Latest / AI in Manufacturing / AI Is Reshaping Industrial Software. Who Wins? : Michael Finocchiaro - Founder, ThreadMoat
Transcript
- speaker-0: Welcome to the AI in manufacturing podcast here on Industry4.tv. I'm your host, Guudzai Mandy Teresa. And my guest on this episode is Michael Finokiaro. I hope I'm pronouncing it right, Michael. speaker-1: Far better than I try I'd pronounce your name, so very good. speaker-0: Perfect, perfect. Yeah, so Michael is the you know founder and principal analyst at Thread Mode intelligence. so what we're going to do in this episode is really, you know, you know, try to step back and look at the bigger picture of you know how AI is reshaping the manufacturing technology stack. So we're going to you know explore you know what's changing across you know different industrial categories, MES, PLM, and you know just your industrial software broadly. And we're also going to look at where the new AI native players are starting to fit in. And more importantly for our audience, how manufacturers should know think about technology decisions in a market that's obviously, you know, becoming increasingly complex. So, Michael, I would like to welcome you to the pod podcast. speaker-1: Thank you. I'm happy to be here. I'm really proud of being a a guest here. So thank you because I speaker-0: Absolutely. So before we, you know, dive in, could you maybe introduce yourself to our audience, give us a a bit of about your background and tell us a bit more about Third Mod. speaker-1: Yes. my name is Michael Finquero. I've come more from the engineering and CAD side, traditionally, but I've taken took in a taken an interest in the manufacturing side since I did a little bit of work with thread with ThingWorks and and Kepware a few years back. I then I started working with a couple of startups and realized that there was this whole ecosystem of startups that are flying under the radar and making so much incredible change to the industry. on the engineering side but also on the manufacturing side. So I started doing a lot of research, talking to folks, talking to s to founders. And I went to Prove It in Dallas in twenty twenty f in February and met a ton of founders there too. I mean that was quite a eye opening experience, especially the presentations from inductive automation and litmus. In fact I was two weeks ago in Chicago for IMTS. That was my first time at that conference, which was of course all focused on manufacturing, although there were a lot of other engineering startups there like Colab. But I met Bill Byther there and I talked to CapVidea and a and a bunch of guys. and then I was at I C C so I met I I interviewed Colby and and and Carl, the the CEO and CTO of Inductive, as well as Jeff Winter. I finally got to inter interview Jeff. That was a lot of fun. I think you've had an interview with him before. and so I've been kind of diving into this space. I've discovered that it's a lot bigger than anybody realizes. The the problem is that the industry reports who claim to measure the market only talk about companies that have an ARR of ten million or more and most of these startups are most. Well won't either won't declare it or definitely flying under ten million. And yet I've discovered over forty billion dollars of VC going into these companies, over a hundred and fifty billion dollars of market valuation. It's actually huge. So I decided if nobody wants to be the person that's measuring it, I'll be the person that measures it. So I created Threadmoat, started the analytics, started interviewing as many founders as I could find, created a market report. Created an analytics database and a front end to that with now MCP support. I have nine categories which go from design to supply chain plus BIM and things that n neatly fit that I call knowledge engineering in those categories. There's two very strictly manufacturing categories factory f futures for execution for E S IoT and auto minted operations for CMMS and all the other kind of operational stuff. Or maintain X would fit, for example. who I met before they exited to Autodesk and I interviewed Nick Haas at IMTS as well. I'll add to say that ThreadMode has been taking a lot of my time and and it's been really exciting to to talk to founders and and to talk to the influencers in this area like you and understand more about the dynamics. Because I I think we can both say, and I think one of your questions later mentions this, that there's a sort of a collapse of all these things into one space. It used to be you could only think about your product design and you'd leave the manufacturing and the simulation to someone else and I don't think the designer of the future can do that. I think there's the engineer of the future will have to be a lot more holistic and these tools are helping us make sure that as soon as I start drawing a line on paper it's manufacturable from a machining point of view and from a factory point of view. So that I find that incredibly exciting and and and that's why that's why I'm here. speaker-0: Yeah, absolutely. I'm no really looking forward to unpacking more of that with you later on in our discussion. But for now I just want to start it by, you know, zooming out a little bit here. So I mean when I made you a couple of you know months back, I think it was two months back, I was really, you know, pretty mind blown by what