have commoditized skills, right? ⁓ like it I think a lot of people are like, ⁓ coding's dead because entry level coding jobs are kind of like now it's just a commodity, right? Like data entry is a commodity. ⁓ and I think like over time there will be things that are commodities and there will be like high value work that will be needed.

⁓ and we see that shift across all sorts of industries, right? Like we work with ⁓ customers that are in the autonomous agricultural space, right? And there's this whole uproar of like, hey, like are farmers going to be out of jobs? ⁓ if they're not you know, are we going to have low skill labor ⁓ kind of be cut in half because these tractors can kind of go and pick and and kill weeds and all that sort of stuff, right?

So it the question is not about like ⁓ Will AI, you know, do my job? It's like, is your job something that's differentiated? and I think just like having a unique take. ⁓ and ⁓ you know, we talked about this earlier, which is like our focus has really been on the the front end, right?

Like what is the actual product that we want to build and how do we really think about integrating it into like a very delightful experience for our customers? And then it's on the back end, which is like, can we validate that and make sure that our like thesis is true, right? Like Does it actually hold ground when we have, yeah, we just coded this like feature, but like how does it actually look in the hands of our customers? ⁓ everything in the middle is now just like AI, right?

⁓ and I think that could be true for hardware engineering, right? It's like the the world could be your imagination. Like it's like anything, anything that you want, you know, want to think about could could happen, right? And you could build it and you could go to a manufacturing shop and you could go and and get it and you know, built and assembled and source, all that.

could be true. then the question is like, what can AI not do? And it's really about like thinking, right? It's like ⁓ like it's creativity, it's being imaginative.

⁓ and I'm seeing a lot of really cool ⁓ you know, jobs open up in these larger organizations, ⁓ especially in Silicon Valley, around like creativity in engineering and less around like the actual like doing of the engineering. Fino: Awesome. Martin Bielicki: What's an example of that that creativity role? Do you have any any tangible examples like a what what the actual job spec is there?

Pradyut: Yeah. There's like ⁓ so if you like if you look at ⁓ like open AI and like anthropic jobs today, it's kind of like no one and they they even like say it's right like this this job may not exist in twelve months kind of funny because they know like you know over time AI's gonna to go. But there was one that was ⁓ about AI in ⁓ physical sciences and it's like, ⁓ like what are different ways AI could right, touch physical sciences? And all they're really trying to do is like, ⁓ w we've we've been focusing Martin Bielicki: Mm.

Pradyut: purely on code, right? And like there's this big market for code. But like what else could happen, right? And like where what are the opportunities AI being embedded in there?

And really what they're just thinking about is like people that are creative, understand a domain, and that can think about how can AI create a better experience in that domain. Right. And that doesn't require knowledge of, you know, like core AI LM infrastructure. That doesn't require knowledge on being a a PhD in material sciences or whatever.

But it's just like the applications, right? And that's the creativeness. It's and that's why I keep going back to like it's not it's not a product manager because like the product manager role like that was there five, ten years ago, is so different than kind of what you're looking at right now. ⁓ and again, like once you understand and can define that that embedded application of AI in these more nuanced and niche verticals, I think it opens up this whole new like industry, right?

That like AI can go and disrupt. And I think like us as startups We're trying to take ⁓ we're trying to take some forward paths towards embedding AI into hardware engineering, right? Whether it's ⁓ on the simulation side, whether it's on the design side, manufacturing, whatever it might be. But like it's happening also at these big companies.

And and that's again the big differentiator. It's how are you actually introducing these experiences to your end customer because everything else will, you know, product can get commoditized very quickly. Martin Bielicki: Yeah. to answer the question briefly about kind of what what Junior Engineers should do.

I think ⁓ first of all engineers are here to stay for for a long time. And I for for a long time engineers are gonna be ⁓ in the loop and that's kind of a a big feature kind of many, many of our customers really want. ⁓ but also kind of looking to what happened in software, I think there's a big rift in the software engineering market in a sense that the best are getting paid more than ever and the lower half are just kind of struggling to find jobs. And in the end, that's probably what's gonna happen in engineering as well.

I think that we we're a long, long way off of engineers not being needed or not being valued. But it's even more important to be in that that top half or the top quarter quarter even. So so yeah, the the advice would be just be the best you can. Fino: Ouch.

