The Founders' Playbook: Operations and Systems Engineering — with Epsilon3 and SysGit

Episode Summary

The episode titled "The Founders' Playbook: Operations and Systems Engineering — with Epsilon3 and SysGit" delves into how artificial intelligence (AI) is transforming engineering and manufacturing processes. Hosted by Michael Finocchiaro from AI Across the Product Lifecycle podcast, the discussion features Max Mednik of Epsilon 3, a company that develops software tools for manufacturing, testing, and operations in complex industries such as space and aerospace, defense, robotics, and energy. Also joining is Steve Massey of SysGit, CEO and co-founder of SysKit, which offers a platform for collaboration on the design layer of engineering projects, focusing on requirements management and system architecture drawings stored in Git.

During the conversation, both founders discuss their perspectives on AI's impact on their industries. Max Mednik shares his skepticism towards generative AI tools like ChatGPT, acknowledging that while they have exceeded some expectations, he remains cautious about their capabilities. Steve Massey agrees, emphasizing that these tools are part of a broader toolbox for machine learning and that they complement traditional methods rather than replace them. Both agree on the importance of data governance and management in leveraging AI effectively.

For PLM and engineering professionals, the key takeaway is the necessity to adopt a balanced approach when integrating AI into their workflows. While acknowledging the potential benefits, it's crucial to understand the limitations and ensure that these tools are used alongside established practices to enhance productivity and innovation without compromising on quality or security.

---

Full Transcript

Michael Finocchiaro

might interrupt you like if someone ⁓ asked a question, right? All right. Well, this is we're live. This is Michael Finicchiaro from the AI Across the Product Lifecycle podcast. I'm very happy to be joined by Max Mednik of Epsilon 3 and the ever-evalent Steve Massey of Cisco. ⁓ We're going to have a great time talking about how AI is transforming engineering and manufacturing. And then since I've got an engineer and a manufacturer guy, that's a great discussion to have. ⁓ So Steve, tell us a little bit about Cisco.

Steve Massey (SysGit)

Yeah, so hey, I'm Steve Massey, CEO and co-founder of SysKit. We're a platform for collaboration on the design layer of engineering platforms, anything complex from a vehicle to a medical device to a spacecraft to a rocket. What that means is we specifically do requirements management and system architecture drawings. We store everything in Git in particular. And so we can use our customers existing investment in DevOps and the security practices that come with it. to move really, really fast, build a number of changes and just compare everything and get to that hopefully, many facts for both state that we can then hand off to someone like Max.

Michael Finocchiaro

Thanks. So Max, why don't you ⁓ jump in there and tell us what Epsilon 3 is doing?

Max Mednik

Sounds good. Thanks for having me. I'm Max. I'm one of the co-founders and COO of Epsilon 3. We make software tools to help with manufacturing, testing, and operations for all sorts of high stakes complex industries, including space and aerospace, as well as defense, robotics, ⁓ energy, and beyond. So we kind of help companies after the design phase to help them actually build and test ⁓ and run their hardware in production.

Michael Finocchiaro

Awesome. That's fantastic. It's cool too to have both the engineering and the manufacturing sides. I haven't done that to me. Usually I've got both. So that's going to be really exciting. It's been now almost four years since OpenAI kind of changed everybody's lives pretty deeply with this LLM thing. I wanted to ask Max, like back in 22, 23 when OpenAI and ChatGPT 3, I think it was 3, right? two, it was three that came out at that point. Were you like super skeptical? Were you like super enthusiastic? Where did you stand on that? Were you bullish or what did you think?

Max Mednik

Yeah, I've generally been fairly skeptical just because I studied computer science in college and I did AI classes and I know how hard it is to do anything that's kind of Feels magical or actually really good ⁓ It definitely has like in some ways Exceeded my expectations, especially recently in a couple things, but I'm still pretty skeptical and I think it certainly has its use cases and it can't help with many things, but it's definitely not like a kind see it, kind of just a magic thing, you just hand it stuff and it gives you perfect solutions to everything. ⁓ I'm generally a bit more on the skeptical side, but I think it has promise.

Michael Finocchiaro

Steve?

Steve Massey (SysGit)

Yeah, like Max, somewhere under all of this is a computer engineering degree. And so that betrays a little bit of the magic underneath the hood. So it definitely has informed us of the limitations. I think what's interesting is that we see LLMs in particular as a tool in the toolbox of available machine learning capabilities. Before LLMs really blew up in the last three years, we were doing a lot of like training of BERT models and

Michael Finocchiaro

So. I think what's interesting is that we see LLMs in particular as a tool in the toolbox, available machine learning capabilities. Before LLMs really blew up in the last three years, we were doing a lot of like training of BERT models and

Steve Massey (SysGit)

some computer vision models in order to generate the outcome we want for our customers. And now

Michael Finocchiaro

some computer vision models in order to generate the outcome we want for our customers.

Steve Massey (SysGit)

with LLMs, ⁓ used correctly, they can either reduce the training time we spend on a BERT model, or they can get us to a really interesting end state when building ⁓ document outputs, for example. But having actually used these, you quickly understand where they will fall over and where you can't trust them.

Michael Finocchiaro

they can either reduce the training time spent on the BERT model or they can get us to really interesting end state when building document outputs, for example. But having actually used these. Right.

Steve Massey (SysGit)

And then that has actually caused us to make investments in like, how can we ensure that we're building permanently correct outputs from like a library, even if there's like an LLM or an agent driving that library, we just, don't, we want to move away from LLMs being inventive and make sure that they're only transitive. So, yeah, so I'm with the right set of bounds. really powerful, but they are not a magical solution to everything.

