Data Centers Are the New Highways
===
Eric: [00:00:00] 250 years ago, this country was 13 colonies on the edge of a continent. In 1869, two crews met in Utah and drove a golden spike, and America had a railroad from coast to coast. A farmer on the plains could sell to New York. In 1956, Eisenhower signed the law that built the interstates. A trucker out of Wisconsin could reach anywhere.
A factory could go where the land was cheap and the people good. Neither one handed anybody a living. They handed everybody a railroad. That's the American idea. It was never about giving something to everybody. It's about giving everybody the chance to be anything.
Naveen: I love that line. Can, can you say that again, Eric?
Eric: It's about giving everybody the chance to be anything. Anything.
Naveen: All right.
Eric: Love it. The data centers going up across this country right now are that kind of project, the railroad and the interstate of this generation, and the people who [00:01:00] gain the most won't be the ones who lay the track. They'll be the ones who ride it.
We're a niche manufacturer in Muskego, Wisconsin. We build cranes and aerial lifts. No software department, never much marketing material, and today we run software we wrote ourselves on tools that ride on those data centers. This episode is about what that looks like on a shop floor. Welcome to Engineered to Lift.
My name's Eric Niemi. I'm the vice president of engineering. I have here our CEO, Naveen Vinta.
Naveen: Welcome, everyone.
That was a nice intro, Eric. Appreciate it. So looking at railroads and the interstates as the [00:02:00] thing that came before for these data centers is a, is an excellent way to look at it.
Eric: So Naveen, w- we build machines people can't buy just anywhere else. Why has there never been much to show for it?
Naveen: So that is, that is a very good question.
So what we excel in is engineering. What we excel in building the machines, doing the calculations, reading the regulations, making sure the machine that goes out of the shop is completely, thoroughly ANSI tested and everything. But developing marketing material for it, and then showmanship has never been a primary focus because of the ROI on it, how much heavy lifting it takes to be able to do that portion of the program, and then what it can bring.
The traditional thought processes will do the work, and people who need it will come. But in this AI native world where everybody's attention is the new economy, attention is the new economy. For what we build, we need to be able to show how [00:03:00] we're doing it and then develop a marketing material for it.
Eric: Marketing material can't take a back seat anymore.
Naveen: Yeah, exactly. What we do is very, very important. That is the crucial, but marketing has to play equally important role to be able to display the work because somebody somewhere out there might be struggling to find this equipment, and then they might be doing things dangerously, putting people lives and then at risk.
So we wanna avoid that at all costs.
Eric: Correct.
Naveen: Marketing will help that.
Eric: One of the tools that we've learned to use is a macro, but what is a macro and what did it take off the engineering's plate?
Naveen: Yeah, I mean, that, that's a very good question. Macros have been there for a long time. As engineering team, we're focused on the fundamentals of doing finite element, element analysis, doing the calculations in the other parts of the program.
So macro is a Visual Basic script that can be turned into a button inside SOLIDWORKS. But to be able to learn that, that's a [00:04:00] different kind of skill set from engineering versus IT. Now, even though if you're experienced in IT, you can keep writing macros till the cows come home, but you need a full-time developer to be able to do that, especially in our engineer to order shop where we don't make similar machines.
Developing macro, each macro can be a slightly different from the last one we developed. A human to be able to write each one and do, the ROI is so very low. Now with the help of AI, so I generally use, I don't know if you wanna call it the SI, like the president renamed the term, right? So if... I generally use Claude, but you can substitute any, I don't know if you wanna call them the frontier lab models or they're like, I'm taking most of the inspiration from the Arlen podcast.
They're like commercial companies. You can substitute any LLM for Claude. If Claude does the heavy lifting of writing a macro, macros become so [00:05:00] expendable, like bullets. You can just write a macro, use it one time, and then just discard it. Previously, when you had to write a macro, you had to maintain it.
The IT person has to be dedicated for that one.
Eric: Right. Every time the software updates, the macro has to update, and- Yeah ... all the supporting documents have to update, and-
Naveen: Yeah.
Eric: Yeah ... update, update, update.
Naveen: Yeah.
Eric: It, it gets to be a never-ending story.
