Tank Talks By Ripple Ventures

David Caputo on Harmoni's AI-driven factory floor software and hardware moat

56 min Listen to the original Episode page
Manufacturing TechnologyIndustrial AIHardware MoatsSeries AAerospace & DefenseOperational Efficiency

Summary

David Caputo, co-founder of Harmoni, discusses the inefficiencies in lower-middle-market aerospace and defense manufacturing, citing issues like tribal knowledge loss and manual data entry. He explains that Harmoni addresses the ISA-95 gap by providing context-aware AI (HAL) and proprietary hardware to operators, rather than relying on generic LLMs or greenfield autonomous factories. Caputo highlights the company's $10M Series A led by Bessemer and its strategy of building a hardware moat through embedded Linux and edge AI to ensure reliability and low latency in regulated environments.

Episode sections

  1. 0:00 – 6:22

    Intro and David Caputo's Background

    Matt Cohen introduces David Caputo, co-founder of Harmoni, highlighting his transition from electrical engineering to RBC investment banking and then to owning aerospace/defense factories. Caputo explains his path from Toronto to Wall Street, noting he entered banking after a fraternity friend's suggestion and built the leverage finance practice at RBC.

  2. 6:22 – 13:54

    RVE Partners and Factory Operations

    Caputo details founding RVE Partners in 2013 to acquire lower-middle-market aerospace and defense manufacturers. He describes the operational reality of these factories, noting that despite advanced equipment, they suffer from tribal knowledge loss, manual data entry, and inefficiency due to disconnected systems.

  3. 13:54 – 21:11

    Harmoni's Origin and Pain Points

    Caputo shares specific operational failures that led to Harmoni: operators using wrong instructions for mirror-image parts causing scrap, and losing a critical machine program on a retired machinist's USB stick, halting production for a year. He also mentions a factory tour where an operator was 'cutting air' to appear busy.

  4. 21:11 – 26:31

    Harmoni's Market Position and ISA-95 Gap

    Caputo explains Harmoni's position in the ISA-95 stack, specifically addressing the gap where human operators exit the automated data pyramid. He notes that while large plants use full stacks, 300,000 US manufacturers are job shops where humans interpret data, causing inefficiencies that existing MES/ERP systems fail to capture.

  5. 26:31 – 32:00

    Product Strategy and MVP Development

    Caputo reveals Harmoni did not use its own factories as a testbed to ensure unbiased feedback. He describes building an MVP with three pillars (labor automation, process control, observability) simultaneously, comparing it to insulating a home where all gaps must be closed for efficiency. Early customers rejected the initial complex dashboards, forcing rapid iteration.

  6. 32:00 – 34:59

    Balancing Customer Needs with Future Technology

    Matt Cohen asks how startups balance building for current customer needs versus future technological capabilities. David Caputo explains that manufacturing is highly regulated and complex, with legacy rules like FAA wet-ink signatures, making it risky for outsiders to build solutions without deep industry experience.

  7. 34:59 – 37:44

    Strategic Decision to Own Hardware Architecture

    Caputo details why Harmoni built proprietary devices with embedded Linux, RFID, and edge AI instead of using commodity tablets. He cites the need for specialized features like tool life management via RFID, visual factory indicator lights, and edge AI inference to avoid cloud latency and high compute costs.

  8. 37:44 – 39:28

    Series A Funding and Bessemer Partnership

    Caputo announces Harmoni's $10M Series A led by Bessemer. He highlights the long-standing relationship with Bessemer, starting from a seed investment, and notes the immediate strategic alignment with partner Kent Bennett during the Series A process.

  9. 39:28 – 43:58

    HAL: Context-Aware Manufacturing AI

    Caputo contrasts generic LLMs with Harmoni's HAL, which provides operational context (machine state, part specs, maintenance history) to operators. He argues that without this automated context, prompting an LLM for manufacturing advice would require hours of manual data entry.

  10. 43:58 – 48:10

    Modernizing the Installed Base vs. Greenfield Factories

    Caputo argues that while greenfield autonomous factories are innovative, they represent a small fraction of the $3 trillion US manufacturing industry. He emphasizes that 90% of manufacturers are small shops, and their value lies in trusted supply chains and safety records, not just unit cost efficiency.

  11. 48:10 – 56:11

    Hardware as a Moat and Business Model

    Caputo discusses the return of hardware investment, viewing physical hardware as a tangible moat compared to software. He describes Harmoni as

Do your own research

  • Verify Harmoni's customer base and revenue concentration by reviewing their website case studies or press releases for named clients.
  • Check Bessemer's portfolio and recent filings to confirm the $10M Series A valuation and terms.
  • Research the ISA-95 standard and compare Harmoni's solution against established MES/ERP providers like Siemens or Rockwell Automation.
  • Investigate the regulatory landscape for aerospace/defense manufacturing (FAA) to assess the risk of legacy rules like wet-ink signatures impacting adoption.
  • Analyze the cost structure of proprietary hardware vs. commodity tablets to evaluate the scalability of Harmoni's hardware moat.
  • Review the technical whitepapers or documentation for HAL to understand how it integrates with existing machine data sources.
  • Look for independent reviews or analyst reports on the competitive landscape of industrial AI and edge computing solutions.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): Everybody focuses on the lightning strike when something big happens, good or bad, but nobody ever actually thinks about, or it often gets overlooked, the path that leads there. So it always looks like luck in the end when you see that lightning bolt strike. But you forget that person was for years and maybe decades climbing up a hill, waving a metal pole. And when lightning finally strikes, it always looks that way. But don't forget. all the work that goes into leading up to that lightning strike. That's my favorite analogy.

0:34Matt Cohen: Welcome back to Tank Talks. I'm your host, Matt Cohen, founder and managing partner of Ripple Ventures. And today I'm joined by David Caputo, co-founder of Harmony. David started his career as an electrical engineer, but somehow found his way into investment banking at RBC, where we actually worked together for several years. David ended up spending a decade in investment banking, but eventually broke out on his own to buy and build aerospace and defense factories in the middle of America. He spent years on the actual shop floors watching well-equipped plants still lose time, scrap parts, leak tribal knowledge, and waste millions of dollars on inefficient systems. That experience is what led David to start Harmony, a factory. orchestration system that uses proprietary devices on the shop floors alongside custom software that coordinates people and machines with the simple goal of wasting less time and making fewer mistakes. With a $3 trillion manufacturing industry that still runs on paper, spreadsheets, and whoever happens to remember how the last job was done, David's bet is that the next wave of industrial AI doesn't need to live on dashboards, but instead at the front lines of production. This chat is for anyone who is building or scaling in the physical world and trying to solve real world problems in physical AI.

1:49Unidentified speaker (UNKNOWN): Now

1:50Matt Cohen: let's jump into the tank for this week's episode with David from Harmony. Thanks for joining us in the tank today, David.

1:58Unidentified speaker (speaker_0): Happy to be here.