you showed me, like the you know what you're attracting a pretty broad range of, you know, industrial software and AI you know companies. So you really get to see a lot of what's happening across the market. So My question is, when you look, you know, beyond like the product announcements that we see every day and the usual AI hype, what are the biggest shifts you are actually seeing in industrial AI right now? speaker-1: Well I'd say there's three of them and none of them is really a product launch. I think that the capital is split between Frontier Labs on one side and and and on the other, industrial AI still needs to show revenue in ROI before everybody writes a check. So there's a bit of a a difficulty there. I think that Adams are back. In fact I launched the physical AI side of Threadmoat yesterday, because There's a whole new generation of software companies like Bright Machines that are doing both software and hardware in order to do autonomous manufacturing or some of the lights out, stuff that Jensen was sort of teasing last year at the GTC in terms of the factories of the future. and so I think those are major shifts. There's also the adoption of op open USD by just about everybody in order to do digital twin of the digital factory. and another one I think is that for thirty years we were worried about integrations and in fact that doesn't matter anymore. With agents, the integration doesn't matter. It's basically the agent does all the integration. The problem is context and intent. Like how do you get all the context of the data, which is of course the entire battle right now on the shop floor between the devices and the intelligent on top of them is getting that context and then the design intent on the engineering side of what was it actually supposed to do? Once the agents can figure that out, then a lot of these things start to roll a lot. speaker-0: Yeah, absolutely. I mean it was super interesting what you mentioned there about, you know, bright machines, you know, going sort of like into the hardware, you know, space. And it that that kind of like reminded me of also a you know, a startup that Walker Reynolds interviewed for the past two weeks. I think they're also building their own hardware and then they're kind of like building a software stack on top of that. I think they are disrupting the PLC. I think it was neuron industries. Neuron industries. Yeah, so the yeah yeah, they they're building an AI native PLC and kind of like you know building some so i i i it's interesting that we're starting to see you know companies of like wanting to own that, you know, whole experience from hardware up to software. So it'll be interesting to see if that becomes an a trend, you know? speaker-1: Well, I can also say that the my one of my big takeaways, 'cause I was in Chicago for IMTS as well, was that talking to people that had been there in twenty twenty four, it was just as big, you know, all four buildings of the McCormick Center completely jam packed with machines and and robots and people running around. There was no mention of AI two years ago. And this year that you couldn't you know, you couldn't throw a ball without hitting a sign that said AI on it. It was Absolutely everywhere. And it was in every single hardware manufacturer. Whether they were CNC guys or they were building new milling machines or it was additive printing or it was the robots. It was everywhere. So this yeah, this integration of of physical AI where we're using AI to inform the physics and inform the the actual action of the bits to atoms, absolutely huge. I think that A sixteen Z in September also said software is the star of the show, but with AI Hardware is now taking center stage. I think there's like ninety trillion dollars of investment that's needed by twenty twenty f twenty forty with more robots and and more parts. So it's gonna be really the ne the wave of the future is gonna be coming there. speaker-0: Yeah, absolutely. So I mean, as you know, sort of like alluded to, you know, earlier, like manufacturing has r historically, you know, relied on, you know, these relatively well, you know, defined systems where you've got your MES, you've got your TLM, ERP historians, CADA and things like that. But you know, today we're seeing new categories, you know, for example your composable MES, manufacturing data platforms, AI core pilot and things like that. I I wanna get a sense of how you see you know the manufacturing software software stack you know evolving and and more importantly, what I'd like to you know no hear from is are we actually seeing new layers being added to this you know traditional stack or really it's just an you know a a a you know a rebuilding of existing categories? speaker-1: I I believe we are. I believe that I alluded to it before that the categories that used to be very separate of design, simulate, build are all blending together because we're the agents can basically take over a lot of the design for manufacturing, design for simulation, design for support, design for service, all those DFX things. I believe the stack and the stack is Probably more useful than the debate. The PLM's already being rebuilt. It's a sixty billion dollar market that was architected in the nineties and up to now hasn't really moved. And suddenly it's exploding with all these new features. They have, I don't know, fifteen, twenty