There's only so many places for this. ⁓ so I I like to ⁓ switch gears and and think about digital transformation. ⁓ you know, I ⁓ when I think about companies being Digimetri, we were t you were I think ⁓ it was prior to you alluded to how companies are evolving and and stuff like that. I think of digital transformation on a scale of one to five where one is sort of Martin Bielicki: Yeah.

Fino: Companies still on Excel and email, which is unfortunately the case on a everywhere. And then like a five would be, ⁓ they've already got autonomous, agentic, ⁓ adaptive digital twins, right? Which is basically nobody. ⁓ when you guys go and talk to customers, do you find that they're closer to one, closer to five, closer to three?

I mean, how how what what's your feeling? Because ⁓ no customer's gonna come on to my podcast and say, Yeah, I really suck at digital transformation, right? But you guys, you don't have to name them obviously, but you guys are in touch with these people. What what's your feel?

Are are companies starting to transform or are they still a bit reluctant? Pradyut: I think everyone I think everyone wants to be like a four or five, right? But like where are they today realistically? They're like a two, right?

I don't think they're like a one, right? Maybe some maybe some Yeah. Martin Bielicki: I'd I'd guess Ram goes there. Too being yeah, too being generous sometimes, but yeah.

Fino: Ouch. Pradyut: We we work with some like ⁓ like fabricators that are still doing prints, like physical prints and they're still doing like, you know, the the physical classic sign offs on on changes and redlines and and that's okay, right? Like ⁓ it's not bad. It's just it's it's work for them, right?

And and for them, like what is five? Like five could just like, ⁓ I have an iPad in front of me and like now I can redline on an iPad instead of like this physical print. ⁓ and and so like I think like everyone wants to be more innovative, but I think everyone also is like grounded in truth that like this is not gonna happen overnight, right? ⁓ there is change management involved, there are processes, there are other business applications, there's compliance factors.

⁓ so really for us, we see a lot of I and I think most people also on the other end, I don't know if they want to be a five. I think they want to be like a four, right? And I don't think they're looking and saying, I'm gonna replace my entire finance or engineering team with AI and like that's how we're gonna run the business. I think they're gonna be like, we're gonna be a 10 times more productive business with the same headcount and with the same resources, and we're just gonna grow market share.

⁓ so what we're seeing is like people that want to be fours that are probably like one and a half twos. Fino: About you, Martin. Martin Bielicki: Yeah, yeah, that that probably tracks. I ⁓ I don't have much to add here.

It's it's mostly I would say, especially on the AI transformation scale, it's like w one or two. ⁓ digital probably a bit better. ⁓ if if you look at those kind of larger enterprises, like automotive air space. Yeah, digital more more so.

⁓ AI not not there yet. And and kind of everybody uses AI but more so in in the ML sense than ⁓ Than Geniis, but more so. Pradyut: I also think like it's important to flavor in like what department you're talking to and who of that department you're talking to. Right.

⁓ if you're talking to the CIO, you're gonna get a very different answer than if you're gonna talk to like the VP of engineering or an and mechanical engineer. So and it's also that answer might be completely different from engineering to manufacturing as well. Fino: Absolutely. ⁓ so what ⁓ so my theory is that if you're gonna wait for the big three to do that transformation, it's gonna be a while.

Whereas if you go to build or bench, you're gonna get the AI without having to wait. And and I'm wondering if when you've done that, when the customer has seen what you can do, has there been an aha moment where like, ⁓ man, if I, you know, fixed broke down the data silos and I had good data governance and I had more data to feed to these AI powered tools. Holy crap, I would get so much more out of this and I don't need to wait for you know, two, three years for the roadmap to mature ⁓ at these big vendors.

Have you ever seen that actually happen or? Martin Bielicki: I was actually surprised a couple of times by kind of big OEMs approaching us with actually and actually asking for quite big visions immediately. So they are thinking about how do we transform our o our whole company and our whole development process with AI. And I've been surprised.

I didn't think this would happen kind of coming would come from them that quickly. ⁓ so yeah, so they've I they actually know what they where they want to go. It's more so that nobody can really deliver that whole proposition yet. So so I I I will th I think that they will be probably picking different players for different initial parts that workflow and then looking to who can then do the whole thing maybe or who can kind of what's the combination to achieve the whole vision, how quickly can can we get there.

⁓ but yes, I was surprised a couple times. ⁓ but then some other customers, ⁓ yeah, you come in with with one use case and then they start thinking, okay, maybe we could apply this here and there. Maybe this department also could need this. ⁓ but generally we we tend to start from the most kind of obvious painful one and then continue the conversation.