1:59Matt Cohen: Super excited to be reconnecting with you after working together for so many years at RB. RBC on the capital market side. You were obviously on the banking side where I was on the trading side. You've had an incredible run after your years in banking. But maybe for our listeners who don't know you as well as I do, give us a quick background, your early childhood educational background, where you spent some of your early time in banking. And after that, before we jump into what you've been building with the team at Harmony.

2:25Unidentified speaker (speaker_0): Sounds great. So I live in New York now, but I grew up in Toronto. And where we first connected was my first job at a college, which I'll talk about in a second. So I studied electrical engineering in school, but you got to put yourself back into the 2002, 2003 timeframe when I was graduating to really understand what it meant to be going into the tech field at that time. It's a very different proposition than it was today. We had just come out of a dot-com crash. and tech wasn't looking like the most promising area to go work. And from my perspective, for no good reason, other than a buddy of mine in my fraternity literally said to me once that all the business school kids wanted to go work in this field called investment banking. I had no idea what that even was or meant, but it sounded interesting. And so I decided to take it upon myself to go sneak into the business school in the evenings and attend all these. information sessions and learn a little bit more about it. And it's one of those be careful what you wish for things, because ended up going to work in that field and spent almost a decade doing it. So I was at RBC Capital Markets. I started a program called the Generalist Program, where you move around and try out different areas of the bank, ended

3:31up landing on corporate finance in Toronto, spent a number of years there, and that's what brought me down to New York, I helped the bank build out the leverage finance practice down. here. It was an amazing ride, learned so much through that experience. But really, it almost felt like a bit of a startup. Joined a team of five, six people. And by the time I left, I think we were close to 70 people.

3:51Matt Cohen: That's amazing. You don't see a lot of people jump from electrical engineering right into banking. Thankfully, RBC Obviously, had that generalist program, which is now obviously an amazing program that has paid off handsomely for them. But what were you feeling like getting into that program early on? A little bit of imposter syndrome, I'm sure, but how did you eventually find your footing figuring out what was supposed to be probably common knowledge to maybe an MBA grad for someone who came in with an electrical engineering background?

4:17Unidentified speaker (UNKNOWN): Yeah,

4:18Unidentified speaker (speaker_0): I mean, let's just say I relied a lot on my peers around me. I mean, I had a great group, guys and girls who I started with who I think they could tell I was a bit of a fish out of water, but in investment banking, at least half is about your attitude and work ethic. And I think that as long as you're willing to grind it out and you're willing to put in the hours and figure it out. Let's just put it this way the math is not very complex. You're adding, dividing ratios, right? I mean, it was very easy from that perspective. But I did bring a lot of the kind of technical and in fact, even the early days of dare I say it, auto. automation into investment banking. Like one of one of the things I'll never forget it, it earned me the nickname computer, but I was assigned what was called the structured product comps. And it was this spreadsheet with gosh, thousands of structured products, which are these kind of exchange-traded mutual funds. And I think it was every week or every few weeks, they had to be updated. And someone had to manually go to every single website in

5:14Unidentified speaker (UNKNOWN): that

5:14Unidentified speaker (speaker_0): list and put pull them up and put in the nav and then go to the next. And it probably took five, six hours. I mean, to go through that many, everyone with a different site and a different place to look. And I remember thinking, you gotta be kidding me. You guys pay these young investment bankers well to go and click on a link and type in a price. And so it probably took me longer the first time. I don't know, can't remember exactly how long it took, something more than five or six. hours, but let's just say less than ten. But built a macro where there was a little button in the top corner and it used VBA and some ancient coding techniques here. But you could just kind of click the button and then it would just sort of run down and update all the prices in the sheet. And I remember thinking, boy, that's a much more efficient way to do it. But it's yeah, definitely got a few quirky looks when people walked

6:00Unidentified speaker (UNKNOWN): by and

6:00Unidentified speaker (speaker_0): saw it VBA up on my screen.

6:02Matt Cohen: You're giving me some PTSD, David. I had the same idea where I had to find the closing prices at the end of every month for like the balance of basket reporting for like the electronic trading desk. And when that Bloomberg macro VBA was installed, that changed my life for sure. Less hours spent pounding your head against the keyboard was important. But after spending a decade in leverage finance at RBC surviving the crisis, you decided to leave and start your own, I guess, small MA shop. In 2013, you started RVE partners, Rift Valley Equity, focusing on acquiring aerospace and defense manufacturers. What was the thesis behind that? How did you even think about getting started right there before Forgoing going after maybe working at a smaller startup investment banking place?

6:49Unidentified speaker (speaker_0): Yeah. So the last. area that I worked with RBC was leverage finance. And what leverage finance is, for those who aren't familiar with that field, is you're basically backing private equity acquisitions of companies using credit facilities, using debt. So think of it like a mortgage provider for someone who's buying a home. So we were providing the debt to back these acquisitions. And in that process, you obviously underwrite all these opportunities. You understand sort of what the thesis is. And then you also see the other side of it when they've gone well, when they haven't gone well. And there was a deep desire for me to try my hand at that exact same game. I saw many examples of it going well and not going well. And we thought, gee, why not us, right? We can do that too. And so the premise was raise some money around doing lower middle market. buyouts. So this is a very different segment than we would have been looking at for leverage finance deals at RBC, right? I mean, I think at that time the smallest deal we'd be looking at was $40 million of EBITDA. I mean, we're talking about focusing on companies sub 10, sometimes sub five, whole different area with a different level of sophistication. And what we wanted to do was bring a sort of large cap mindset and also systems, processes, and tools to these lower middle market companies. So our whole thesis was we're going to roll up our

8:09sleeves, get involved in these companies, help actually improve the operation. So not add overhead, but add resources through our own talents and skills around data and analytics and support. It could be everything from helping assess a new investment in a piece of equipment to putting ERP. systems in place or quoting systems or tracking the platform data or historical usage metrics, whatever would help inform either the operations or business development of these companies by building systems and tools. So that was our thesis. And again, it's one of those be careful what you wish for. We close the fund Fun to raise the money. And I remember celebrating. And then there was this oh crap moment where we realized, well, now we've got. we to do something with this. Raising the money is it's like getting out of the starting gate, right? In a race. And so we closed our first acquisition fairly early on and built a platform around aerospace and defense manufacturing. So that was really the bread and butter of what we were acquiring. Acquired a Ford shop down in Texas, Crow Precision Components, a machine shop in Tulsa, MM Manufacturing, a composite shop up in Massachusetts. called Aerobond Composites, and was amazing, kind of building this group of companies, learning a lot over that time, going through a lot of factories through the due diligence process. And yeah, it was a very informative experience. And

9:27Matt Cohen: what did the reality look like though, from operating in that like fun model spreadsheets Sheets of analyzing the capital of the businesses that they were requiring versus actually seeing the body? bodies on the ground and how they were operating these factories on the floor. How did that appear differently the first time you got to really get your hands dirty?