startups doing model-based system engineering, and another fifteen I can find that are doing just requirements management or work instruction. So that whole stack is is splitting out and it's it's we'll talk in a minute about the the incumbents. I think that the AI startups are attacking every single level of that lifecycle, whether it's in the engineering side I just mentioned. But on the manufacturing side, I think that we still have the same ISA ninety five levels, but that we're just getting f more and more ways of integrating them in interesting ways. companies like T D Engine that are revolutionizing the way we visualize the data because they've started as a time database and now they've put all these analytics on top of it. And Jeff Tao is probably one of my favorite people to interview because he's like the most enthusiastic person that that I probably know. but you're also seeing companies like well I I a friend of mine just came out of stealth IQ brain and he wants to sit on top of PLM MES and R T E R P as an orchestration layer to make sure that the bomb is coherent across all those stacks, which we couldn't do before agent AI at all. You also have companies like Hilos Labs that are putting a a product graph To replace the CAD file. So we're just moving the geometry around rather than CAD files. And then I mentioned OpenUSD or we could even say QIF as a way of ensuring that all these systems can communicate with the exact same files without having to pay exports and imports and you know reconversions where we lose fidelity and we lose context. So I'm not sure we're adding new layers. But we're certainly seeing a lot more options for each layer and then each company can recompose the stack based on their specific needs. The the kinds of equipment they have. now that basically all of the high bites and the litmuses and all the the tulips are talking MQTT and OBC Way. So there's really what you have on the shop floor is is not that it's not gonna limit you because all of these guys are talking both those protocols, Arlen and Thomas of course at did a great job of of evangelizing them. And that's what I my takeaway was from the ICC, the inductive automation conference last week, was just this composability and this contextualization is becoming pretty universal. And so I think it gives customers a lot more choice to find the solution that's really tailored to their specific manufacturing needs. And they don't rather than having to take a very brittle legacy solution and trying to modify that to fit the their needs. It can kind of go the other way, right? speaker-0: Yeah. Okay. Yeah, I mean maybe before we go on to you know talk about, you know, the you know AI native tech tops, you versus the incumbents, I wanna get a sense on which parts of the stack do you do you think are more most vulnerable to disruption, you know, right now? Where is speaker-1: I think the the insight part because that's where AI c can do it so much faster than anybody else. I think the I mean in terms of the connect and collect, that's pretty rote now. I mean everybody is doing the connect and collect, the the visualization, but I think it's really that insight layer. And then what are we going to connect that insight layer to? Is it I don't think I think OE would become less of the the the benchmark and there would be other ways of looking at Well, it's not just efficiency, it's really how many things am I putting out there, how much energy am I saving, also, because I think that that's another big consideration. it's gonna be on how much innovation can I push through the factory huh much faster from my in my my my engineering team. and again, I I guess it'll also be different whether I'm doing discrete or process or batch or continuous, because those Those move at different velocities, right? I think v discrete's moving at a much faster velocity than those other two. Although I am seeing a lot of startups coming into the process industries trying to revolutionize it. So I I I think it's more of that insight layer, but it probably will will filter down into the other ones. It's just well, I I did find one interesting solution too. Maybe you know a Albert of Fluxian solution systems, have you talked to him? So he he said, Well, the problem is that you you have these older machines on the on the shop floor and you it costs a lot of money to pull a machine out and put it back in again. So what he does is he just goes into that machine and and rips out all the stuff that was inside the machine and puts in a new new controllers, new AI edge based stuff so that the old machine can still work, but with using the new technology, which means that We're s not replacing the the old rusty machines on the floor. We're just re engineering them and then re upping them so they can talk to this new world of again, these layers of insight on top that are gonna help us move a lot faster and do a lot more. speaker-0: Yeah, absolutely. speaker-1: And if that made sense, but that that's sort of how I see it. speaker-0: Yeah, absolutely. Yeah, I mean, you know, talking about, you know, the you know, old systems or old machines in the in the in the shop floor, I think it also applies pretty much to sort of like you know how how how easy it is for for startups to really, you know, penetrate, you know, the the that environment. To use an example, I think you mentioned