Pradyut: Yeah, we we take a a slightly different approach. Our you know, you you get kind of two buyers. You get like the buyer that's buying AI and you get the buyer that's buying value, right? And if AI happens to be hard of that value, then great.

and and we like the the latter ⁓ kind of buyer, right? Which is hey, I have a a problem and I'm looking to solve for it. If AI is part of that solution, amazing, right? But like if AI is not part of it, that's still okay.

⁓ and I think that's like really important to our philosophy. It's like we're not just we don't go to companies and companies will come to us and say, Hey, I want to use AI, how can I use AI? Right. It's that's you go to Accenture for that.

you really you're coming to us because you're saying, Hey, I have a very specific problem. I'm looking to solve for it, and your solution, you know, ⁓ claims to, you know, automate engineering documentation, right? Now part of that could be deterministic, part of that could be through inferencing and AI. ⁓ and every vendor will have their own approach in terms of how they claim to automate insuring documentation.

⁓ and it's like who is delivering the best value or the most amount of value, right? In terms of solving for that problem. I think when you look at the legacy players, they have these big claims, but they they fall short, right? They're maybe delivering 10 to 20 percent of the the proposed value.

Whereas we're able to ⁓ sure, leverage AI in many cases, but at the end of the day, actually deliver on the value that we're we're telling people, right? ⁓ And I think that it just the it's differentiating between the smart buyer and the buyer that's buying AI to buy AI. Fino: But there was I'd say a an an issue in your response though, because if if the customer goes back to Accenture, Accenture and Cap, they're all gonna propose another module of Windschill or Street Experience or Team Center or SAP.

They're not gonna they don't know who Bench is. They don't necessarily know who Build is. So you don't want that that to be the gut response 'cause it'll never get back to you again, you know? Pradyut: Yeah.

But like I I don't Think I don't think our our win to ⁓ an organization is hey we have AI and Winchill doesn't have AI, right? Or I mean Winchill has AI, right? It it really is kind of saying, you know, here's exactly what you can do in Windschill and the value that it has, and there's value there, and then here's the value and the philosophy that build has. And and the and it's like it's really just like philosophical differences at the end of the day, right?

Fino: A bill. Of course. Pradyut: ⁓ are you an organization that is ⁓ going to spend three million dollars on windshield and use it as a vault, right? And there's a lot of other value inside of it, but maybe your organization is not ready to adapt that value.

So even though that there's things there, you just can't do anything. Or are you looking at an organized or you're looking at a tool like Build where your organization can adapt it in in days or weeks or months, right? ⁓ but you can go deep into like the engineering value that you can drive. You don't have to go into supply chain and operations and manufacturing, just drive a lot, lot more value vertically into the engineering organization.

So I think it's just like a philosophical buy and it's like really it comes down to value. If somebody's looking to solve for problems, we're here to help solve for those problems. ⁓ and we'll give them our roadmap. Right.

To Martin's point, they do kind of lead with this what is the five year plan? But let's back into making sure we're solving for the one year plan for sure, right? ⁓ but they do want to know that you're gonna grow with them because their business is gonna grow and your system needs to grow with their evolving needs. ⁓ and there's different ways to solve for it.

But and I think the to just kind of end on this one note, the the the thing that we where we really win is like just kind of say, Hey, this this was X tool and this was it in 2020, and this is X tool, and this is it in 2026. I can name you on one finger or on one hand. You know, the number of changes that have really been happening, right? In in six years.

And now look at build or look at bench or look at any of these newer players, and you could probably, you know, get all of those updates done in a week or in two months, right? So the pace of innovation that we're able to deliver is far outpaces ⁓ the any of the incumbent tools. And that's like true for most industries. Fino: Did you wanna jump in, Martin?

Martin Bielicki: ⁓ I think that covers everything. Fino: Okay. Because I was just hoping that the thesis is more if you use ⁓ well it ⁓ just I guess a quit the question's more about the vi how do you guys get more visibility to the startups when the all the the public space is taken by the bigger vendors? How how do we make more space for your solutions to be more visible?

And I guess that's why I created Threadmo and I do these podcasts to give you guys sort of a platform. Martin Bielicki: I think there's a couple different approaches. I think first of all just comes down to your product. And I think a lot of the the big players are maybe not as focused as we are.