9:47Unidentified speaker (speaker_0): Yeah, I mean, what I would say is value creation is a lot harder than it seems. It's it's something where, especially in investment banking, where you operate at a hundred thousand foot level in everything you do, right? You're you're diving in an inch. deep into something that's a mile deep, and you generally see the success outcomes. And so if you're even talking to a company as an investment banker, you're in the top very small percentile of companies out there. So being at the hundred thousand foot level is different than being on the ground. And so there's a lot of hard lessons learned on what actually happens on a factory floor. I've seen things that work and things that don't work. And I'd say probably the biggest takeaway from that experience was just the reality of what I saw in these shops compared to what I was expecting to see. When you think about a factory that makes landing gear for a military platform, you're probably picturing an advanced operation in your head. But when you get in there and you see scratch pads of paper and not a single electronic device on a factory floor, mistakes being made. left, right, and center, and the craziest things occurring that you would never expect to occur, and just the complications and reluctance to introduce technology on factory floors.

11:04Matt Cohen: The acceptance of technology, did that stem from like initial just like worry of job replacement of sorts or de-leveraging the value that an employee may have had prior to the technology coming in? Is that the the guarded wall that a lot of these people have in their minds before they accept it? And was this also a bit of an age gap between the people on the floor willing to accept versus maybe the younger ones that were willing to accept it too?

11:31Unidentified speaker (speaker_0): You know, I'd say that's more of a factor Actor for robotics. Like if you're starting to get into physical AI, stuff that is very apparently intended on you know, replacing human operators. Most technology products out there are sort of efficiency enhancing, or at least claiming to be efficiency enhancing. There's thousands of products, you know, that target a lot of specific areas within manufacturing, most of them kind of tech driven, whether it's iPads being rolled Roll out across the floor or new applications where they have to log in and enter the information digitally instead of writing things down. There's a ton of these tools out there. The reluctance stems from the complication factor of adding yet another system. Because in pretty much any shop, there's some baseline technology that everybody has to use. You have to punch into a system to say, I'm here and I'm getting paid and I'm on the clock. And then you have to punch into some type of ERP system to inform it. what you're working on. That's pretty much status quo at any factory. Once you start

12:31introducing additional systems, what you're doing is you're requiring operators to use more and more tools to enter the same information to tell that tool what's going on. So now I have a maintenance prompt in my machine. I'm going to go log into the maintenance software. Okay, I'm Matt. I'm working at machine ABC. I'm working on part. DEF and I'm having this specific type of issue. And you're kind of entering all this information just to get that done. And then you got to go use this other system when you're going to fill out a check sheet because I just inspected something. I gotta go log into that. I'm gonna go enter all this information, tell it what I'm working on, who I am. And the same thing in some shops, you could have six, seven, eight different systems. We've seen even more where these operators are having to use all these different systems. And believe it or not, they're put in place to introduce efficiency. But as you can probably imagine, once you start stacking these types of systems, you're introducing anything but efficiency.

13:26Matt Cohen: Yeah, it kind of feels like your old macro spreadsheet days. And then combine that with what we've got now in AI with all the agents, thinking they'll add more value, but they end up adding more chaos sometimes. So you evaluated hundreds of factories, you spent more than a decade acquiring and operating them. Walk us through the moment where you started to see the disconnect between the machines and the ERP and the lack of coordination between the two that allowed you to come up with eventually the idea that you've been working on lately with Harmony.

13:54Unidentified speaker (speaker_0): Yeah, so I mean Harmony really came out of multiple very real pain points that I was experiencing in my own factories. I will never forget walking into the first shop we acquired and getting there bright and early. The shifts start at six, six thirty in the morning. It's kind of like the trading desk. And and watching seventy people line up at an ADP terminal and one by one punch in, and then watching that same group of people as soon as they were done, line up again at an ERP terminal and one by one punch in and it out. It literally reminded me of the Flintstones. I mean, I couldn't think of a a better analogy for what was going on there where they punch in with the dinosaur. And I just remember thinking, like, is this really how it's done? And asking the shop manager and them saying, Well, I mean, you show me another way to I mean, yeah, you gotta tell ADP you're here and you gotta tell your ERP system you're here when there's no other way to do this. And that was just the beginning. I mean, all kinds of things. were happening. And because remember, we're talking over many years here, but you could imagine this one time, and I've told this story before, where we make wing ribs in my Tulsa shop, and these are for a business jet platform. And there's a left hand, a right hand, and they kind of sequentially go down as you can imagine what the inside of a wing looks like. And so they have kind of these mirror image parts, and we're

15:13making them in my shop. And the operator takes one of these wing ribs out from the machine, it's done its cycle. And there's a couple of manual operations that they have to do on it once it comes out. They pull up the instruction. We have fancy work instruction software. We'd invested in, believe me, I was investing in all the technology I could in these factories, and pulls up the instruction, follows the instruction, bores a hole in the part, and then realizes he had pulled up a left-hand instruction, but it was a right hand part.

15:39Unidentified speaker (UNKNOWN): of

15:39Unidentified speaker (speaker_0): the machine. And so they look real similar, but for that mirror image thing. And the problem is you end up with a crazy expensive piece of scrap. And that type of problem following the wrong instruction was happening over and over again. And I started realizing: wait, work instruction software doesn't actually solve that problem. This is a different type. The nature of this problem was different. And there was another one in my of shops. in New Jersey where we were making these defense components. And there's this one operator where the standard way to distribute the programs that run on these machines is every operator sort of has their own USB stick, and we would give each operator their programs that they're going to run. And this one guy had been running the same program for many years. And then one day he was a great employee, probably our best machinist. And one day he retires. and he moves back to Poland. And let's just call him Jimmy. And so Jimmy moves back to Poland and and then we say, Okay, let's give the next guy who's replacing Jimmy the program. So we give him the program, he goes to run it. He's like, Hey, this isn't even written for

16:40this machine. I mean, this program is written for some machine we owned ten years ago. And then we started realizing what had happened, which was the only current version of this program resided on Jimmy's USB stick. And Jimmy was deep in Poland at this point, long retired and untraceable from our perspective. We had no way to get a hold of him or get that program back. And so we had to re-engineer this part for a year. And and it literally took a year because that it not just developing the code, but you actually have to put these parts through all kinds of test procedures once create new sort of engineering. fixtures and programs to make them. And so it was a massive cost, as you can imagine, to shut down production on a part for a year. But I just remember thinking, there's got to be a better way.

17:24Matt Cohen: That's unbelievable. I mean, just to think how many of these existing technologies were basically sold to create more efficiency, less redundancy. And in fact, they created less coordination and more work to be done. So what was the kind of final straw that made you realize you had to build something rather than keep buying factories and just stitching stuff together?