T D Engine. I think they are really, you know, branding themselves around this idea of being a replacement of, you know, OSI Pi, right? Or Aviva Pi, right? Exactly. And and and and and I guess w what I wanna kind of like get a sense on is is what advanta what advantage would you know a startup have over a you know an established you know you know company or where do you see the established vendors having a bigger advantage you know than you know the this the AI startups. speaker-1: Well I'd say the startups win where the incumbents architecture is the problem. Ziemens PGC, DESO, Rockwell Automation, they're all sitting on architectures that have built before AI even existed. I mean they they these these things are built back in the the origins of them are in the back in the eighties and nineties. So they haven't really moved. They've they've tried to evolve and for with varying results, they've sort of tried to catch up to this AI wave that passed them moving back in twenty twenty two. the thing is their data models are built before AI existed. So their data models are kind of rooted in this old way of doing things with the V and the and and all that stuff. a the the startups can just move at light speed, right? You you don't need six months to two years to put it together your MES system. You can do it in Less than thirty days. You could do it in thirty minutes of if you believe the presentations that prove it, right? I mean they they were Litmus was showing what, three factories in in three minutes that they put together, which was just crazy. so I think that the what they could do is a level of speed and commitment that the big vendors can't match. There's no way that they can go as fast because they don't have the startups the the incumbents have too much a technical debt and and just this legacy weight behind them. the incumbents win, however, where data and regulations are important, right? Where they they already have the relationship with all the the regulatory agencies and they've already implied imported sorry and implemented all the things they needed for that. And so with and they also have the proprietary data formats. so I think that they they'll still be they'll still be it it let me put it a different way. Like someone was asking me this just yesterday and I like, well, these the you you saw the the Louis Valentine, you know, like three, four weeks ago he hooked up Astra, I think it was to ion shape or inventor, and he built a jet engine just with one prompt and everybody's like, my god, you know, the AI moment is here. The open But I don't it did kind of freak everybody out, but I don't think that's industrial. I don't think A SOLIDWORKS user that loves using SOLIDWORKS every day is tomorrow gonna start using Claude Code or or Codecs. I think they still want to use SOLIDWORKS. There's still this weight, this hubris that we have because we're humans and we have software we like to use. So there will be disruption, but there's also hey, but I've always used it this way and I still want to use it this way. There'll still be resistance, so there'll still be a place. However, there is this idea of velocity. I Interviewed Bill Byther both at Priv Prove It In at IMTS. Some he was talking about how how quickly he could put in his new his new factory operating system in in in days rather than months. Mind Device was talking about doing it in two weeks, whereas c competitor had asked for a hundred thousand and and a a year to put it in. I got the same from several other people I talked to. So I think that that's A question of velocity. and there's also well, there's also things like Tulip whereas they the Tulip has put some vibe coding in so they can easily just write up as a prompt to generate all the pr the the the front line screens that they need for the workers, which is a huge change over having to actually code all that stuff and then test it out. It's it's just crazy. The I think that It's a whole new paradigm. it's gonna take a while to completely come to fruition. But there's gonna be things are certainly gonna move, I think maybe more on the user interface and before the data models move. But even that's gonna shift too, because eventually we're gonna have large models that are trained on that specific data for that specific industry with those specific data models. So Even the data models will not be a barrier to AI being able to take all the rote tasks out and leave humans to do the creative stuff. I don't that was sort of a roundabout answer, I apologize. I hope speaker-0: Yeah, absolutely. I mean I'm i I I guess that's that's that's you know also the the open question, right? You know, clearly, you know, as you mentioned, obviously the user interface, having the AI startups, you know, sort of sit on top of the existing, maybe using the existing systems as systems of record and kind of like, you know, building an agentic, you know, you know, you know, operational layer on top of that. But again, to your point, the idea that LLMs are going to become, you know, powerful, but they would still need access to that proprietary data. For example, if you've got a company that has been building some some workflows, models on Aviva Pi for for the past, you know, I don't know how many years, that is a massive amount of domain expertise locked in there. Is is you know, Aviva Aviva going to be, you know, willing