Naveen: Now, this is like a utility. You something do on a spreadsheet, you just write a macro, use it, and then discard it.
Eric: Yeah.
Naveen: Who cares? The next time you need a macro, we'll ask Claude to write one.
Eric: Right. Hey, Claude, fix that.
Naveen: Yep, exactly.
Eric: So then, uh, what we're saying is that, uh, Claude being the harnessing, it's- Yep ... manipulating all the softwares underneath it without engineering ever seeing it or anyone really ever seeing it.
Nobody likes the, uh, not being able to see something.
Naveen: Yeah.
Eric: So, uh, SOLIDWORKS in the background, uh, give us a little more information on that.
Naveen: So SOLIDWORKS in the background, the macro does the clicking. It's like a [00:06:00] human, what would they do in SOLIDWORKS? The repeated motions, the drudgery. When you have a macro, turn it into a button.
You can click that button, and then it can do 20 to 40 clicks that a human has to do every time. If they miss one of them, they have to go back, and then it becomes drudgery while you're in the-- Well, let's say when you're in the zone, designing and then thinking about the problem, thinking about the customer, this drudgery might kind of disrupt.
You need to use a different part of your brain to be able to do this rote work. A macro automates it, and then every time you click a button, you don't have to Intently remember the every sequence, it does all of it. As humans, engineers, you check the end result and then just bless it off. That's exactly reinforcement learning with human feedback.
So recently, Eric, I talked about, uh, SendCutSend. So we saw their interface, right?
Eric: Yeah.
Naveen: So-
Eric: It's pretty quick ... shout
Naveen: out to SendCutSend. Yeah.
Eric: Yeah. That's, it's pretty remarkable that they've been doing this a hot minute, but-
Naveen: [00:07:00] Yeah ...
Eric: it's, uh, you just give them the file and they, they give you back a quote. It's like, wow, that's-
Naveen: Yeah, they generate a quote in real time.
With our customer base, they're not very price sensitive. Couple tens of dollars, couple hundred dollars, depending on the price varies 10, 12%, they're okay with it, but they want speed. How quickly can we supply to them? Because if we don't have the right machine, if they have to stop the work, that is much, much more expensive than getting the right part.
Eric: That's the crux of the matter is, uh, you've designed something, now someone has to make it.
Naveen: Yeah.
Eric: And if you can't make it in-house, you gotta find a vendor that can do it, and typically has to be in short order.
Naveen: Yeah.
Eric: And it, it might be weeks down the road before you get to that portion of the design, and here you are behind the eight ball, and-
Naveen: Yep,
Eric: yep
the vendor says it's four to six weeks. And you go, "Uh, now we gotta find another vendor that can do it quicker." It's-
Naveen: Yeah. And then the volume also matters because we [00:08:00] need these one-off parts. The, our volume is not really that high. When your volume is not that high, the vendors you're interacting with, they don't re- really treat you as cream of the crop because we're gonna not get as much business from you.
Eric: Right.
Naveen: And then the most of the time if you want a replacement part is waiting for the quote. So hopefully others... I saw some others also taking inspiration from SendCutSend. They can immediately generate a quote, and then the part can be here by next Tuesday. So that, that, that is very beautiful. In popular culture, we always hear about Android, the gigafab Tesla is doing, SpaceX and everything.
Those, those are leading examples, but there's somebody like SendCutSend and others who are trying to revolutionize how we do things. I was hearing from our purchasing manager, Rob, yesterday. At SendCutSend, his issue was, "I cannot speak with anybody. If something goes wrong, there is no way to correct it." I [00:09:00] think that is the crux of the matter.
They're able to, like, generate that kind of speed because they eliminated all of those intermediate steps. All the
Eric: human factors go away.
Naveen: Yeah.
Eric: And not necessarily bad, but not necessarily good either.
Naveen: Yeah. But with that, they're able to serve a lot more people-
Eric: Right ... instead
Naveen: of...
Eric: Yeah. And, and a lot of the smaller companies are who they're serving.
Yeah. It's not the, not the big giga corporations. They have their own machine shops- Yeah ... and their own, uh, special vendors that do stuff just for them.
Naveen: Yep, yep. So I, I really like how they were able to, like, generate a quote on the fly once you upload a DXF or a STEP file, and then that, that was pretty neat.