17:46Unidentified speaker (speaker_0): It's funny. There's no one event that precipitated it. I could sit here and tell you stories all day. There's another one, well, I'll tell it very briefly. We were touring a factory and we're watching this was a company we were gonna buy, we didn't end up acquiring, and the operator was cutting. this big, beautiful part. And I remember thinking, gee, it's really quiet. I can't believe this is such a silent machine. What a beautiful machine. And then the spindle's moving, it's doing this whole thing. Then we walk around the back of the machine and see there's no chips coming out of the belt. And I guess what had happened was it's kind of like a show. They were just putting on a show where the operator was sort of put an offset on the machine and was cutting air the whole time. You're standing in front of the machine looking busy, and all these kinds of things were happening. And like, and again, something so preventable where you could see the production efficiency and the quantity of output and just measure that against how long things should take and realize that problem was going on. And so I could tell you stories all day. So there was, it was no one thing. I guess it was like you might remember there was one Simpsons episode where there was this like electric cupcake.

18:46And I forget it was with Homer or Bart, someone keeps grabbing the electric cupcake and getting shot. shocked every time. And it that sort of felt like me at one point, where I'm like, This has just got to stop. I'm not getting shocked anymore. You gotta do something about this. You

18:59Matt Cohen: can't keep saying dough forever and hoping it'll

19:01Unidentified speaker (UNKNOWN): change, I

19:01Matt Cohen: guess. So you co-founded Harmony with Adam Ellis in 2023 after COVID. Tell us about the partnership. How did it form? What were the strengths that you and Adam brought to each other to the table? And when did you guys officially decide to go in all in? on factory orchestration?

19:16Unidentified speaker (speaker_0): So believe it or not, I didn't want to build a company initially. I mean, my goal was to buy some software and fix this problem. When I realized that didn't exist, my next step was: okay, we'll build a tool for my companies. Let's go and build some kind of internal captive solution and asked all my managers, hey, who should we call? Who should we use to help us build something because that was the enough is enough moment. And one name came up multiple times, Adam Ellison Ohio. You

19:47Unidentified speaker (UNKNOWN): got

19:47Unidentified speaker (speaker_0): to call Adam Ellison Ohio. And then I separately found a blog post that he had written. It was a pretty insightful piece about connecting machines to ERP systems and using sensors. And remember thinking, this is exactly along the lines of the type of stuff that we want to do. So I'm going to reach out to this guy. So this was a cold email reach out saying, hey, love what you're doing and would love to have you look at a project. And we we had one phone call, and two things came out of that call. Number one, I realized he was a really good dude. We got along really well. And number two, we both realized that this problem is so much bigger than my group of companies. The fact that this type of solution didn't even exist in the manufacturing industry. And we were having to build it in the first place was was enough. And it was literally one of those UN, yup, let's do this kind of moments. And and that's kind of how it all started.

20:40Matt Cohen: That's not normal for a startup to get joined together through like a connection of one call. it

20:45Unidentified speaker (speaker_0): It was yeah. I mean in reality I'm kind of condensing it a little bit. There might have been a few calls. but Yeah.

20:51Matt Cohen: Very special recipe that you were able to form with Adam. Yeah. But can you explain? We've chatted over the years about where you do and do not play when it compares to other players in the space. Harmony sits in this wild West no man's land kind of between the middle of ISA ninety five stacks and ERPs and MES. Maybe explain to our listeners who aren't so deep in the manufacturing world. how you fill this gap on a typical shop floor every day.

21:19Unidentified speaker (speaker_0): Yeah. If you can't imagine, this is a topic I really enjoy talking about. So this, the ISA 95 stack, for people who aren't familiar with what that is, a standard that came out back in 1995 for the manufacturing industry that sort of outlined how data flows within a factory. And so there's five layers of this stack. right? You got the top layer, which is your ERP system. That's sort of outlining what should be happening, what demand do we have for new parts? When is it due? Right. Everything that you can imagine is dictating what is eventually going to happen on the shop floor. And then it goes down from there. You have an MES layer right before that, which manages the Is actual execution that happens on the shop shoff floor. So it takes what should be happening and then translates that into. okay, who's going to do what on which machines, and then what was actually produced when that was done. Below that, you have your SCADA layer. And the SCADA layer is meant to optimize the actual production activities itself. So think about a Coca-Cola bottling plant where you got a line,

22:19you can turn up the feed rate, and you're going to offset that against more scrap being produced because you've got kind of less time. to sort of switch from one process to another. And so it's the connection between what should be happening on the floor and what is physically happening on the equipment. Then below that, you got your sensor PLC layer. That's kind of what's actually operating the machine itself. Every machine has a control panel. So this is getting right down to the bottom layer. And then the lowest layer itself is the actual machine and the physical moving components, motors actually. actuators, low-level controllers, that kind of thing. So you got this whole stack. And then I say 95 identifies how the data flows from the top, the demand, right down to a part being produced, and then back up to the demand to know that it was done. This system worked really well in large factories. Okay. And the Toyota plants, like I said, the Coca-Cola bottling plants, the anyone doing anything repeatable in a high volume has pretty much implemented this system to some degree, usually fully, effectively. The problem is most of the manufacturing that happens in this country is not that. It's a lot more disconnected workflows, job shops. Like there's hundreds of thousands of shops in this country. I think the the number I usually go to is 300,000, plus or minus penny depending if you're looking at.

23:39factories, companies. But let's just say 300,000 manufacturers in the US alone, 90% of these shops are fewer than 100 employees, right? These are not highly automated plants running the same product. They're changing over multiple times a day. And the problem in that whole stack that I just outlined is that middle layer that I call the SCADA layer. And the issue with that layer is that you don't have a knob on the back. of a human operator's head to like turn up and down the feed rate versus quality, right? All of the data leaves this sort of automated pipeline and goes into some type of physical form that human operators then interpret, execute. So that it sort of escapes that pyramid. The human operators handle a bunch of things like which program they're going to run, follow. following some instruction, operating the actual machine itself, goes back into the pyramid to then do the, you know, sensors and the actuators, comes back up when a part is done, and that human operator organic comes, goes outside of that data pyramid again. Some unit operator goes back to the system and says, Hey, I'm Matt, and I finished 10 pieces. Now I'm done, and then eventually gets back to ERP. So it's that exiting and entering of that ISA 95. layer that has been impossible to solve for the manufacturing industry that I believe is the root cause of almost all

24:58inefficiencies and almost all scrap that occurs. And

25:02Unidentified speaker (UNKNOWN): the

25:02Matt Cohen: reason why this wasn't getting solved for the smaller manufacturers was because of the cost to build something, the dispersion between all the use cases. Was this not something that someone before you thought? of solving? What was the hindrance to solve this in the first place? It couldn't have just been a cost or scale problem.