to just open that up for, you know, training and things like that. It's kind of like, you know, a tricky thing to to to to to see where it goes. speaker-1: And yet they don't have any choice because there's also the the the silver exodus, right? You have a whole generation of people going into retirement and all that knowledge and all that domain expertise is walking out the door. And so the manufacturers are demanding that the the vendors open up because they need to capture the this information. Otherwise they're just not gonna be able to make that thing they could make yesterday because, you know, the guy that did it just left yesterday and Left no notes and nobody knows. So th there's not that that's where AI is really gonna shine, is is mining the that that all that context and all that intent so that the next generation of of agents and the next generation of people on the shop floor will will be able to continue to build the same innovation innovative products. I mean that that's part of the whole reindustrialization thing in the US, right? Is how do we How do we do that? How do we keep going without having to outsource everything to another continent which had all it the disadvantages that came with that? speaker-0: Yeah, hundred percent. Yeah, that's actually, you know, a a a good point, right? Manufacturers don't really have a choice, you know, with all, you know, the you know, knowledgeable people retiring, so they're gonna be forced to really try and capture that you know, tribal knowledge somehow, some way. You know. speaker-1: And and and that's a problem both for you on the manufacturing side but on the engineering side as well. Mm-hmm. In the engineering side it's all that stuff that's captured in F T and A and all these notes. And so there's a whole generation of startups collab being one of them, but there's a I've I've got it you know, at least fifty on the database that are going after that problem. How do I get all that design intent and make sure that it carries forward? So if that those people go away, I I don't lose all that innovation. And so it's really it's becoming one of the big drivers I think behind investment and and one of the things that keeps factory owners and and people that build stuff up at night is how do I keep that stuff in house? and and how do I keep humans interested? Because also I think a question I ask on my podcast to the stuff founders is what do the humans kids need to do to not be AI'd out of a job? And the answer usually is, well, the fundamentals, because computers LLMs don't understand physics, right? They're not physicians. They can look at a math equation and then look at other examples of solutions of it and then even reverse engineer Lavier Stokes without actually solving it. But at the end of the day you have to actually solve it f from front to back, not from back to front. Otherwise the physics doesn't work. so I think that there's there'll always be need for the human in order and because there's things that the machine will never be able to do. speaker-0: Yeah, absolutely. Yeah, I mean obviously, you know, AI represents a significant opportunity for, you know, startups and even the end users and themselves and also the incumbents. And so obviously what that has done is is is created this this rush of sort of like you know, AI cold rush. And so almost every software company now, industrial software company now has got an AI story, whether it's a new startup or, you know, it's a you know an existing, you know. vendor or things like that. So that kind of like you know creates some noise which becomes problematic, specifically when I'm talking up to, you know, manufacturing leaders. So from your perspective, how should you know manufacturing leaders or, you know, engineering leaders for that matter, you know, look, you know, for when when it's like evaluating whether an industrial product is, you know, is actually quite a meaningful technical and or business differentiation, how should they go about, you know, doing that, you know, evaluation to sift, you know, the the signal from the noise. speaker-1: I think they need to ask whether it's just a wrapper. Because Frontier models are really commoditizing, right? They're just a thin layer and everybody's using the same models. Everybody's using Astra. Everybody's using GPT five point one or or six now. and so I think it's more it's more f the defensibility comes from proprietary training data from a workflow integration that's deep enough that if you have to switch to something else it's painful, which gives you that kind of mood. And then the regulatory readiness, one of the most important ones. So I've got a lot of startups that are focused just on one industry because that way they understand all the regulations and they can really focus on one vertical as opposed to a horizontal solution where you have to do it all. if I look at one of the engineering examples, I've talked to Greg Mark at Backflip AI, who was previously at Mark Forge, one of the big 3D printing things companies that was acquired. And he said that a wrapper around a frontier model is easy to do. What's awesome is when you've got a foundation model that's trained from scratch to understand what CAD does. So we can actually the foundational model can do the r work rather than the just being a co pilot or or a a chat bot. And I think that what we're