Michael Finocchiaro

Yeah, I've got several friends of mine that have been doing using GSD and Ralph recently and to put guardrails around all that stuff. ⁓ So let's talk about development for a moment. I imagine that ⁓ since ⁓ first it was Windsurf, which then quickly became Cursor, but there were also Codium and there was a whole bunch of a flood of tools back two, three years ago. And now it's sort of centered around cursor, anti-gravity, ⁓ co-pilot, and so forth. I'm supposing that your developers are using AI in their daily workflow. Is that part of the daily thing you want them to do? And how have you integrated AI into the developer workflow? Because I think that some people might be skeptical thinking, well, AI and code, well, you It's just good for scaffolding or for doing a user demo, know, just throwing up a demo quickly. But I think it's actually far, far more useful than that.

Max Mednik

We definitely have been encouraging ⁓ our engineers to experiment with it, use it. ⁓ It has been useful in certain things, especially for some quick prototyping ⁓ or kind of like helping to plan out changes. ⁓ So we are definitely using it across the board and it has ⁓ improved productivity. think certain things that ⁓ it's pretty good at, ⁓ writing lots of boilerplate code, testing, building stuff that... is not very unique or has been built in lots of places that there's so many examples and it's kind of reinventing the wheel. So it literally could just build the wheel very quickly. that kind of stuff, it's actually has been good and it has boosted our productivity pretty significantly. We had one engineer that literally in a day built basically 75 % of Google Drive. Like what is Google Drive? Using AI as single engineer, obviously it's not everything in the last 5 % of Google Drive.

Michael Finocchiaro

Ha ha ha ha!

Max Mednik

is like what requires the 300 or 500. Yeah, exactly. But for a quick and dirty initial thing that is like that, it can do that. So we use it, but we're also very cautious with it. And we always have a lot of human code review and a lot of manual testing and all, just because of the nature of the software that we build and our customers and just the very sensitive situations that they use it in and that it has to always work. And so...

Michael Finocchiaro

The 80 % of the time, right?

Steve Massey (SysGit)

Yeah.

Max Mednik

it's not okay for us to ship something that's like, it has some bugs because AI built it or whatever will fix it. Like that's not okay. So we, we try to use it to be, to move quickly, but also have enough safeguards and checks and human code reviews and all that stuff to keep ensuring the quality. I think that's kind of the role of human software engineers is like kind of applying the level of taste and level of, you know, checking of the outputs.

Michael Finocchiaro

Do you find that your product managers are using it for doing PRDs and things like that? Is it really, I think that the biggest revolution is really for the product managers, right? Because suddenly they can build these PRDs in seconds rather than days.

Max Mednik

Yeah, for sure. We see that too. ⁓ You still kind of need to be careful just because it could be like kind of just random stuff that doesn't really apply to the product. But yes, we have seen that be useful as well and definitely helps accelerate.

Michael Finocchiaro

Have you seen the same thing, Steve?

Steve Massey (SysGit)

think what, if anything, really shows who really knows their fundamentals of computer science, because you need to have enough understanding of architecture, enough understanding of the tools available to actually generate the correct set of prompts and everything in order to generate code that is viable at the end of the day. And so you can't just roll in and say, hey, Siri, vibe code me an enterprise product for manufacturing companies.

Michael Finocchiaro

Hey Siri, vibe code me and enterprise.

Steve Massey (SysGit)

You will have a very bad time, but if you have a well-defined framework for your application to begin with, you have an understanding where the boundaries are, you have a corpus of unit tests you've already developed, you can reference all of that and you can mix in the things that are working with your product already that you've already written and the human has already written on your team, and then use that to help influence what the LLM is doing.

Michael Finocchiaro

It's great. ⁓

Steve Massey (SysGit)

It ends up being like really helpful. And so our team does, we use cloud code in particular, because I think there's like a, a little, it feels a little lower of the ground to be honest, than some of the more curated things and honestly like the credits are cheaper and it's all going to Opus anyway. so like we found that's been really interesting. And then separately, it's been really helpful for like, honestly me keeping documentation up to date.

Michael Finocchiaro

Yeah

Steve Massey (SysGit)

I can record a video of me using the application. I have a pipeline that'll generate, ⁓ you know, the correct updates to the make doc server, right? But like there's a way to do it that produces garbage and there's a way to do it that makes it really, really valuable and helpful. And we've also seen that with the team as they, you know, work alongside Claude and the rest of the folks in the organization.

Michael Finocchiaro

That's cool. Yeah, love the cloud code is just it's amazing. I said I was also using anti gravity and kind of blown away how well the Google's doing that. But yeah, cloud code also you ran out of the credits and there's something like, hey, of money. ⁓ Have you, I was, maybe think about, ⁓ in some ways though, the AI thing has sort of also isolated us. Is it still able to help us collaborate as like developers? Because I'm not, I mean, I haven't played around as doing several developers on the same project using AI because Kotco is sort of an individual thing on your fastest device. More of a question of Git and using Git correctly to get that right.

Steve Massey (SysGit)

Yeah, I'll take this because I like, I'm like really excited about this process. Before even we adopted cloud code or any sort of LLM assisted ⁓ product development, we were doing something kind of interesting called mob programming. It's where you can have two to like two to three engineers work on the exact same tick at the same time. It sounds inefficient, but it's actually incredibly efficient because for what you end up ⁓

Michael Finocchiaro

We were doing something kind of interesting called mod programming. It's where you can have two three engineers work on the exact same thing at the same time. It sounds inefficient, but it's actually incredibly efficient. for what you end up...