Eric: So not just SOLIDWORKS, but what are some of the other things that we're working on, Naveen?
Naveen: Yeah. This, this might come as a huge list, but I'll touch different aspects of how we're doing, especially for a different small business owner who's listening. They're not big enough to be able to hire [00:10:00] somebody for that specific role, but we cannot find that unicorn that's gonna be so fractional split across these segments.
Some of the things that we're doing, we're exporting STEP files out of SOLIDWORKS, and I am using Blender to be able to generate all the marketing materials. There will be a catalog for each one of our machines that is... Claude is able to run via MCP Blender and then the things that we have come, we will, we will put a link in the show notes to one of our magazines.
That, that, that is pretty nifty. Previously, if you had to get that, that is at the level of somebody like a Ferrari or McLaren would be able to do that.
Eric: Right.
Naveen: Now, we can do it because the learning curve with Blender, that is all done by Claude. So previously, our time sheets-
Eric: Yeah ...
Naveen: that is as utilitarian as it can get.
We used to have a, like, physical manual punch card. Now, I developed an app with, uh, with the help of Claude. [00:11:00] Somebody... I know I'm going in the weeds. This is a little bit of... We use Supabase as the database. We use Railway to host it, and then Claude does the development. As Claude as an MCP to GitHub, it writes the code, and then it auto deploys to Railway, and then you can see it in production.
The whole CI/CD is streamlined. Now, folks can log in through their a- phone, and then it's also geo-located. When they're at the shop, they can log in their time, log out. That is also the courtesy of Claude.
Eric: That's a huge time saver, for sure. Yeah. That we would have to have someone in personnel or payroll- Yeah
looking at the time sheets, logging it into one big data sheet- Yeah ... and then processing it and writing the checks. Now, it's all automated now.
Naveen: Yeah. So it's automated to the point that Rob opens a page, everybody's hours appear against the right machine. He just clicks them, click Approve. When it comes to running payroll, Claude looks at the time sheet [00:12:00] and then auto updates the hours in our- PEO just works, and then everything happens.
I mean, we're leveraging this as much as we can. My cloud subscription comes at about four or 500 bucks, but I think we are getting four or five times the value of it.
Eric: Yeah. It's, it's been an amazing tool for sure.
Naveen: Yeah.
Eric: Even I'm using it to find odd information of, hey, what are the dimensions of the weld seam on a tube?
Naveen: Yep.
Eric: I mean, give me an idea. Yeah. It's not in the CAD-
Naveen: Yep ...
Eric: because it was never thought of at the time. Now we're, we're running into situations where, hey, we're having to grind that out.
Naveen: Yeah,
Eric: yeah. That costs a lot of time.
Naveen: Exactly.
Eric: So how do we fix that? Well, give me the dimensions of it. I can put it into the model and put it in the right place So we don't have to go back and-
Naveen: Yeah ...
Eric: etch it out or scratch it out of there and-
Naveen: I like your example yesterday where we were talking differences between DY and DX.
We can tell it to look at the regulations, give us a primer where you can [00:13:00] digest that information much more, and then you can help us understand, bring us all up to speed, because it did the heavy lifting. You can understand what it put out. If we look at it initially, we might not able to understand if it's saying- Yeah ...correct or not.
Eric: Yeah, a lot of it is buried in a standard that's a couple hundred pages long. You gotta sift it out and go, what is the correct section? Then, and it references back to this and back- Yeah ...to that, and it, it, it can get cumbersome. Let's, we're definitely looking at that.
Naveen: Yeah. It take out, takes out the drudgery, but also your experience is, like, so valuable here.
The output that it gives, the expertise you have, and then you're able to judge the output much, much, much better than anybody can.
Eric: But it, it also helps me explain it to those who have-
Naveen: Exactly ...
Eric: near zero experience with it, and even those that have been in the industry a while, but they haven't worked in that part of the standard.
So we gotta get people up to speed, and this is gonna be a great tool to do that, for sure.
Naveen: Yeah. The [00:14:00] initial worries of, like, so how things transpire initially, right? COVID, we had a different thing. Everybody has to lock down. Then afterwards, it was not that good. With AI, I think the shift is happening where initially it might take away the jobs, but now because of AI, there's more jobs that are being created.