seeing so there there's also in my recent post I had one that went sort of vir viral where I explained where I see four domains of of AI impacting industrial software. There's engineering AI, so how do we change design and design for manufacturing and PLM to be AI native and able to use all that advantage with human in the loop, with the deterministic outcomes because we put in the guardrails. You have industrial AI, which is more of our topic, right? When how do we design factories and design processes so that these are also u using those advantages without creating safety issues or or things like that because of you know the probabilistic nature of AI. There's also physics, AI with an S, which is bringing real physics into both manufacturing for machining. Like how how do I calculate all those all the physics around bit speed and what kind of material is ki impacting which other kind of material, how do I remove everything? How do we make sure the acceleration and deceleration of the bit It's just right so I d I get a maximum lifetime out of the bit. These are just insane amounts of physics behind that. And all those physics now are part of these new AI models that a whole new generation of CNC software companies like Limitless and Manukai and and Lambda are Lambda function are coming out with. so I think that and then the last one is physical AI, which was we talked at the top, the the bits to atoms where you're actually using all that to create hardware itself and then using the robots to build the robots kind of thing. so I think that the the the I in thread mode I I score the each of the start ups that I look at with seven v dimensions in order to show the differentiation. I use market opportunity, team execution, technical differentiation, funding of issue growth a metrics industry impact and could have been mode so I can kind of get a better idea of of how they they sit relative to to the incumbents. And I think that as the the the proof point of course is how how fast the plant runs at the end. If if you take there's a lot of companies that already have adopted the the UNS architecture, right? And they're already seeing insane amount of gains in terms of efficiency and and velocity. And better safety and better throughput and all that stuff. I mean you talk to those guys more than I do, but it's pretty insane that the amount of gains that we've seen and those can only really be done by by the startups, right? The incumbents are are not capable of moving at at this that same kind of velocity. speaker-0: Absolutely. I mean I I think kind of like as you mentioned that you know a lot of sort of like you know companies to to watch out for those who are really more mainly providing access to a general purpose, you know, foundational model, sort of like acting as a as a rapper, so to speak, on around the whether it you know Chat G B T or it's it's Cloud or all those no foundational models. So but that also in a way introduces a a risk, you know, for for for for for for companies. Companies who may today build the build their workflows or build their knowledge inside these, you know, you know, tools. And as the as the blast radius of of of of the of the foundational models LM sort of like increases, it will obviously start to, you know, kind of like you know, eat up those you know, startups as it were. And the situation there might be that some vendors might find themselves locked into those particular software and not being able to to move and you know that to utilize the capabilities of these, you know, you know, foundational models. So, you know, as far as making that decision of whether to lock yourself in or not to lock yourself in, how should you know, you know, you know manufacturing leaders think about technology as they, you know, adopt What should they, you know, sort of like prioritize to make sure that they don't, you know, find themselves locked into a platform? speaker-1: Well I think there's several aspects there, Lockin. is it just file format? Because we already have standards like QIF, step, JTU and OpenUSD. It's more of an incentive. I mean the the the incumbents profit because they lock the the customers in and every integration they build sort of f preserves their data gravity because they c the data still stays in their system. The buyer's defense is structural. They need to own the data model by against an API rather than a stack. And be suspicious of open when it's coming from a platform vendor because open could just mean, well, I you're I'm still keeping your data, I'm just allowing you once in a while to use it somewhere else. and open standards of course can also be be acquired. I mean vendors profit from lock in and every integration is preserving their data gravity, but there's these integrations are designed and those integrations are designed to to maintain that lock in. And so if you're bidding against an API. Like that's what CESMI is pitching with I three X, right? Is building us an API rather than in a stack. that gives you a lot more flexibility and and reduces that risk of of of lock in. I also went to a baseball game with Bob Van der Kohlin ThreadCloud. I don't know if you met Bob yet. He flew over from New Zealand right yeah, he's a great guy. Great f fun at a baseball game too. Thanks, Bob. you can also easily get bought, right? Price goes go up and open standards become closed