Steve Massey (SysGit)

⁓ losing in single screening, the development of a single feature, like half of that time spent in getting code into the mainline code base is spent on design reviews and product reviews. And so you can just have your design review immediately because two humans basically co-piloted the authoring of the software. And then your product person can jump in and authorize it immediately. And we've been able to like move really, really fast to the point that

Michael Finocchiaro

Wow

Steve Massey (SysGit)

We've lent, we just kind of lent into Kanban for the last two years instead of having to sprint cycles because of that. We still did the planning and so being able to drop in Claude. So you have two humans and Claude actually has accelerated them even faster because you have two humans making sure Claude isn't doing something wrong. And it's actually made us really, really fast. And if anything, as a fully remote company, ⁓ it's brought people closer together because

Michael Finocchiaro

Because you have two humans making sure a bot isn't doing something wrong. And it's actually made us really, fast. And if anything, as a fully remote company, it brought people closer together because

Steve Massey (SysGit)

You know, they can move faster, they get the easy wins and the things that were like really hard to do before. Like there's a lot less like hoodie up, hacker late at night, listening to techno and a lot more like collaborative development because of it.

Michael Finocchiaro

they can move faster. And the things that were really hard to do before, like lot less like hoodie up hacker late at night with the techno. I love that.

Max Mednik

I love taking...

Michael Finocchiaro

Is that the same experience you've had, Max? ⁓

Max Mednik

Somewhat. haven't yet. I mean, in terms of the broad ideas and experiences, for sure, our team also uses Cloud Code and finds it really helpful. ⁓ We've also developed some of our own internal best practices, the best ways to use it, and some kind of meta patterns around that kind of stuff. We haven't tried the full everything as a team or this kind of thing. We've done some hackathons. We've done a few different modalities and mechanisms. But yeah, but generally in terms of using the tools, ⁓ they're definitely promising. And if you use them the right way, kind of like the ideas you said, ⁓ I think they can definitely help accelerate.

Michael Finocchiaro

Yeah, so generally in the schools, they're definitely permitting and they're just using that right. But they're definitely permitting that. Okay, well, it just makes me think of my friends using a G, have you guys looked at GSD and Ralph? They just came out like weeks ago. GSD is getting shit done. It's a French guy came up with this, it's just all guardrails. It builds all the MD files and then like boom, it just keeps you.

Steve Massey (SysGit)

I don't think so.

Michael Finocchiaro

in line with your coding standards. then Ralph is like the, you know, the character on the Simpsons always asking questions. It's just like, just do it. Stop asking me every two minutes. Just do it. It's certainly interesting. Um, but it makes me think that maybe are we seeing with AI in this situation of coding, are we going to see Agile finally go away? Because, know, Agile in the beginning was supposed to help developers turn into a hammer to hit over the heads of developers. Right. I think that. I think we're going to see AI just basically wiping that away, right? We're going to have a new model or a new philosophy behind code, right? I mean, it doesn't sound like sprints are all that useful. That's what you were saying, Steve. Or Max, you can jump in if you want.

Steve Massey (SysGit)

Okay,

Max Mednik

I don't know. I feel like some things will probably change and accelerate, but I think some of the fundamentals will likely remain. I think some things that change, for example, the role of the PRD or knowing what you're going to build starts becoming very important and important earlier because the thing can build it faster. So you kind of need to know what you want to build. ⁓ If you don't have clarity around that, it's not like, it's going to take me two weeks to stand up the database and write just the generic infrastructure. And then I need to customize.

Steve Massey (SysGit)

go ahead.

Michael Finocchiaro

Yeah.

Max Mednik

No, you basically go from day one, you're working on all the layers very quickly. So knowing the spec, knowing the end goals and the guardrails and constraints that Steve mentioned, which are really helpful in making sure the model is performed the way you want, starts becoming really important ⁓ earlier. So I think the way that you work maybe changes a little bit, maybe some of the agile processes could adapt and improve and further accelerate. Like I don't know how many people that run sprints now, maybe they stop running them. Like I bet it's not going to go away, but maybe some things can get updated to move more quickly. I don't know. What do you think?

Steve Massey (SysGit)

Yeah. I mean, I sprints is the social construct. If they're working for an organization, we'll continue to operate. We actually just briefly did a sprint as like a kind of an outlier from our Kanban approach, just to meet a big workshop we did last week. And everyone's pretty actually excited about doing that. So we might mix it up later this year and do that. But like one thing that we noticed exactly is that pre-work is much more important. The quality of our tickets has never been higher because you need to be able to fully describe the desired end state for the robot to even do anything useful with the input. And that comes out from a more mindful approach of product review and then generating the PRDs and then developing all of the content, the pre-work that goes into the rest of the platform. Because again,

Michael Finocchiaro

The quality of our tickets has never been higher. You need to be able to fully describe the desired end state for the robot to even do anything useful with the input. And that comes out from a more mindful approach of product review and then generating the DRD and then developing all of the content, the pre-work that goes into the rest of classroom.

Steve Massey (SysGit)

These LLMs aren't that smart. They will try to do the bare minimum to make you happy and converge on that. So you need to like really detail everything out and you need to have a human like basically monitor it.

Michael Finocchiaro

Yeah, it's just Reddit on wheels, right? More or less. ⁓ So like, ⁓ let's talk about how AI is actually embedded in the DNA of Epsilon 3 and Cisco. I suppose that ⁓ nobody, no serious startup could even do software without having some level of AI in it because no investor is going to give you a dime otherwise. mean, Max, how have you embedded AI? it, so I like to ask, like, you know, is it something a user actually sees or is it sort of the, in the plumbing somewhere or maybe it's both? that's sort of the question.