Eric: Yeah, and it's, it's not that it's taking away jobs, it's taking away that drudgery- Exactly ...to allow our high eng- engineering group to move on to the next great thing.
Naveen: Yeah, everybody-
Eric: That p- ...to step
Naveen: up what they can
Eric: do ...new product development is important and advancement in technologies. We're, we're using some old tech that- Yeah, yeah ...hey, it's, it's, some of it's going away.
Naveen: Yeah.
Eric: I mean, vendors aren't-
Naveen: End of life, end
Eric: of life. Yeah. Yeah. Vendors aren't manufacturing those parts anymore. How do we adapt? Well, we need more engineering time. Well, where do you get it?
Naveen: Yeah.
Eric: Well, AI takes away the drudgery- Drudgery, exactly ...so now you have that time to devote to those new product development-
Naveen: Yeah ...
Eric: ideas or product sourcing.[00:15:00]
Naveen: Engineers with all their expertise have to, like, chime in on thus high-value ideas instead of, like, looking through the drudgery documents.
Eric: Yeah, for sure.
Naveen: One other example, I'm, I'm sure you have heard or other audience might have encountered this in different scenarios. When it comes to radiology Radiologists look at the X-rays and scans and everything, right?
When AI is able to do a much better job, initially the thought process is radiologists are scared for their job. But the way it turned out, we need more radiologists now because the drudgery is being taken off. Each radiologist has the capacity to see more patients. What is it? The, I'm sure, I'm, I don't know if you heard Jevons paradox.
When something becomes cheaper, people tend to use more of it.
Eric: Oh, sure.
Naveen: Yeah. Yeah. Think of it, agriculture, when we used to like till the land with cattle and then manually. Now [00:16:00] because of the machines, we're not using those machines on the same amount of land. We expanded so exponentially that it's, it's a joke.
Oh, why do we only do that? Now we have the capacity to do more. Exactly how the AI story will play out.
Eric: Yeah. It's, my grandfather was a farmer- Yeah ... and he used to till the land by horse-
Naveen: Yeah. Yeah ...
Eric: and plow. Imagine what the tractor was to him.
Naveen: Exactly.
Eric: Oh, wow. Yeah. You know, it, you put some gas in it and you go.
Naveen: Yeah.
Eric: Yeah. And as long as you keep putting the gas in it and check the oil and the water, you could plow f- 24 hours a day if you wanted to. The horses had to sleep. Exactly. Tractor doesn't have to sleep.
Naveen: Yeah. Yeah. So everybody work is going to go up I think, uh, I think- Those are some of the use cases. We can...
I'm also developing our own MCP to be able to work with Outlook and NetSuite, where we'll be interacting with Claude as a harness. It ca- it has visibility into [00:17:00] both of those with whoever is using it, their permissions. It's gonna be much, much more effective. So these are the things that we, we have been doing in-house.
I come from an IT background, so I'm able to, like, navigate this without, I don't know how deep is the pool. I'm able to navigate. If any small business owners out there want to have a, like, free consultation, this is the current state of affairs. We wanna bring in AI. What's a thoughtful way of doing it?
What is a low-hanging fruit that we can handle today? I'm open to have that conversation. Please reach out. I'll be more than happy to guide you in the right direction. So that is what we are doing in-house, Eric. Recently, I've come back from IMTS show, and then we've arranged a bunch of demos with somebody who's going to leverage AI to help manufacturing businesses.
Could you speak to s- what some of the folks that we have spoken to? Octonomy, right?
Eric: Well, we looked at Octonomy, and, uh, it looks pretty fruitful- Yeah ... uh, in the fact that it's more of a documentation [00:18:00] software and a control for the engineering group, that it looks, "Did you complete all these tasks? Where's the file located?"
And it can compile a document that can show every, where did you put the FEA? Where did you do the load analysis? Mm. Where did you put the hand calculation? Where did you put the, where are the drawings located for the project? Hey, did you go through and do an interference check? Did you do a thorough look at the GD&T?
Naveen: Oh, wow. Okay.