standards. So if you're betting if you're b if you're betting more on APIs, it's gonna be a lot easier. I think OT is already betting on those standards. I mean that's why MQTT and OPC Way are now are there. and I think that a lot of the better solutions are also going with more tool agnostic design. Estari pitches that a lot for being not vendor locked. Collab has this sort of coexistence strategy, Because they sit on top of the DASO stack and the PTC stack and the Zeman stack. So you know it's it's it's an interesting problem. I it's a question of of building base on on APIs in order to keep your your leverage. speaker-0: Yeah, absolutely. All right. I think one, you know, topic that I want to you know spend some time on here with you before we, you know, you know, close the session is the MES, because MES is a you know particularly interesting, you know, you know, topic for me. I mean the category is is now being approached from you know many different levels directions, right? how do you see you know the role of MES evolving, right, as as obviously these no AI systems, data platforms and these composable architectures become more common. speaker-1: well it it it MS yes definitely survives, but it's not necessarily the center because the on the outcome needs to be right every time and the probabilistic stuff is not gonna work. That's really what MES is for, right? So that's why it has to survive because it is for the one source of truth in exactly what gets built and where. but the intelligence is moving up and out, right? We're getting more of that AI pulling in the context for for the MES, providing the context for that. And like I was I interviewed Craig, my friend Craig Scott Afuse at IMT at ICC and he says the he he had Claude and Cursor write an entire MESI ninety five over a weekend. So it's like not even rocket science, right? You can basically bang these things out. w when I interviewed Jeff Winter, he said the ME he I think he accurate said that MES is an anchor system, right? It yeah, it doesn't negotiate semantics. So it really d it remains a an anchor. The thing is well how do we how do we build around that? If you talk to Rise, Kurt will say he's MES but he's actually more of a mom because they're more of everything around that manufacturing data as a hub to control other operations and not just the execution. that's sort of w the c a camp of of startups that talks about MES staying. there's another camp like the that I would say that it gets unbundled, sort of like the way that Tulip Nate Nathan Lender talks about composability is not modules or configuration. It's really just this software, the the way that you build the the front frontline screens. the when I talked to Bill Byther of Machine Metrics, he was talking about how the MES, they don't really have the real time data. So he that's why he built his operating system on top of the real time data rather than historical data. Litmus talks about evolving where all of the AI data apps will run and having that foundation. and the other guy I interviewed at ICC was Aaron Simley of HighBite, where he sort of discourages the partial use of MES. He really wants to make sure that we're the data is flowing correctly in and out and not being captured at any one particular point. I think that the ISO stack survives speaker-0: So just j just to in interrupt you, wha what does partial use of MES mean in this context? speaker-1: Well he was saying that I I think that he they that some people are trying to decompose MES where they they're using high bike for a piece of it and th but not not the entire thing and he's saying that he it's better that you leave MES alone but then you're you make sure the data in and out that the contextualization is happening elsewhere. That you're not trying to recompose the data while as you're doing the execution. Regardless, I think that ISO ninety five is the right stack and and it's the the right way to look at it and it's the right way we've built UNS anyway. and I think that that's that's really the key. When I talked to Carl Gould, the CTO of Inductive, he also talked about the that the hardest unsolved problem so far is migrating legacy, MES and SCADA to the new universe and that's gonna be the next frontier, I think. But that's where AI is gonna be really strong. But that's really like the entrenched area is that legacy scatter, legacy MES is really gonna be the the front the next frontier to really bust for for the AI startups. speaker-0: Yeah, absolutely. And I think it's a it's a very again, MES is it's a very complicated one because even before the AI, you know, revolution for lack of a better word, i it's it's something that has always been on the card on whether you build or buy. So now with all these, you know, coding tools, it's interesting to see how the dynamic is going to, you know, play out. Where I mean, companies were building E S when it was expensive to do so, right? And now it's it's It's almost, you know, gone down to zero. So speaker-1: Like Craig talked about, right? It was just over a weekend. speaker-0: Exactly. speaker-1: Well, but you still need the determinism. It's still gotta be deterministic and the machines have gotta run and they've gotta have the right pieces at the right pl workstation at the right time. And if it's not deterministic it's just not gonna work. So the guardrails and the human intuition