Max Mednik

Yeah, so there's basically two big levels. One is the stuff we've kind of just been talking about, like internally and for our developers and for how we build the product. We're certainly leveraging it to be able to move quickly ⁓ and help build more quickly. That's one piece. But in terms of our users and customers, there are a few areas of our kind of platform where there's actually AI type functionality and we're continuing to add more and more. ⁓ based off of customer demand as well. One area is in terms of bringing in documents and bringing existing files and documents, it used to be like a little bit more of a manual process or it had to be in a perfect format. Now we basically allow users to bring in existing kind of old documentation that might be in more scattered or disorganized formats and the various models we have can basically help parse it and bring it in. Other areas like in terms of generating new stuff. you used to have to like, for example, our procedure portion where you can write, know, checklists workflows and procedures used to have to bill all those manually. Well, now you can with natural language say, make me a procedure to test this component of a rocket engine and do this, this and this. And it'll give you the skeleton of it. It's obviously not going to be perfect and you need to review it and tweak it, but at least give you kind of some. guidelines potentially or help you ⁓ get off of the blank page. There's also some stuff with asking questions about your data and answering those kinds of questions. So those are some of the areas we're continuing to add a lot based off of customer demand. One interesting thing though that we've noticed is our customer base, because we support aerospace defense and some of these kind of more sensitive industries, there's kind of like a split in probably most of them. they are very paranoid about AI and they very explicit security policies banning it. it's a little bit of tricky balancing act because for investors and for the customers that love AI, we want to be AI forward and have that kind of brand as well and innovative, we also don't want to scare the customers that are like, we are banned from using AI. So if your thing says AI anywhere near it, we can't use it. And so we can fully turn off all the AI. We've also built...

Michael Finocchiaro

Hmm

Max Mednik

all the AI functionality in a way that's in GovCloud, it's secure, it's local, it doesn't store prompts, it doesn't store results, it doesn't use any open AI, none of those services have access to it. So we don't have access to it, we don't train on their data. It's like hyper, hyper secure, kind of locked down version of AI, because some people, do want that. But for others that are still paranoid, we can just fully turn it off. So it's kind of an interesting kind of... balancing act that we're trying to do, but that's kind of at a high level where we're at with the product.

Michael Finocchiaro

Thanks Max, how about you Steve? know you're in a similar environment with a lot of A &D customers.

Steve Massey (SysGit)

Yeah, and I like what Max says, where there's a mass amount of paranoia. think when you, some of the more enterprise-y folks have made an investment in their own LLMs. if they, in acquiring and securing their own foundational models from the major providers. And so we just make sure we can deploy the platform in a way that can connect to whatever their end choice was. ⁓ We'll always experiment with, does this feature still work with an offline model like,

Michael Finocchiaro

Thank you. We'll always experiment with, hey, does this feature still work with an offline model like,

Steve Massey (SysGit)

you know, GEMMA or GPT-OSS, usually it's okay. ⁓ But like, like what Max was saying, like as a company, customers come to us to, to use a workflow that is valuable to their organization. That workflow has some implicit understanding to work, interact with our data model underneath. And

Michael Finocchiaro

you know, GEMMA or TTSS? Usually, it's okay. But like what Max was saying, like, as a company, customers come to us to use a workflow that is valuable to their organization. That workflow has an implicit understanding to work, interact with our data model underneath. And...

Steve Massey (SysGit)

the real benefit of using these AIs is like a one-time cost of like vacuuming up. like semi-structured data from other sources or producing semi-structured or human facing artifacts after the work has been done inside of our platforms. And so that translation layer is really where most of the, think the LLMs really shine and where we've seen. I don't know, I mean, think we're moving towards an age of, you know, we're making lower investments in like building the perfect CSV in border, right?

Michael Finocchiaro

⁓

Steve Massey (SysGit)

we're allowing more flexibility at the edges of our data model as a company. And it almost makes it easier to collaborate with other companies because of that. And so I think that's an interesting thing that I've noticed over the past year or so.

Michael Finocchiaro

So you guys both seeing because that's interesting you mentioned like customers training own models is that really becoming something that's a little bit more ⁓

Steve Massey (SysGit)

What?

Michael Finocchiaro

mainstream

Steve Massey (SysGit)

I don't-

Michael Finocchiaro

or that's really sort of the rare exception.

Steve Massey (SysGit)

I don't mean training their own models. mean, they go out like a large defense prime will have a relationship with, ⁓ right. And so they have taken the responsibility of us having to show them like figure out where the LM is gonna be hosted. And they just give us an endpoint we can connect to. It's often the same foundational model. It's just passing their security postures.

Michael Finocchiaro

with open AI or anthropic or whatever. Okay. Right. Cause I've talked to two or three startups that have actually built their own model, like their own, like Leo AI is maybe the most visible one where he built his large mechanical model. I'm wondering if that, are we going to see that more and more? It used to be like impossible. I just needed billions of man hours, but now with the new GPUs, it's probably a heck of a lot faster than it was. But I don't know. Is that something that you guys had thought about or it was just like, no, it's just too much work. And besides I can get as much out of a just leveraging the stuff I've already got. Or maybe it's just ML that you need, don't need really the LLM side.

Steve Massey (SysGit)

think what they're probably referring to is they've built an engineered pipeline that can use best in class machine learning building blocks to get to their end state. It's what we're doing as well, right? There is an LLM that can drive a, you know, effectively a Python object model with a graph database under the hood. You really want to dig into it. And I can say that's AI because there's an LLM involved, but like fundamentally they've engineered a larger system that like uses

Michael Finocchiaro

Yeah.

Steve Massey (SysGit)

building blocks from the rest of industry. So Max, I'm not sure what you think of that.

Max Mednik

Yeah, yeah, yeah. I kind of feel the same way. And I think that all of this kind of stuff is definitely stuff that we're exploring and using and people use different labels and different ways of calling stuff. But yeah, for sure.