And they just kind of deliver the same value in the the verticals the applications that we choose, that we can. So so I think the core is just what can your product do versus what can theirs do. And I think for a long time they cannot match what we do broadly as a startups because just because of their speed and just historically they're just have just not known for being the fastest or the quickest to innovate. So so I I'm not worried about that part that much.

⁓ otherwise it's about building trust with these big companies. So right now I guess trust is built mainly through like pilots and through kind of proving them that it actually works. I think trying to increase the I guess ⁓ the position that starts holding in on our clients' minds would be would be interesting. And with that, some programs like ⁓ you know those programs like plug and play or or different kind of accelerators or whatever they are, you know, those innovation platforms, they can help.

And there's I guess different results that come from them. Some of them really elevate your profile and they give you a great introduction and then ⁓ the framing of the the sales completely different. Some of them don't, prank me, right? ⁓ so yeah, so in the end, I think the core in my opinion is just the product.

And I think ⁓ being able to showcase the trust in some way of the of the client, usually a pilot kind of with your approach to kind of how you build the product, is what you need. I think the question will become ⁓ yeah, w what happens when when ⁓ I don't know the salt systems releases something that's 10 times worse. But kind of kind of does the same thing. And my of course our thesis that but we still win.

And I think then the product speaks for itself. But but I think that's gonna be the testing moment. Fino: Yeah. Anything to add, Pradu?

Pradyut: Yeah. Well, I was just saying like it, you know, capabilities kind of like a checkbox, right? But like the experience is what really matters, ⁓ at the end of the day, right? So like, yeah, they could click a checkbox, we can do this, right?

As Martin said, but if it you know, it's a it's not a ⁓ a binary, can we do this, can we not do this? It it really is about well, how can we how well can we do this, right? And how well have we integrated into the core experience so that you can actually gain value from this thing that we've built. And I think startups You know, more than ⁓ incumbent ⁓ vendors are very customer obsessed and we really care about like what how are you feeling?

And we we take the time to actually have those conversations with customers. Whereas I think legacy, you know, vendors get caught into this kind of like innovators' dilemma situation where it's like, I'm gonna be focusing on my top, you know, fifty customers and I'm gonna index on problems that ⁓ that they have. Fino: Yeah. Pradyut: And that might cause a a ⁓ negative impact to the rest 2,000 customers that I have, right?

Because from a balance sheet perspective and from a a PO, those ten hundred customers, whatever they are, are 80% of my revenue and I need to keep them happy and I need to grow them. And that's just a different philosophy, I think, startups have. Fino: actually I d some good questions in the chat that just came in like a minute ago and I know we're almost done, but I'll I'll pose the first one to Martin and the second one to Pradu. How do we do how do we do that?

So the first one is from Susan Koons, my friend over at Zemens, and she wants to hear about how AI agents are beginning to automate CAD and CA workflows into end. Martin Bielicki: great question. And I think ⁓ the end-to-end point is a very interesting one because a lot of a lot of the customers we speak to want end-to-end workflows ⁓ and have end-to-end orchestration. But kind of to our to our previous point, I don't think that they are ready necessarily organization-wise, to do end-to-end workflows via agents.

And I think that there's gonna be need to be kind of evolution in how they structure their teams internally for them to be able to do end-to-end workflows. Currently, where we are seeing the biggest ⁓ adoption, kind of the most pool, is kind of end-to-end workflows, but kind of contain within one department or one team. So for example, an end-to-end workflow for the CAD department, but not CAD simulation systems. They want that for sure.

That's where we're going and that's what our product will allow, will enable. But I do think there's gonna have to be some reshuffling internally on who manages that. Because if we have an optimization problem going from CAD through sim through through requirements and then back, who manages that? Who who's who's the overarching team?

Is it CAD now managing the whole process? Usually we see then the CAE team not being very willing to give up their ownership. So so so that's kind of what we're seeing now. And I and I think the tech will come sooner than the organization readiness.

Fino: Of course no. That's great answer. Thank you very much. ⁓ that reminds me too, like we were having ⁓ I've had two podcasts so far on just e bomb and in bomb.

And it's to me, it's also it's the same problem because at the end of the day, you could do it in the RP. You could do it in PLM, but whose budget is that? The engineers on the PLM and the ⁓ manufacturing's RP. So everybody's gonna say, okay.