on what actually works regardless of what it says on paper, that's still gonna be important. speaker-0: Yep. Hundred percent. So yeah, I mean just kind of like to to to to to sum it up a bit for us and kind of like give us a more of a an an outlook or forecast, or rather maybe a prediction for that matter, where what do you think will be fundamentally different about, you know, the manufacturing software landscape, say five years from now. And I I often ask people to make a prediction here on the podcast, but I find that sometimes five years may be Too long that you know, you know, things are changing right now. So give us speaker-1: Yeah. speaker-0: a a a rough estimate of no how soon you think things are going to change and w what what what's what's going to change in your view. speaker-1: Well, I think that the humans stop being the integration layer. I think the agents do all the integration. I think that the it the system keeps the context, but the human supplies judgment at all the decision points because the AIs do not have judgment. institutional memory, as we already talked about, stops retiring with the people when they leave. It stays inside the systems, including failed ideas, right? Because we need to learn from our failures, not just from for our successes. I think that autonomous design probably isn't gonna happen in our generation. there are always gonna be y an use need for a human to be in there somewhere. the realistic path is bounded with specialized agents that report a confidence level and then an engineer who wants to to learn how to ask for that and and use it in their design. So I think that you know, like five years ago LMs were not able to do any reasoning at all. Five years from now probably they'll be able to reason. So a lot of the like almost all the road tasks will be done by agents. It'll be the humans will be there too for the judgment calls, for anything that requires creativity and of course all the the deterministic to guarantee the get deterministic outcomes. I think of course, as I already said, it c it will capture the contextual Stuff and and the failed ideas so that these things will be able to ensure we don't make those failures in the future. I think that there's also the this multi-agent can cognitive platforms with the knowledge graphs is gonna become more and more important. I think that's why the the startups that are talking about knowledge graphs are are are growing and and getting a lot of investment because suddenly that's the way that agents think and that's the way that that you're going to improve your your your determinism by having a an event mesh and an edge fog cloud distribution. I think you're gonna see that coming in three to five years. OpenUSD will continue as a vendor neutral exchange format. QIF hopefully will take a bit more place too as a better way to move the 3D data into the factory floor without it needing to have a seat of a a an incumbent vendor. And then I think that there will also be multi agent CAD CAM agents Rescheduling C and C's and things like that. There'll be neurosurrogates and CAE and quantum solvers. I think that the whole there'll be a lot of new stuff that will come and it will be pretty mind blowing. it'll just be a question of how fast we apply it and how safely we can do that. but I think yeah, definitely if I was gonna do the magic ball, I think we'll have this multi agent CAD, we'll have CAM agents doing a lot of the C and C stuff for us. and then some neural surrogates and and quantum solvers because that that's definitely going to I think we're gonna break the quantum barrier within six five, six years and that's gonna be a big change as well. speaker-0: Yeah, absolutely. Great stuff. So, Michael, you know, thank you so much. But before I let you go, I kinda just want to give you an opportunity here to, you know, let us know how best, you know, people can engage with you, what are your services you have on offer. I know you, you know, you s as I mentioned, you know, at the start of the session, you showed me this data set that was really mind blowing. Can you just give us a a sense of how people can engage you and what you have on offer? speaker-1: Absolutely. So the easiest way is through LinkedIn. I've got thirty thousand followers on LinkedIn, Infinicero. I think you're gonna put it in the in the lower third. You'll have the spelling of my name. threadmo.com is where I park all my commercial stuff, all the these analytics on the market, and the market report. I also have a website called demystifying PLM dot com where I talk about the history of CAD and how we got here in terms of PLM and CAD and and manufacturing. and and I'll be at A sixteen Z Tech Week next week in San Francisco, down in LA the week after that, so October twelve to fourteen, I'll be at Barcelona for the tech talk. I'll also for the Europeans I'll be up at Slush and Helsinki and the in the sauna. so that'll be fun in November. And then I was invited to be on the board of of prove it. So I'll see you also in Dallas 'cause I in February. speaker-0: Michael, thank you so much for your time here and you know sharing your insights with our audience. I really appreciate it. speaker-1: I really appreciate it too. And I've I've it's exciting to see people like you pushing these great ideas and and educating everybody. So I really appreciate the work you're doing too.