Michael Finocchiaro

So, we've talked now about how you use AI in the development process and how it's embedded in Cisco and Epsilon 3. And you were both relatively skeptical going into the beginning of this. Now, where do you guys see it going from here? We've lived through the LLM thing and then last year was more agent to agent MCP. I'm not sure where you guys think we're going next. there going to be yet another like, well, now it's cloud code. think actually MCB and then cloud code changed the game. What do you guys think is like the next game changer going to happen? Where do you think that move's going to happen?

Max Mednik

think definitely the agentic stuff, we've been getting a lot more requests for that. We built an MCP. And I think for our kind of more AI forward customers, they wanted to be able to ⁓ write their own AI prompts and interconnect different systems that can then, the AIs can access their Epsilon 3 data and can potentially manipulate it and do stuff like that. So we've definitely been seeing a lot more requests around that stuff. ⁓ It's hard to tell what's in the future, but I think something that's exciting, at least for the people that I work with, is further kind of embedding some of these capabilities into the actual hardware and interfaces with the hardware. So that's something that some of our customers use Epsilon 3 for. can actually write, for example, test procedures and tests and operational things that can access the telemetry of the hardware, even on the lab engine, sending commands and get data back and... do pass-fail criteria and kind of analyze it and stuff like that. It's all very, it used to be very manual, but I think ⁓ if you leverage some of these ⁓ different models, could potentially either automate the creation of these things or have to kind of help monitor the data that's coming out of the hardware and sensors, potentially automate some of the testing and building activities. Kind of the future that I think is at some point gonna happen and it's probably. decade away, if not more, but I mean, I could be wrong. It's the more like Iron Man, like Jarvis, like, hey, build me this big suit and it grabs the parts and it puts all of it all together. And it's all like voice activated and engineers just kind of tell it what to do. And it does all the stuff behind the scenes. I feel like at some point we're going to get to that future with 3D printing and all these cool things coming together, but it's probably a ways away. But I think some of these foundational things are helping move in that direction.

Michael Finocchiaro

Steve ⁓

Steve Massey (SysGit)

I'll agree with Max, and I think what he's getting at is that ⁓ what has been sensationally in the media described as the age of scaling is over, where it turns out you can't just continue to throw GPUs at an LLM, continue to throw training data at an LLM to get a proportional increase in quality. I think what you're actually seeing is that that is rolling off. There is actually a good enough from the LLM foundational models that we have today and the rest of the work for the next, you know, until the next breakthrough is back to fundamental research and then engineering pipelines from building blocks. And like exactly what Max has described is like, yeah, the LLN is great. And honestly, it was probably good enough a year ago for a lot of things we're doing today with it. ⁓ And now it's how does that fit into, you know, my computer vision model? How does this fit into ⁓ the code that the team pulled together to drive that piece of hardware? How well does my pipeline work on an edge compute device with limited resources, right? So, yeah, I think that's where we're going with this. know, and it's, yeah, I mean, it's gonna be really interesting. I think we'll get to that. I don't know, I'm optimistic on the Jarvis thing. I think that's a great comparison. I came out at a conference last week, the panel I was on, and like, I think we'll get there much sooner, to be honest, because I think the building blocks are there. It's just how do you wire it all up and have all the pieces. and get it close enough. So probably won't build a rocket, but you might build the Group 3 UAS.

Michael Finocchiaro

And get another But we also need a lot more power, right? That's the other problem is those fusion plants that we desperately need to power this time. It makes me think of something else like...

Steve Massey (SysGit)

yeah.

Max Mednik

sure.

Michael Finocchiaro

Sorry, just had a brain fart there. I was just thinking about the Jarvis thing. are we going to also see changes in terms of the DevOps change? Because I think we already have ML ops, so are we going to start having AI ops so you can swap in and out LLMs? I don't think that that's very mature yet. I'm not sure that they're... So I think that's another thing we're going to have to figure out, right? We're have to do some kind of infrastructure as code for... AI, ⁓ LLMs, all that other scaffolding we're going to need over time. That's going to be an interesting evolution too, right?

Max Mednik

Sure, we've done a little bit of that kind of stuff like documenting these kinds of patterns or templates and these types of things to do exactly what you're talking about, Michael. So I agree with you.

Michael Finocchiaro

And do you think also, like I didn't even, I should ask you about when we were talking about the products, but like, you seeing customers asking also to change the user experience? now are we going, are they starting to ask for more chat like interfaces rather than clicking here and clicking there? Or I mean, on the machine floor, maybe it's still an iPad or some kind of a touchscreen, but will we get to the point where we can just ask it a question, right? Like even vocally just say, and what tool do I need? And it just tells you or points to it rather than you having to click and take your gloves off and then touch the screen again and all that kind of stuff.

Max Mednik

Yeah, we've had a lot of people asking for both voice as well as AR, kind of stuff. The voice thing makes a lot of sense actually because some of our customers, for example, they're the engineers like on a ladder, like doing something. And it's like annoying to carry the iPad or a second person needs to be next to them to like tell them, okay, connect this to this. And then did you do it? And so like a voice, what's the next step? Connect this, okay, did you do it? It's like, it would be a lot easier. ⁓

Michael Finocchiaro

You

Max Mednik

AR potentially as well, help you, like, oh, this thing connects into this little spot right here, like these visual overlays. So there's definitely some use cases. I think it'll still take some work to get there, but definitely at least the chat and voice kind of interaction is pretty straightforward, and we've definitely been hearing requests for that kind of thing.

Michael Finocchiaro

And so Steve, from an engineering point of view, since you're working more with the engineers than the guys on the shop floor, ⁓ is that changing too, the interface between ⁓ the engineer and the machine? Or are we still ⁓ driving our CAD and using our little mouse and clicking here and clicking there all day?