So the question to ⁓ to to Prad, who this is from Brian Stack, the founder and chief CEO at ⁓ digital twin uh.co.uk. So ⁓ UK startup doing digital twin.

He says, ⁓ as I'm awakening in the space, I'm and just having validated our digital twin for machinery and our an RD facility, I'd like to know your take on the future of engineering. We also run human-aid robot robots here. An engineer can look at a machine and go and create a fix and repair a part the AI ⁓ how do you see the wait a second, I I got lost in the reading of that. We also run human aid robotics here.

An engineer can look at a machine and go and creatively fix or repair a part. Pradyut: Ha ha. Fino: Whereas the AI will try just try to replace it. So how do we get around those kind of things?

And then he said, How do you because he said the executive teams are trying to decide how they can hire less skilled staff because the AI tools are gonna do everything. Pradyut: Okay. I think the question got lost in Fino: Yeah. ⁓ I think he's asking just about the future of engineering.

How at at what point can ⁓ are we gonna be to trust AI to do more of these engineering c tasks? And how do we keep skilled humans in the loop as opposed to just unskilled people that are just hands to do stuff that the robot can't do? I think that's sort of what he's trying to get to. Pradyut: Yeah.

I think it's like a crawl walk run approach and it's like trust but verify ⁓ is like our thesis, right? Especially a a build, right? Like it's like, okay, we we think that the AI could do a lot of this, right? Especially ⁓ on like the software side, but as I mentioned, we always have human in the loop on the verification, the QA side of things, right?

So it is always trust but verify and I think like that'll hold true for like hardware engineering as well, right? It's like, yeah, let the, you know, AI go from Texacat or whatever you want to do, but then like you know, the first step is always gonna be have a drafter, have a mechanical engineer look at drawings, make sure everything's annotated well, make sure like it's up to like DFM standards. So I think it's like you can achieve 90% of the work ⁓ in a relatively short time using AI, but it's leveraging humans still to verify that the data is accurate and like it's actually represented the the output that you were hoping to achieve.

I think over time that it'll It'll, you know, obviously grow into AI owning that verification layer. And then it's like AI verifying, AI verifying AI as we're seeing we're seeing today ⁓ in the software world. But ⁓ you know, I think it is that crawl, walk, run approach paired with trust but verify. Fino: ⁓ so but just to close out, thank you very much.

⁓ I was just wondering how where where can they ⁓ where can folks meet you between now and the end of the year, what trade shows you're gonna be at where they can come and get a live demo bench and build and and meet you guys. Martin Bielicki: Yeah, so we actually were in in startup Autobahn in Stuttgart today, as we're just wrapping out now. So so yeah, my CTO is there and my co founder as well. and ⁓ then we are also visiting ⁓ Detroit for the American Automotive Summit in St.

Fino: Nice. Nice. How about you, Pratt? Pradyut: While you're there in September you should swing down to IMTS ⁓ in Chicago.

⁓ yeah, so we'll be we'll be there at IMTS, we'll be in in DC in a couple weeks at C D F A ⁓ Yeah. ⁓ we'll be no I'm I'm just gonna be I'm gonna be just attending there for for a day. ⁓ yeah, yeah. I'll I'll share I'll share any like notes that I have with you.

Fino: Yeah, I was about to say IMTS is about that time, right? Martin Bielicki: Yeah, could do. Fino: ⁓ you're gonna be there? Awesome, man.

Are you demoing? Are you on the stage? ⁓ I'd love to hear your feedback afterwards. Thanks.

Pradyut: We've got a few. Like we've got like partners conferences. So like we're we're part of like the Net Suite S D N program and Net Suite's like our biggest ERP integrator. So ⁓ we'll be at Suite World ⁓ later this year in October, I think.

⁓ and then obviously IMT. Fino: And if there's another threaded event, I can count on you guys, right? I'm hoping to do one in Munich ⁓ in October. I'm still trying to figure that one out.

⁓ all right. Well thank you very much. I hope ⁓ that people I saw thank you for the questions ⁓ from the audience. And ⁓ next week I'll have I'll do another one of these.

⁓ trying to remember who that is. Who am I with next week? ⁓ well anyway. ⁓ it'll be fun.

⁓ it's been great. Thank you guys. ⁓ we'll see you guys ⁓ next week on the AI Across the Product License Podcast. And ⁓ take care.

Thank you. Pradyut: Thanks, Michael. Martin Bielicki: Thank you. Fino: Bye bye.