25:20Unidentified speaker (speaker_0): No. So I mean part of the problem is you just have it's a bit of a many to many problem. You've got dozens of ERP vendors, more hundreds actually, if you really want to look across different industries, right? You've got hundreds of types Typ of equipment on the bottom side. And so you need this sort of many many-to-many to many adapter that's going to work with all of those ERP systems and all of those machines. And so there have been vendors who kind of move further down. Like you have a lot of guys who develop really advanced MES systems, or frankly, a lot of companies who operate only in the MES space that will move down into the sensors and actuator space. And so you have areas and pockets of integrated stack, but what you don't have is any type of universal tool that you can sell into the market or you know offer to the market to solve this problem in a kind of generalized way. And so that's what we

26:14Matt Cohen: must have started with some wedge because you can't just go in with a full orchestration system, right? Your product has three pillars, I understand labor automation, process control, and observability. Which one did you start with first? And I assume you chose to eat your own dog food at your factories before selling it and offering it to others, correct?

26:31Unidentified speaker (speaker_0): So actually the answer to that is no. And there's a really good reason for that. And so obviously that would have been a very low-hanging fruit way to start off your company. Boy, isn't it great to start off with some captive customers as a test bed. But the most important thing for me was that if I'm going to dedicate my life. to this mission, I want real signals of efficacy. And the last thing I want is to bring a product where everyone's going to say, Oh, yeah, this looks so great. This is exactly what I wanted, but not really believe that simply because of who's bringing it or introducing it. And I didn't want to use them as a test bed for it. And so my whole premise behind this was. let's build a product, an MVP, right? So Adam and I spent a long time heads down building. And we built an MVP. And to your first question, that MVP actually did include all three layers, but nowhere near to the same degree that we've penetrated those layers today. But we had aspects of all these layers. And there's actually a very important reason why. It's not just because we wanted to build more, the classic, like try to do too much too early, but it's because I view efficiency in a factory a bit like heating a home. So imagine you're saying, I want to get my home to be

27:51really energy efficient, right? You could go out and buy yourself a highly efficient furnace, right? Some brand new heat pump furnace that's going to use geothermal energy and do everything. ultra efficiently. But if you leave one inch gaps in all your windows, it is not going to make a lick of a difference, right? All of that efficiency and heat flows straight out the one inch gap in your windows. And let's say you address the one inch gap in your windows, but you have no insulation in your walls. You're sort of moving the problem around the plate. And so factory orchestration is not a piece. piecemeal kind of thing. The concept of factory orchestration is that you have to close all the gaps if you actually want to realize efficiency. And so we had the beginnings of each area. So we had a smaller number of labor automations. We had a more primitive process control, but process control nonetheless. And we had a simpler observability system.

28:49Unidentified speaker (UNKNOWN): It's

28:49Matt Cohen: unbelievable that you took the approach or the path that almost no founders would take, which is the harder path because you wanted to have a higher efficacy rating on the quality of the product you're offering. Most people we see at the early stage say, Well, my parents have a factory that we started with, or my cousins, or some type of family member, or close a relationship. And yes, that is an easier path to getting adoption, but eventually it comes down to, well, are they actually telling you the truth and helping you get better? Some people will say yes, they're my family. Of course they're going to be honest with me. But are they really going to tell you the hard truth about how shitty this is and break your heart at the same time? And I think is what you wanted is you wanted to be told that it was crap and not working right before you actually started to put real sales and effort behind it, correct?

29:36Unidentified speaker (speaker_0): Absolutely. I mean, obviously it can work. And I'm sure there are many successful startups who have started in sort of let's just say friendly customers. So I don't want to knock that as a path overall.

29:47Unidentified speaker (UNKNOWN): Of course.

29:48Unidentified speaker (speaker_0): I just think you have to be real vigilant because at the end of the day, you're only going to be fooling yourself. And it is not a light journey to be a founder. And I think that the more real information you can get, the better. And frankly, those, as you could imagine, I mean, the first couple customers, again, like I'm sure most startups, I mean, that was really tough. We went in there with a half-baked product. And there were many moments in those early customers where they pretty much told us, hey guys, that this just feels a little too early stage for us. And so I think we're just gonna pull the plug here and come back to us when you guys are a little more ready. And that forced rapid. iteration and development. And frankly, just grit around the idea and making it work. And I think, but for those experiences where I'll never forget we launched our first set of dashboards, and we were so proud because we had built these dashboards that you could customize them and put, I'm going to have a bar chart here and I can show whatever I want. And you pick your axes. And now I'm gonna going to have a line chart up here. And then I'll have this other table down here. And we made this message. massively configurable dashboard and we launched it to this is when we ran our first customer with only a pilot, and we went to go show them and we heard nothing for a couple of days, and we were waiting for the applause. Like, boy, that was the best dashboard we've ever seen. We kind of came back saying, guys, it's too complicated. Like

31:07when I go into other tools, I just go and there's just a dashboard there, and it already just has all the charts I need to look at. And I have to go in here and I gotta like. think about it like what time period do I want and which machines do I want and what are the metrics? And they're like, Can you make it just easy? Where I go there and they're all just and so we kind of said, Huh, that's really interesting feedback, and that makes total sense. But it's just an example of how getting the value of getting real feedback. And as you can imagine, we iterated on that new dashboard pretty darn.

31:36Unidentified speaker (UNKNOWN): Very quickly.

31:37Unidentified speaker (speaker_0): There's hundreds of other examples like that. but that's the value of real feedback.

31:40Matt Cohen: We see a lot of like YC startups and incubators going after the real world problems, right? The hard problems on the factory floor and the manufacturing facilities of different kinds of industries. But sometimes the things that you want to build as a founder don't align necessarily with what the company or the customers want. Keep it simple, stupid, sometimes is the best way to think about it. How hard is it, though? to build for what your customers want today before they know what they want tomorrow? Because you know what they need in years from now? You want to bring your customers forward into the future of where the technology can go while also servicing their needs at the very beginning. How do you kind of find that balance as a startup?

32:23Unidentified speaker (speaker_0): Boy, there's a lot to unpack with that question because I think that there are many examples where you get young bright minds coming into a new industry who build something absolutely game-changing and they've never spent a day in that industry before, but they come with just a totally fresh perspective and maybe a unique take on the technology that they're using. Manufacturing is a bit of a different animal of an industry. There are a lot of things that you could look at from afar in manufacturing and say, boy, that makes no sense. Why is that packet of paper traveling around with that piece of work around the shop floor? That's absurd. I mean, in today's day and age, we should just be digital and we should have these packets. But there's decades old rules that date back to FAA regulations where if you're making a part that goes on a plane, you need the operator themselves to physically sign a wet ink signature beside every part. And I mean, there's thousands of these little nuanced things. Like there's certain software that you're prescribed to use by the OEMs that sort of lives in the sandbox environment that can't integrate or touch other systems. And there's a million little things like that where if you were to come at an industry from the outside and just put your head down and build something that you think they're gonna want

33:43and that you think will solve the problem, but you don't actually understand the nature of the problem well enough, I think that can bite you. And that doesn't apply to all industries. And so I hope that's not a discouraging comment for someone who's building something without having spent time in the industry. But just manufacturing it is a bit of a different animal in that it's a high highly complex, highly regulated, in a lot of cases, environment. And I think what makes us unique in the way that we approach the problems. So we haven't talked a lot about my co-founder, Adam L spent decades in the consulting industry with working with manufacturers to solve these exact problems that Harmony solves today. And I obviously lived them and experienced them. And so what we built, when we put our heads down to build not the whole enchilada, but let's just say pieces of the whole enchilada to make sure that we had a complete solution, it was highly informed by many iterations of personal experience.