Steve Massey (SysGit)

Yeah, I mean, like ultimately the goal of our company is to move away from drawing boxes and lines between them to build a shared abstraction. ⁓ You know, what makes that really important Max touched on, which is yes, there might be a human to machine interaction that feels more like natural language, but your response isn't exclusively taxed, right? There's some sort of semantic visualization that needs to be presented to the user in order to understand, Hey, this bolt needs to go into this hole.

Michael Finocchiaro

Right. ⁓

Steve Massey (SysGit)

Or from our perspective, ⁓ if you change the as-built mass of this part, it violates these three requirements, your verification's out of date, and it actually screws up this action diagram, and now you can no longer meet your mission. And so how do we visualize those changes in a way that's familiar to an engineer who spent his last 20, 30 years learning how to do a certain way of engineering? So it's not just like, I have a chatbot. It's how do we present that? that semantic information in a visual and interpretable way so that human can make the decision at the end.

Michael Finocchiaro

And I'm glad you said that. It reminded me of the other question I'd like to ask at this point, which is for the younger engineers that are going to be watching this video over time that are probably already feeling a bit of anxiety because AI is coming to take your job kind of thing, what kinds of things would you suggest they work on and they focus on so that they're not as replaceable as just your average Python programmer, right? Which that's not really a great job description anymore, right?

Steve Massey (SysGit)

I mean, just fundamentals, honestly, like understanding why a system is built, like, and then what the discipline is actually doing. Cause the LLMs don't have that context. The LLMs are a glorified autocomplete that can converge on something that can make you satisfied at the end of the day. Like fundamentally, that's how the math is working out. So, ⁓ I've actually used my underlying computer engineering degree more often in the past two years, just by thinking through how we're building things and showing the team what to do. I've had to add it.

Michael Finocchiaro

to make use of the time that we have. Fundamentally, that's how the... ⁓

Steve Massey (SysGit)

just to communicate how we should be architecting our platform. And like, I think that's really, really powerful and interesting and it'll actually make, you know, that degree program more valuable going forward.

Michael Finocchiaro

Cool.

Max Mednik

Yeah, I agree. still think that setting the fundamental engineering disciplines, whether it's computer science or mechanical engineering or whatever your specialty, I think is still going to be extremely important because you're going to need to know how to use these tools, check them, give them the goals and constraints and all, and make sure that the thing's actually going to work in the end. And I feel like it's going to probably encourage people. to be better engineers and to kind of learn more and be smarter and everything. ⁓ Like I think the kind of super low level, I don't know, just kind of entry beginner, like those kinds of jobs maybe are potentially parts of them could be replaceable. But I honestly think there's gonna be a role for even more engineers later. ⁓ I was talking to a friend of mine about whether ⁓ his kids should learn computer science and he was like, why should they learn computer? science, it's just all going to be automated. But I kind of disagree. I think there's still going to be a really important role for even, I don't know, probably forever. I don't know, maybe in a hundred years, the AI is going be running a little more, just going to be like hanging out. But I think at least for the near term, there's still a very important role for humans to know what they're doing with respect to engineering.

Steve Massey (SysGit)

Yeah.

Michael Finocchiaro

Ha ha ha ha.

Steve Massey (SysGit)

Yeah, I really think it's a fun topic to talk about because what it's doing is mean your smaller teams can be more impactful, which means you'll have more teams tackling problems as opposed to having more people tackling fewer problems because of the challenges inherent in writing software, right? Like we're not teach, like we'll teach kids assembly in college if they want to take those classes, ⁓ but we're not writing assembly on a day-to-day basis, but people still need to know how to do that.

Michael Finocchiaro

Maybe the. What it's doing is mean your smaller teams can be more impactful, which means you'll have more teams tackling problems as opposed to having more people tackling fewer problems because of the challenges parent-running software has. We'll teach kids assembly in college if they want to take those classes, but we're not writing assembly on an end-to-end basis. But people still need to know how to do that.

Steve Massey (SysGit)

And like, it's kind of nice to know how the computer is working under the hood. And

Michael Finocchiaro

And it's kind of nice to know how the computer's working under there.

Steve Massey (SysGit)

then the next layer up went to C, the next layer up, which was like portable assembly. Well, Hey, I'm not writing, updating registers anymore. And then the next layer up was C plus plus and Java. And it's like, well, Hey, I'm thinking an object. Next layer up is like Python and TypeScript. Right. And there's still people at every step of the way. ⁓ and like, it just, it's easier for the bike.

Michael Finocchiaro

And then the next layer up went to C, the next layer up was a portable assembly. Well, hey, I'm not writing registers anymore. Then the next layer up was B plus plus and Java. And it's like, well, hey, I'm taking an object. The next layer up is like Python and Python. And there's still people at every step of the way. And it's easier for the like.

Steve Massey (SysGit)

people to tackle more problems as smaller teams. And if something really requires you to drill down, you'll still be able to do that. So I don't know, I think it'll make a better future of more people tackling more interesting problems because you can get more done with fewer people.