Eric: And that's some of the stuff that we're looking for that you can do all these things and hand sketch them on a napkin, but-
Naveen: Yeah ...
Eric: inevitably, the wind will blow away that sk-
Naveen: Yeah.
Eric: Exactly ... napkin, and then you're stuck. So if we can feed a system that has the tools in the background to do those calculations, now it's documented.
It's in one source for the project. Now you can go back at it. In six months, somebody wants something similar- Yeah. Yeah ... you can then look at those parts. [00:19:00] Okay, we, they're, they're usable. They have the right strength, the right materials. Let's grab that and go.
Naveen: Okay. That's excellent.
Eric: Um-
Naveen: I'm sure SOLIDWORKS itself has also been making strides like everybody else.
Everybody is in this, I don't know if we are past the FUD, is what they usually call it in technology, fear, uncertainty, and doubt with AI. I think everybody is got that FOMO, what if we don't do AI, we will be left behind.
Eric: Yeah.
Naveen: Can you speak to some of the efforts SOLIDWORKS itself is taking to bring AI natively?
Eric: Yeah. SOLIDWORKS is looking at a number of tools, ARA and a couple of other different subsets of AI packages that will be built into SOLIDWORKS moving forward. They're not quite there yet. Um, y- they can create a drawing, but there's some issues with the scaling. There's some issues with, hey, it didn't put all the dimensions in.
It doesn't automatically do GD&T.
Naveen: I see.
Eric: And those type of [00:20:00] things. It, it's got the basics. Okay. The information is there in SOLIDWORKS, but from, from my standpoint, it isn't quite there yet.
Naveen: Okay.
Eric: And we've gotta move forward quicker.
Naveen: Yeah.
Eric: Yeah. That we gotta get that drudgery out of the way, and how do you do that?
You look for other sources for AI that might be years ahead of where SOLIDWORKS is from the AI standpoint of the code in the background. So we're looking at Mech Agents is another one. We looked at, uh, another simulation software that was out there. We've looked at some for doing assembly instructions, manuals- Yeah.
That's right ... that kind of stuff. That's, that's the drudgery that we don't want our engineers tied up doing. Okay. Yes, engineering has all the information to build those products on the back end, but to take the engineering time to do that, it just- Is not, yeah ... the ROA is, ROI is not there.
Naveen: Yeah.
Eric: We gotta get away from doing that and automate [00:21:00] it or outsource it.
Outsourcing is Scant at best.
Naveen: Yeah.
Eric: So we want it the way we wanna see it, and we're gonna leverage AI to do it. Yeah. That's just it.
Naveen: So because of leveraging all this AI, my ... I mean, as a small business owner, my philosophy is I would rather have my employees relax and then have a little bit of a calm working environment, just have their thoughts slowly come to them, but we're gonna push the machines, like, as hard as we can, full throttle, see what they can do, because machines are not expensive.
We wanna save the employees. We want the hem- them to have the time to think, and then go through the workday with a lot of productivity and then ease.
Eric: Yeah, the, the machines have come a long way. There's, uh, supercomputers out there. Yeah. The price of those has come way down.
Naveen: Yeah.
Eric: I mean, even a, a desktop computer with serious, some 128 gig of RAM and video card [00:22:00] from the gods to make all this stuff happen, that's, that's
if that's the limitation, it's a pretty quick investment to go, or return on it to go, "Hey, it only cost us five, $10,000 to have a supercomputer that can do all this stuff on its own." You give it the softwares and away it goes. Now you don't have to- Yeah ... you don't even have to think about it too much.
Just provide it the information.
Naveen: Well, I think we talked about it comprehensive enough. Folks who listen to us, we came here to listen to manufacturing. Why are you also talking about AI? Because it has become that prevalent, and then it is taking everybody's game up. We are seeing it firsthand. That's why we wanna educate our listeners.
But also ask to our listeners if there's anybody out there who is leveraging some of these tools that we have not come across. When you see high ROI, please send them to us. We'll be thoroughly evaluating them for our purposes and then put recommendations out there. I think that concludes our episode.
Thank you for [00:23:00] tuning in. We'll be back with the next one.
Eric: Thanks for watching.
Naveen: Stay informed, stay protected