34:39Matt Cohen: I mean, you guys did something that was pretty unconventional rather than running on commodity tablets. You designed your own proprietary devices with embedded Linux, RFID, edge AI hardware, and a full stack to connect. control from bare metal to cloud. That is very unusual for a startup, especially in the SaaS land, to be doing that. Why was it important for you guys to own the hardware architecture?

34:59Unidentified speaker (speaker_0): There are some very good reasons. Trust me, I wanted to not own the hardware architecture. But when we actually took a step back, there was going to be too much compromise in what we were trying to achieve using off-the-shelf hardware. I mean, if you think about it, like our RFID, for example, I mean, we have patents today on how we use RFID and what we're using it for and some of the workflows that involve RFID. RFID is this amazing technology that's been around forever, right? You think about loss prevention and retail, right? Those little stickers that go on pretty much anything in a retail store. That's RFID. It's They use them at ski hills now. I mean, there's different applications. Nobody was using them in manufacturing. I mean, there was a couple of niche products that was using that. We're using them for kind of tool life management and tool identification, but they were very niche applications. But the way that we're using them around a shuffler to identify people, jobs, anything really. I mean, you could track your tools, your gauges, your fixtures, know where things are, know where things have been, understand the workflows, and then all that data layers into all the AI stuff. And so it's a really unique application. but yet it's not something that you're going to find in an IoT device typically, right? If you do, it's not going to have the 18 other things that we require to be able to do what we do. And another

36:13example would be our visual factory indicator light. We have a highly controllable indicator on the top of the device where we can set patterns, which it's kind of delivering its message. So Is it pulsing? Is it flicking? Is it solid? What colors is it showing? And it sounds so simple. You're probably thinking, ah, light. I mean, this doesn't sound like advanced technology here, but the way that we're able to control it and tie it in to deliver real-time efficiency or OEE, as we call it, indications as you're walking a shop floor to visually see what's going on, having that and having that not just be some little light, but a proper factory grade indicator that you could see from. across a hundred thousand square foot facility. I mean, that was a very specialized use case. And there's a ton of other things without getting too deep into the design, like how we load programs on machines, supporting hundreds of different machine types, the way that we monitor what's going on in this equipment and the TPU chip. We were using a Google product originally. Now we're using another much more advanced product,

37:13but having edge. AI processing capabilities where we can do high-speed inference right at the machine is game changing on what you could do. Because if you're having to stream constant data to the cloud across thousands of devices doing real-time inference, I mean, everyone understands token cost these days, or frankly, just GPU cost in general, because this isn't all language models. And it just becomes impractical to do anything. And if you do, you're not you're gonna lose the real-time nature. So this was all very strategic stuff. So we took the hardware. I'm glad we did.

37:44Matt Cohen: I'm glad it's paying off for you. Obviously, you recently announced a $10 million Series A round led by Bessemer. Can you tell us a little bit about that process? How you found the partner at Bessemer to lead, and why you chose them to be your Series A investor?

37:58Unidentified speaker (speaker_0): Yeah. So Bessemer had backed us from very early on through a schedule. scout investment and had just been great partners throughout. Given the initial small amount that was invested, the level of attention and focus that we were getting and really just spot-on kind of guidance as to what they're seeing was really amazing. And so I think from that perspective, we definitely had a great relationship with them. And when we met Kent, or early on, Kent Bennett, who's the partner at Bessemer, who we work with, it was an I would say an immediate connection. But it's one of those things where when you meet somebody who really understands and and gets what you're doing and why this is important, you sort of know it. And and this was definitely one of those moments where after the first meeting, you know, both Adam and I were as energized as the Bessemer team. And I think that was when we really kind of stepped back and said, was that it? Like, is this where we're going to do our Series A? Right. And so, and then from there on, it just kind of naturally evolved. And they've been really amazing partners. We're still in the early days, but like I said, we've worked with them for a long time and we know the team and have a lot of good things to say about that firm.

39:08Matt Cohen: That's amazing. Obviously, they believe in the mission and what you recently announced with how the context aware of where manufacturing AI process. product is different than what most general purpose models can do. Talk about how AI is only useful on the shop floor if it has both the operational context and the delivery mechanism. You're talking about how compute costs are obviously expensive if everything is shooting up to the cloud. You found a way to solve that. The that last meter problem, getting it right at the right time, at the right person, at the exact moment. of work you need it. How have you solved that perfectly for these manufacturing facilities that you work with?

39:45Unidentified speaker (speaker_0): Yeah, that's also a great question. And so if you think about the most rudimentary way to implement AI in a factory, it would be: let's put a laptop at every machine, every work center in this factory, and pull up Claude or Chat GPT and basically just say, here you go, guys. Like here, you've all got access to AI. now. And it could even be some trained version of Claude or ChatGPT for manufacturing, right? So let's just even go one step further in this rudimentary implementation. You got some trained LLM, right, that you can give all your operators access to. The problem in manufacturing is context. And by context, what I mean is, let's use the example I've made you my machinist example throughout this podcast here. So Matt math and machinists is punched into machine ABC running program, DEF, making part XYZ, right? So for you to get anything useful out of AI, you need to give it a lot of information, right? As anytime you want to get a good answer out of AI, it's all about prompting. In manufacturing, there's a ton of data that's happening thousands of times a second, but also just the static engineering data. that's driving that process. So it needs to know, okay, you're Matt. So let's start with that. So imagine you're prompting Claude. You're gonna say, okay, I'm Matt. I'm an operator in company XYZ. We make these types of parts. I'm currently working on machine

41:04DEF and I'm making part XYZ. Just so you know what part XYZ is, here's some PDF schematics, right? So you got gotta to now tell it what you're making, what the part is. Then you're gonna have to say, okay, and I'm running some programs on a machine, and here's the programs that are running. You'd have to then somehow get those programs on that laptop and then get them all loaded into your contacts. Then you gotta say, and I've made seven parts, and here's the results so far. Here's the 18 measurement points, and here's where things are coming. And I'm having so you gotta then somehow get that data digitally. It's probably written on a piece of paper in most factories, but let's say you had a digital. input. So you gotta go get that data, throw it into your context, and then you gotta say, okay, well, the last 10 times someone's made this part, here's what's happened. And then you could go and extract some information from your ERP reporting tool and say, okay, well, Jimmy made 10 good pieces and two bad ones. Bobby made 12 good pieces and no bad ones, but here's the program Jimmy was running, and here's the program Bobby was running. And Billy was on this machine and he made 18 good ones, but it was a different day, and that machine was in this maintenance state. Oh, but by the way, here's the maintenance state of my current machine. And the last time we ran maintenance on this, and here's all the data from the last three hours when I've been running from all the vibration sensors. And so as you could probably imagine, I could keep going with that example for a very long time. And

42:23the problem is context. And you would be sitting there. for hours prompting, feeding context data into the prompts of your LLM to actually get an output. And you could go right down to the exact make of the machine, when it was bought, what's the history. So when you ask Hal a question at the device and you say, hey, my tool's chattering, what's going on? All of that context is a foregone conclusion. That is provided, and frankly, the right context. is provided based on the question that you're asking. And you're going to get very informed information. It is a very powerful tool.