Michael Finocchiaro

people to tackle more problems with smaller keys and if something really requires you to drill down, you'll still be able to do that. I think it'll make a better future of more people tackling more interesting problems so that you can get more done with it. Well, and it's not like a simpler one away. If you're dealing with stuff on the edge and you're doing really electronics on the edge, that's going to be a simpler anyway, right? mean, on your Raspberry Pi or whatever. It's interesting to me because the dynamics in LinkedIn are so weird, what posts make it. But I just did this one post about the math of geometry and it just went completely viral. That was like six months ago. I put it up two days ago and it went viral again. Because I think that there's still a need to understand how these things work at the end of the day and it's not just about, anyway. ⁓ So let's talk a little bit about ⁓ digital maturity. know Steve, had a cool thing to say. I look at digital maturity for companies as being, well, why are you even talking about it? Because without digital maturity, you can't really get anything out of AI because the AI is only as good as the data you give it. It's even a bigger example of garbage in, garbage out than we've ever had before. Uh, I usually look at digital maturity on a, on a scale of to five, one I'm still doing email and I'm still using Excel 95 % of my day. Um, and then agentic AI, I'm basically like Max was saying, I'm off having coffee and surfing while my agent is taking care of all the dirty work. Um, I don't think very anybody's at five or maybe very, maybe that probably nobody. Um, and your experience, the two of you with the awesome companies you're already working with. Are you seeing companies more towards the one or are kind of getting toward the middle of three? ⁓ Let's start with that question. What are you guys seeing in terms of digital maturity today in 2026?

Max Mednik

I don't know, for the customers that I mostly talk to, I feel like it's a lot closer to one for the most part. Maybe one to two based on the scales you just described. There's still so much Excel, know, MATLAB simulation, I don't know, just these core technical tools, which they serve their purpose and I think they'll probably still keep using them for a while, but definitely still a lot of email, a lot of Microsoft Word, a lot of documents. And that's kind of where we can help potentially, because we'll help digitize it, we'll help kind of make it more intelligent, more usable. But there's still so much of that just like literal paperwork. ⁓ And so I think that there are definitely some exceptions. Some of our customers, some of the people we talk to are very kind of high tech and they're trying to automate a lot of stuff and they want to get away from everything fully manual. So there definitely are... pocket, but I think it's still quite the minority that are kind of heavily adopting these things. I think it's still pretty early days. I what is the New York experience, the people you talked to?

Steve Massey (SysGit)

I'd agree it's one and two, but there's like a ⁓ few outliers that are like, let's at least try to get to five. And they're willing to like make a massive investment in figuring that out. But those are the outliers. It ultimately comes down to like, listen, we live in a world populated with humans, right? And humans want to continue to do things and we will continue to do things. And there's a way we interact with each other. The point of the AI is to help. ⁓

Michael Finocchiaro

one or two, but there's like a few outliers that are like, let's at least try to get to five, and they're willing to like, make a massive investment in figuring that out. But those are the outliers. It also comes down to, like listen, we live in a world populated with humans, right? And we want to continue to do things, and we will continue to do things. And there's a way we interact with each other. The point of the AI is to help,

Steve Massey (SysGit)

take that semi-structured information and get into some sort of ontological framework so you can have a predictable non-inventive outcome from downstream machine learning or data processing. that's fundamentally kind of, that's not going to change. And ⁓ there's just a massive corpus of information out there that is still in that incredibly messy human layer to help digitize and pull into our various ontologies.

Michael Finocchiaro

that's not going to change. And there's just a massive force of information out there that is still in that incredibly massive human layer to help digitize and into our various technologies. And why do you, I mean, I know in my world, the PLN world, probably 99 % of e-bombed, in-bombed transformation is still in Excel. You already mentioned Excel, Max, but why do you think we're still stuck there? Is it because the UIs of the other products are so bad that it's just easier? Do you use Excel or have we just not done our job and giving them the way forward? Or is it just laziness? I know how to use Excel. Don't want to lose or learn anything else. What do you think it is? it or both maybe?

Max Mednik

I've thought about this. I think it's a lot of those a lot of the things you just said it's just there the tools are universal like Microsoft Office basically everyone has it everyone kind of knows how to use it and you can make it do all sorts of extremely Fancy things like I basically saw a few customers that in one Excel workbook with literally 50 tabs and thousands of formulas and macros They had like a full ERP everything it's like they like 10 different software tools that you might buy off the shelf, they basically just rebuilt in Excel. It's kind of amazing, but it's also extremely difficult to maintain. you have bad finger, you can just break the whole thing. If that person, one person knows how it works, then they're gone, like whole thing dies. So I think it's just, it's very powerful. It's very flexible. Everyone has it. And it's a little bit of kind of just like psychology and behavior change. if you're...

Michael Finocchiaro

Riddle.

Max Mednik

and a status quo bias, you're just kind of used to the way it's been working. And especially in more larger, more legacy organizations where there's some hesitancy to adopt new tools or ways, it's kind of just safe, it feels safer maybe to stick with it. ⁓ I do think you're right that maybe other tools might not have good UIs or just it's hard to, it's not like people feel kind of some hesitancy to learn something new or they don't want to make the investment, but. Yeah, I do think that there's definitely ⁓ a lot of this kind of ⁓ inertia with these old tools. don't know. What do you think, Steve?

Steve Massey (SysGit)

is the minimum viable amount of data processing that everyone knows at the end of the day. You don't need a computer science degree to enter data into boxes in a grid. And you can use the functions in Excel as a programming language to build interesting structures out of that. And that's like, I don't know. mean, it's good enough for many purposes. And I think from our perspective as a company, like, hey, at least it's in more structured than a Google Doc.

Michael Finocchiaro

data processing that everyone knows. You don't need a computer science degree to enter data into boxes in a grid. You can use the functions in Excel as a programming language to build interesting structures out of that. It's good enough for many purposes. And I think from our perspective as company, at least it's in more structure than a Google Docs.

Steve Massey (SysGit)

And then we can use that as a starting point because like for decades, you look at this, like massive, like organizations, like companies have been built on just processing business data by putting some guardrails around not breaking it. Right. Look at Oracle, right? Like that's basically where they came from. Right. You look at Microsoft's businesses as well. Right.

Michael Finocchiaro

And then we can use that as a starting point. Because for decades, you look at this massive organization. Companies have been built on just processing business data by putting some guardrails around not breaking in. Look at Oracle. That's basically where they came from. Look at Microsoft's businesses as well. Right?

Steve Massey (SysGit)

Like how do you scale information beyond I'm an Excel spreadsheet and it's fine for three people.

Michael Finocchiaro

Like, how do you scale information beyond either the mental strategy or the spine of the people?

Steve Massey (SysGit)

But then how do I make sure that there's no failure points along the way? I don't know. I'm going to think some of the AI stuff we're doing is going to make it easier to start with Excel if you need to, and then vacuum it up into Epsilon 3 and Cisco when you're ready to have those guardrails.

Michael Finocchiaro

but then how do I make sure that there's no failure? I'm gonna think some of the AI stuff we're doing is to make it easier to start with a solid piece and then backing it up into Epsilon 3 and Fiske when you're ready to have the work. Nice. So the second part of question is now that the company is at maturity level N, which is probably closer to one than 10 to five, and you come in with Epsilon ⁓ and you come in with Cisco, is adopting a software like bleeding edge AI informed kind of software, is that, is this service an epiphany? Is it a ripple effect where the company suddenly realizes if they had good data governance and they broke their data silos and they made this data available to the tools, well, damn, we'd have a much better process. We get to go to market faster, we'd have better quality and. Be easier to manufacture, less injuries maybe on the factory floor as well. ⁓ Are you seeing that already happening?

Max Mednik

I think so. And obviously, I don't know how much of a contribution, like I can't claim like, oh, everyone's super successful because of us, like, but I do believe that there is a role for that. And I have heard anecdotes of customers tell us like, after they set up Epson 3, they had, you know, less failures, they were on schedule more, they avoided, you know, making some mistake that could have been really dangerous. So I think there's definitely those ripple effects also.

Michael Finocchiaro

Hahaha.

Max Mednik

I think there's something that, psychologically, when you start adopting a new system or tool, you start rethinking how you do things. And you might find smarter ways of doing it or reconsidering, like, why are we doing this way that is just super manual? we maybe don't need step five of this thing. We could simplify it and automate certain things. So it forces people to reconsider how they work and to try to make it better, which I think is a good forcing function as well. ⁓ I also, I've also kind of been wondering like, are the people that adopt Epsilon 3, do they just happen to be the ones that are smarter, more forward thinking, more innovative? And so we kind of just support them as opposed to they're kind of stubborn and really like old school, but then they bring Epsilon 3 in to change their mind. I don't really know if that's really happening. I think it's honestly people self-select and, and then yes, they end up being more successful. And I think as a group, like we've been finding our customers are generally

Michael Finocchiaro

Hahaha Hahaha

Max Mednik

they're growing, they're raising money, they're achieving their mission of justice. And I like to think that we play some small role towards that, but it's hard to know which way the constellation arrow goes. ⁓ yeah, what do you think, Steve?

Steve Massey (SysGit)

I mean, like you're better quantifying the changes to the data and you can visualize those changes over time. where you were before moving a little chaotically and maybe you didn't have that data now, even just the presence of that can let you identify and quantify how fast you're moving. And then you can then take a step back and look at like end to end, you how did this experience compare to like the last program I worked on? Like, am I having a better time doing this as a human is...

Michael Finocchiaro

compared to like the last program I worked on. Am I having a better time doing this as a unit?

Steve Massey (SysGit)

Maybe not immediately, like the metrics are difficult to pull out of that, but like qualitatively, that's a significant thing to help teams move faster.

Michael Finocchiaro

Is maybe not immediately like the metrics are difficult to pull out of that, like qualitatively that's significant thing to help teams move faster. Because I was just thinking that hopefully it would also be management would also realize, hey, you we need, if we had better data governance and. Well, that's cool. I think that it's been a great discussion. I really appreciate you guys taking your time today. I don't know if you guys had some closing thoughts.

Max Mednik

I really appreciate you inviting me, Michael, and it was an awesome conversation with you, Steve, as well. think closing thoughts, we'd love to continue the conversation offline on LinkedIn, wherever. If you're a team that's starting to build hardware or test it or operate any sort of complex systems, we'd love to support you. And in general, I'm always excited to talk about the best ways to use AI, where people see it going. And so we'd love to connect with other people online as well.

Michael Finocchiaro

Awesome.

Steve Massey (SysGit)

Yeah, likewise. mean, again, thank you, Michael, for helping us on and Max, it's great to hang out with you today. Yeah, I mean, if there are teams that are just starting out in the design phase and they have multiple engineering disciplines they want to work with, or if they have a massive legacy program they're working on and they want to bring their design review capability on the hardware side into something that looks more like Git or is quite fiddly Git, we'd to talk with them as well. ⁓ And of course, anything with AI, you know, engineered machine learning pipelines overall.

Michael Finocchiaro

Awesome. Well, I think I told you guys this morning by email, but The conference is definitely on for Miami, Threaded Miami on April 13th. So maybe I'll see you guys there. ⁓ I'll probably try to do live podcast from there too. So that could be a lot of fun. There'll be some round tables. So maybe we'll see you there. And if everybody else will see you on the next podcast, I'm just trying to line up the next two victims of my interviewing skills. But this was a lot of fun. I thank you guys very much. ⁓ And we didn't have any questions because I don't know, the Riverside link to LinkedIn didn't work. The thing did work on YouTube. But I'll be editing this and putting it up so you guys will, and I'll tag Steve and Max so that if you guys have questions, you can send the questions directly to them. So thank you very much to everybody and we'll see you next time.

Steve Massey (SysGit)

Thanks, Michael. Thanks, Max. See you guys.

Michael Finocchiaro

Okay. Thanks Steve.