42:58Matt Cohen: Getting access to HAL on your tablet, on the shop floor, right in their hands?

43:02Unidentified speaker (speaker_0): Right at the front lines. And it's a completely new way to deliver AI tools. It's a co pilot It for your machine operators right up front.

43:11Matt Cohen: With all the context that you've built

43:12Unidentified speaker (UNKNOWN): around it.

43:13Unidentified speaker (speaker_0): And that's not it for HAL. I

43:14Unidentified speaker (UNKNOWN): mean,

43:14Unidentified speaker (speaker_0): there's actually HAL in the cloud as well, which is again, it's a different use case, but you can ask as management, like overall questions about what's going on in your shop. And you got access to all kinds of stuff like RFID locating data, ERP data across all your parts, machines, where's my inefficiencies today? All that stuff. But and while that's all great, what's most game changing is bringing that to those frontline operators. Right.

43:38Matt Cohen: That's incredible. I want to get your thoughts, David, on this sort of manufacturing industry of America, the reindustrialization, the reonshoring. We had Aiden from Eclipse on. And there's a huge urge to see the changes happening at the ground floor in American manufacturing to give our listeners a sense of just how big it is. It's a three $3 trillion dollar industry, 12.6 million people, and yet more than 90%. of those 300,000 manufacturers have fewer than 100 employees. So why is it so important to modernize that install base more strategically with like future factories rather than adding more people to these factories, which we are not able to do? And also the wave of robotics that are coming online as well can totally upend this industry. So what are you trying to do? And how do you see this all playing out?

44:29Unidentified speaker (speaker_0): So my belief is these all need to exist. I believe that you need the new autonomous factories to drive innovation. I think what these guys are doing is amazing, and I think it's good for everybody. But you gotta remember it's a three trillion dollar industry. Okay. So the greenfield autonomous factories that are coming up today, and you've probably seen all the headlines. just as much as I have, right? There's a lot of capital being raised backing these greenfield factories. That segment as a whole could get to 100 billion of revenue. Okay. And like that's a lot. That's a lot in for any venture scale industry. That's a lot of revenue. And there's 2.9 trillion left that's being made the old way, right? And that's not even inflation adjusted. Start throwing inflation on the three. 3 trillion, maybe it's still 3 trillion left by the time that autonomous factories have adopted it. And so I think it's important to accept that it's a great field for driving innovation, but there's a lot of parts being made in these brick and mortar mom and pop shops that aren't misplaced by being there. They're actually there for very good reason. If you were to go to an OEM, and say, Hey, I just opened up a autonomous factory and I got this amazing fleet of robots and we are making landing gear parts.

45:48And you're gonna call up Boeing and say, We're making landing gear parts. Okay. And we're gonna save you 20%. We're gonna sell that main landing gear to you for $500,000 instead of $600,000, $620,000, right? So they're gonna say, okay, interesting. Send me your portfolio of landing gear and the safety record of all those landing gear that you've made and show me your ability to manage 50 different outside vendors who have to do all the different regulated testing approvals and HVOF coatings and heat treating and 18 other processes that uh that used us go to through. Show me your partners and your networks and all this stuff. And then when you look at the economics of it, can you even get to 20% cheaper when only 10% of the the time and cost that goes into a part is actually the machining operation itself. The rest is actually managing all these for 50% of it is just the material cost. But then of the remaining, the rest of it is most of that goes to actually managing all these outside vendors. And so it's a tough proposition to say, hey, I'm going to go steal that landing gear that has been made in this factory for 20 years. because I got a more efficient factory. Boeing could not care less about whether they're going to save 10 or 20% if they're giving up one ounce of safety and risk in that part because I think everybody's seen what happens when you have incidents or accidents with and most of what we make in this country are critical parts. That's

47:08why they're made here. It's the key component of the machine, whether it's the landing gear and in that example I just gave. or the hot point

47:16Unidentified speaker (UNKNOWN): in a paper

47:17Unidentified speaker (speaker_0): deposition machine or whatever it might be that we're doing, it's the most critical component that they want to keep close because you're iterating on engineering revisions and you want to make sure you have secure supply of it. And so that's, I think, the big thing to understand is that there's a reason these parts are being made where they're being made today. And our mission is to help that base, that installed base. of companies out there. There's a lot of them. As you you quoted some numbers, there's a lot of them, and we're helping them get more efficient to defend themselves against this threat that is coming. They're getting pressure from autonomous factories.

47:50Matt Cohen: Obviously, there are headlines of how much funding is going into the hardware space right now and the manufacturing space with Travis Kalnik and Adams coming back in. Why do you think hardware is back in style after a decade of everyone? was a software investor and treating hardware as a liability? Why do you think this is now the place people want to be playing and focusing and investing in the full stack, right? Rather than components of it.

48:14Unidentified speaker (speaker_0): I mean, look, my favorite quote, and I've lived this, believe me, is hardware is hard. And I forget who said it, but lots of people have repeated it. I think, including Elon Musk and others. I mean, building anything physical is really hard. And doing that then at scale is even harder. And and I think that it's a real tangible moat in today's world where software has a lot softer barriers than it used to. And now that doesn't mean let's go have a piece of hardware just for the sake of having a moat. I mean, the hardware has to be crucial to what you're doing. But when it is and you can figure it out, the hardware is really a critical. piece of IP that is very hard to replicate. And remember, there's different hardware models and a lot of we're technically a I mean, we we don't really refer to it as hardware as a service, but we're, I guess, hardware enabled SaaS, I think is probably a good way to describe what we are. So I mean the hardware is an enabler for our SaaS product. And so again, different model than we're building widgets and selling widgets. So, you know, there's different ways you can

49:15sort of play in the hardware space.

49:17Matt Cohen: How would you say your opinion of technologists, software providers has changed after owning factories for a decade and now being a startup software creator founder for three years? What has surprised you most about the gap between like the operators and what they need on the factory floor versus what technologists and vendors actually deliver?

49:39Unidentified speaker (speaker_0): The biggest gap to me is along the lines of what I mentioned earlier around that adding technology, just adding it, can solve the problem in and of itself. And I think that if you were to talk to operators on a shop floor, and believe me, I talk to many, it's one of my favorite things actually to personally go and do harmony installs because you learn so much from talking to these folks. And they're very jaded by software. It's not

50:08Unidentified speaker (UNKNOWN): another one. That's

50:09Matt Cohen: what they always say.

50:10Unidentified speaker (speaker_0): Yeah, exactly. And we have to fight that coming in, right? Because that's the environment that we're walking into where it's like, all right, here we go again, another one of these kind of tools, huh? And but we explain to them obviously how we're different and they usually get it and understand. I mean, we're an empowerment. tool for those machinists. And I mean, in a lot of cases, these tools are put in place to monitor them, right? It's sort of like someone strapping an ankle bracelet on you. It doesn't really feel like something that's helping you, right? And and a lot of these tools are exactly that. It's a little box that gets installed inside the machine, sending some kind of telemetry data to the cloud with this magical idea that, okay, we're gonna, we're gonna measure. you, and that's gonna somehow make us more efficient as a factory. And so what something I like to say is when you're evaluating any tool and this predates Harmony, is think about how is this going to change anything that actually happens on the shop floor? And if the answer is it won't, that's probably your answer in terms of what kind of ROI and return you're gonna get. Because if widgets are gonna be made the exact same way tomorrow. that they're being made today, but you're gonna have some pretty dashboards, that's great and it's nice. And I bet you'll enjoy the dashboards, but don't expect an ROI from the product.

51:23Matt Cohen: Yeah, that's an amazing way of thinking about it. I love that. Before we wrap things up, we always ask our guests for their fast favorites. So, first off, your favorite podcast.

51:31Unidentified speaker (speaker_0): So I'm guessing I'm not allowed to say tank talks. You know,

51:34Matt Cohen: unfortunately not.

51:35Unidentified speaker (UNKNOWN): That name has been bad. I have

51:36Unidentified speaker (speaker_0): loved I have loved listening to your podcast. But I would actually say the older episodes of 20 VC, I really enjoy. Now it's kind of become a bit large cap focused, but in a lot of the earlier episodes, there are some really nice, in depth founder interviews that get into the early stage building stories of some really amazing companies. The Eleven Labs one is one that inspired me quite a bit. Matt is just an amazing operator. And frankly, we adopted some like the pod system and stuff from him. And and so it's it was just a really insightful for me. So I've enjoyed it. I

52:13Matt Cohen: it's a good one. I like that. Next is your favorite newsletter or blog.

52:16Unidentified speaker (speaker_0): So there's a blog or newsletter. I'm not even sure the difference these I think it's a newsletter called Breaking the Bottleneck. And it's a manufacturing newsletter and it's run by a guy named Adita. Aditha Ragupathy. And he used to work for Schematic Ventures. Really very bright guy, knows manufacturing really well. Met him early on in our journey. And I've really enjoyed following that.

52:40Unidentified speaker (UNKNOWN): I like that.

52:40Matt Cohen: I like that. Breaking the bottleneck. I'd say first the difference between newsletter and blog is newsletters Letters come out on a repeatable basis, probably. Blogs come out a little bit more infrequently, right? That's what I would say the difference. is. Favorite tech gadget.

52:53Unidentified speaker (speaker_0): So I am gonna go with my X Real One glasses. And I know make me sound like an Uber nerd, but I've I've really embraced them. It's not AR or VR. It's sort of like just a screen. And I use them as at the Farmborough Air Show a couple of weeks ago on the flight back. And it is just what do you

53:13Unidentified speaker (UNKNOWN): do with them? You bring your work.

53:15Unidentified speaker (speaker_0): Yeah, you load your shows on your phone, plug it in your phone, and you got a hundred and seventy inch screen, but they're just sunglasses. And so like my wife sitting beside me, you can turn to the side and have a normal conversation. The screen just goes away when you turn your head and then it comes back when you look forward.

53:31Unidentified speaker (UNKNOWN): Wow. And that's I gotta check those out.

53:33Matt Cohen: X Real. Okay, cool. X real.

53:35Unidentified speaker (UNKNOWN): X

53:35Matt Cohen: is your favorite new trend.

53:36Unidentified speaker (UNKNOWN): So

53:36Unidentified speaker (speaker_0): I'm gonna going to go with reindustrialization, and I know. that's going to sound

53:39Unidentified speaker (UNKNOWN): a

53:40Unidentified speaker (speaker_0): little bit cliche, but reindustrialization is real. And we're seeing it happening here. It is not a US concept. I mean, it's happening in Canada too. And I think it's the new world order right now where countries are really learning that while in the old world you could just rely on one another and you got your supply coming externally. When things don't work out as planned, you need to have security. And especially for critical areas like defense, but it's even broader than that because it could be very disruptive. I mean, we couldn't get hardware during yeah, during some of these times. So

54:16Matt Cohen: it's amazing. Yeah. How about your favorite book?

54:19Unidentified speaker (speaker_0): So I'm going to go with crossing the chasm. And that's for anyone building hardware, you have to read it. Yeah, the lessons learned in that book were invaluable to me. And yeah, we really follow that. Where you kind of pick your ICP really narrowly and really own it and broaden out. I mean, that was it was it was it's a great book. If you're gonna build anything physical, read it.

54:41Matt Cohen: Totally. And last but not least, your favorite life lesson.

54:44Unidentified speaker (speaker_0): Okay. This will be that everybody focuses on the lightning strike when something big happens, good or bad. but nobody ever actually thinks about, or it often gets overlooked, that the path that leads there. So it always looks like luck in the end when you see that lightning bolt strike. But you forget that person was for years and maybe decades climbing up a hill, waving a metal pole. And when lightning finally strikes, it always looks that way. But don't forget all the work that goes. into leading up to that lightning strike. That's

55:18Unidentified speaker (UNKNOWN): my

55:18Unidentified speaker (speaker_0): favorite analogy.

55:19Matt Cohen: Love it. Well, thanks so much for joining us in the tank today with David Caputo from Harmony.

55:24Unidentified speaker (speaker_0): Thank you.

55:25Unidentified speaker (UNKNOWN): Hey

55:26Matt Cohen: everyone. Thanks for tuning in to another episode of Tank Talks. We hope you found today's conversation as insightful as we did. If you're enjoying the show, we've got three quick things to ask of you. First, hit that subscribe button on your favorite podcast platform so you know. never miss an episode, whether that's Apple Podcasts, Spotify, Google Podcasts, or YouTube. Next, follow us and stay up to date on upcoming episodes and behind the scenes content on social media with Twitter, LinkedIn, and Instagram. And lastly, share the love. If you found value in today's episode, share with a friend or colleague collie who'd could benefit too. Your support helps us bring in more amazing guests and keeps the Tank Tots engine running. That's it for today. Until next time, keep disrupting and innovating.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights