WEBINAR: DON’T GET LEFT BEHIND IN 2026 A PLM Insider’s Guide to Thriving in the Age of Agentic Al and Integration

Michael Finocchiaro on 30 Years of PLM and the AI Startup Boom
The Integrate Intelligently Podcast — with Jeff Brickler (CADTALK) and Michael Finocchiaro (PLM Demystified)

Michael Finocchiaro — a 30-year veteran of CAD and PLM at IBM, Dassault Systèmes, and PTC, now an independent analyst tracking the PLM startup ecosystem — joins Jeff Brickler to trace the history of CAD, PDM, and PLM, unpack why 440-plus AI-native startups have emerged in this space, and talk through what "advise, assist, automate" actually looks like for engineering and manufacturing teams adopting agentic AI.

Speaker note: two speakers — Jeff and Michael. Jeff hosts, frames each topic, and drives the conversation forward; Michael carries essentially all of the detailed PLM history and career narrative. One name in Michael's college story (a colleague who went on to become one of the youngest billionaires in history) came through the recording unclearly — I've left it generic rather than guess at a specific real name.

Welcome & Introductions
Jeff: Welcome, everyone — we'll give it another thirty seconds for people to join. I'm really excited to have Fino today; we're going to dive deep into PLM.

Michael: Probably too early for a beer on your side of the ocean.

Jeff: Yeah, it's about five o'clock for you, though — beer o'clock.

Michael: Depends where you are and what you do, but here on the East Coast it's 11 a.m.

Jeff: So I appreciate you joining at the end of your day.

Michael: Not a problem at all — looking forward to this, it's going to be fun.

Jeff: Thanks for joining.

Michael: You guys actually spelled my name wrong on the screen, by the way — it's F-I-N-O-C-C-H-I-A-R-O, not the other way around. It's wrong in the little name tag too.

Jeff: Thank you for catching that — sorry about that.

Michael: No worries. Today we're going to talk about what everyone's talking about — AI — but specifically through the lens of PLM, since that's where you've spent your career.

Michael: Welcome to the CADTALK Integrate Intelligently podcast — great to be talking. I follow you on LinkedIn, so this is a bit of a privilege for me. Thanks for having me.

Jeff: It's my pleasure — I was glad you all reached out two or three months ago to set this up. Looking forward to learning more about CADTALK and talking about what's happening with AI, the PLM vendors, and the startup community. This'll be fun.

Jeff: AI is all the rage right now. The way I came across you — just to plug CADTALK for a second — is that we work in what I call the messy middle, between PLM, CAD, PDM, and ERP. We handle the back-and-forth communication between those systems, and that's how I found you on LinkedIn and reached out. My own experience with CAD, PDM, and PLM really comes through CADTALK — I'm in that ecosystem, connecting these things, but you've spent a career, a lifetime, at all the big players. Tell us about your journey — how you got into this.

Michael's Career Journey: From CAD Graphics to PLM Architecture at IBM, Dassault, and PTC
Michael: I'll try to keep it quick. I actually started on the CAD side — back in the early-to-mid '90s we still had Unix workstations, about the size of a washer-dryer, with graphics worse than what's on my phone today. Those disappeared around 1997 when the Nvidia GeForce came out. I was tuning graphics drivers for those systems — always loved the performance side — then moved into PLM and became the go-to guy for tuning that, starting with hardware vendors IBM and HP, two of the big Unix workstation providers alongside Sun, Digital, and Silicon Graphics.

I did PLM and became sort of the generalist "windshield guy" at HP for a while. Then HP decided they hated everything with a three-letter acronym, so I left for PTC, since they already knew me from tuning their systems — and PTC's performance back then was genuinely nothing to brag about. I was the guy flying to two or three countries a week trying to tune their systems and telling PTC to fix the underlying issues — "it's software, it's fixable."

Eventually I got recruited to IBM again, right as Dassault was pushing their V6 platform and IBM couldn't get the information they needed. I joined IBM just in time to be sold off by IBM back in 2010 — the Dassault deal, 700 of us for 600 million dollars. I spent about seven years at Dassault in various roles — R&D, the Enovia brand — always as the platform expert and performance architect. At one point, partly my own idea, they split the PLM application (Enovia) from the platform itself, since it had all been one big blob before. Once they split it, I couldn't stay in the Enovia app group anymore, since I was the platform guy — I drifted into presales for a while, never quite found my footing, and left in 2017, after having lived through and partly guided Dassault's move into the 3DEXPERIENCE platform.

I've been independent since 2017, with a short stint at a PTC reseller where I learned the Thingworx and IoT side of the world — so I've got one foot in engineering and one in manufacturing. For the last year or so, I've been doing various consulting missions but also studying the startup ecosystem closely — writing a newsletter, running two podcasts ("The Future of PLM," with various thought leaders, and one on AI across the product lifecycle), talking to startups. I've now got 440 startups logged in my own database — 12.6 billion dollars in VC funding, roughly 50 billion in market cap, at least six unicorns in there. It's much bigger than anyone realizes, bigger than I realized when I started. I've talked to about 70 founders so far, and the rate of change is genuinely astounding.

Jeff: I agree completely — it's cool to see this space get attention, because when I talk to customers, we're in a less glamorous part of it, the middle. But it's great to see people excited about PLM and CAD specifically, since so much of the AI spotlight goes to consumer stuff. Second question — what's your actual educational background? Software developer, engineer?

Michael: I'm a mechanical engineer — University of Florida, BSME. Never finished my MSME, partly because I hated my professor, who eventually got fired despite having tenure. But yes, mechanical engineering background — I quickly gravitated to computer graphics. In fact, the reason I got my first job at IBM: the CAD lab had received an IBM 5080 — this refrigerator-sized, dorm-room-fridge-sized machine — and couldn't get the graphics working. The guy running the lab was so frustrated he called John Akers, the head of IBM at the time, directly, which is insane. Akers basically told his team: get a student up here for seven months, have them fix it, and I never want to hear from University of Florida again. I was that student.

The amazing coda: the same group I worked in, the year before I got there, had another student in that exact graphics job — porting CAD and 3D packages onto IBM graphics hardware attached to Unix workstations. That guy went on to become one of the youngest billionaires in history. He was in that group a year before me — I never actually met him.

Jeff: Wow.

Michael: You never know where you're going to end up.

Defining Terms: What Is PLM, and What Is Agentic AI
Jeff: Let's define our terms for anyone listening who might not be as familiar. PLM is product lifecycle management — really about helping engineers manage the product beyond what CAD or PDM do on their own. And Agentic AI — beyond ChatGPT-style interaction, it's about taking actions independently in the background, orchestrating workflows and getting things done autonomously.

What's Worked Well Over 30 Years: From Clunky UIs to Cognitive Digital Threads
Jeff: You've been doing this three decades — I've been at CADTALK eleven years. What's genuinely worked well over that time?

Michael: What's worked well is how we've adapted technology to change the way engineering actually gets done. It used to be pencil and paper, then computers, but early user interfaces were pretty clunky. Over time we've smoothed out the user experience considerably, and I think we're seeing an even bigger leap now with AI-native applications. Making things easier to use also democratizes access — even a non-programmer can use live coding tools today and build something that actually works, and the same goes for engineers scaffolding ideas and having AI help build them out into full applications. I've interviewed 20 to 25 startups on my podcast now about how they use AI daily to build and modify their own products.

Data governance is also improving. Companies used to silo data heavily by department — engineering had its data, didn't share much with simulation unless forced to, and manufacturing was basically "throw it over the wall and hope someone catches it." What's working now is more robust, even cognitive, digital threads that make that cross-department data connection much easier.

Advise, Assist, Automate: A Framework for Agentic AI Adoption
Michael: I really like a framework from Aurorius Barry, who runs AI strategy at PTC — "advise, assist, automate." It captures the mission well.

Jeff: From what you're describing, it sounds like the first step is advise — give people the information, essentially read-only. Second is assist — here are some options, which do you want me to run. Third is automate — a full workflow with guardrails to actually execute a task.

Michael: Exactly, and I think it works well as a framework.

Jeff: As an engineer yourself, working with a lot of engineers and manufacturers — my experience is people adopt this much more readily if they feel in control, rather than an AI just doing everything for them out of the gate. People get uneasy — "whoa, wait, what exactly are you going to do for me?" So walking through it step by step, letting people see the value at each stage before moving on, matters. And I imagine advise, assist, and automate will all have a role throughout, not a strict linear progression.

Michael: Right — I just think the three-word framing is a good way of describing the evolution as we adopt this into engineering and manufacturing workflows; I wrote a whole series about it last year. What's also interesting, tying back to what's worked over the last 30 years, is that we used to work in strict single disciplines — mechanical engineers stayed in their lane, then it got handed to a different engineer. Now we're seeing a real blending: tools powerful enough to give me design-for-manufacturability or design-for-sustainability insight as I work, which makes engineers more multidisciplinary. Jensen Huang talked last year about how every physical factory will eventually have an "AI factory" associated with it — and I think what our industry is building right now is the foundational building blocks for exactly that.

Breaking Down Engineering/Manufacturing Silos
Jeff: To double-click on breaking down silos — we've played our own role in that over the years, as the glue between engineering and manufacturing. We have a product specifically meant to help break down that wall, so engineering can design in a way that's actually manufacturable. And I'd agree that AI is going to accelerate that further — those departments won't stay so walled off, communication will happen more naturally, and people will get what they need, where they need it, more easily.

Michael: Exactly.

A Brief History of CAD, PDM, and PLM: How the Big Three Formed
Jeff: Let's talk through what PLM used to be, what it is now, and where it's going. You mentioned PTC, Siemens, and Dassault — the big three. I imagine they were all involved early in developing PLM itself?

Michael: Going back further — before PLM, the first real CAD system was actually built for manufacturing, by Dick Sourcer's team, who ended up acquiring what became CADDS from a company called SDRC — no, sorry, it was three-space, then that became CADDS, the first real modeler. That same group later split off and created Shape Data, which became Parasolid — the kernel that underlies what's now NX but also SolidWorks and Onshape.

At one point you had maybe 15 to 20 CAD companies, which consolidated down to four or five, plus AutoCAD, whose 3D business is smaller than the other three but still significant. There was also a tiering effect — for a long time only high-end systems existed: CATIA, Unigraphics, Pro/ENGINEER. Then Windows-based tools arrived around 2000 — SolidWorks, Solid Edge, and similar. The biggest CAD vendors realized they had all this CAD data and no way to manage it, so PDM emerged first — how do I manage the actual files? Pro/PDM, then Pro/Intralink on the PTC side; VPM in the Dassault world.

In the course of developing those tools through the late '90s, people started realizing — if I'm already managing the files, why not manage the lifecycle around them too? That's where the idea of "it's not just the design, I also need to hand this to manufacturing" came from — a lifecycle spanning engineering and manufacturing. That kicked off a massive wave of consolidation, roughly 2005 to 2007: PTC rebranded Pro/ENGINEER as Creo; Unigraphics got bought by UGS, then by Siemens, becoming Siemens Digital Industries Software; Dassault had already acquired SolidWorks alongside CATIA, giving them both a high-end and mid-market offering — and really, all three of the big vendors ended up with that same high-end-plus-mid-market structure: Siemens with NX and Solid Edge, PTC with Creo and Onshape, Dassault with CATIA and SolidWorks. PLM effectively wrapped around all of that, partly as a way to sell more software.

Around 2000 you also had a couple of newer companies — Aras, and bom.com, which later became Arena and was eventually bought by PTC, once PTC decided they didn't necessarily need to sell CAD to sell PDM or even full PLM. Bom.com can probably claim to be the very first cloud-based PLM system. Aras did both cloud and on-prem. There were others too — MatrixOne, later bought by Dassault and folded into what became Enovia V6. From there, PLM's scope just kept expanding well beyond engineering and design — into service lifecycle management, model-based systems engineering, ideation — and these vendors turned into genuinely enormous mega-corporations.

Jeff: Even into manufacturing itself.

Michael: Exactly — both Siemens and Dassault went all-in on virtual factories, plus CAM, tooling, routing, and machining. PTC went more toward asset lifecycle management, buying ThingWorx for that — then, quite famously, sold it off last year and refocused back on PLM and CAD, plus requirements management and Codebeamer.

There was a similar wave of consolidation in simulation over the last three to five years — Dassault had acquired Abaqus to create the SIMULIA brand; Siemens has been acquiring aggressively too, CD-adapco, Altair most recently for their simulation tools. Last year you had Hexagon's MSC business sold to Cadence, and Synopsys buying Ansys — a lot of consolidation in simulation that ties back into the broader PLM story. And if you want to go further, Siemens bought Mentor Graphics, becoming an ECAD vendor and a mechanical CAD vendor at once.

Jeff: They've all become these giant, all-encompassing platforms.

Michael: There's been a genuinely monolithic quality to all of it — every vendor wants to be the single platform that sells the customer everything, no mixing and matching between vendors.

Jeff: Right, no one gets fired for buying the one platform that does it all.

Michael: Exactly — which is what makes it both refreshing and surprising that I've now found over 440 startups in this space. The real spike came around 2022, roughly the same time OpenAI released ChatGPT-3 and LLMs took over. There were startups before that, but 2022 saw a big jump of people saying: I'm not going to wait three years for Dassault to build this feature, or however long for Siemens to ship a co-pilot — I'll just buy someone else's. And the customers for a lot of these startups turn out to be Boeing, Anduril, various drone companies — companies that want to move much faster than the big three can, since the big three, being so large, aren't as agile. You're seeing real adoption of what looks like "risky" startups that are actually quite stable — Collab raised 72 million last year, NeuroConcept raised 100 million, Leo AI, only two years old, raised 9.7 million in September as an engineering co-pilot. Real money, real products, used daily by engineers building real things.

Jeff: I didn't realize there were that many startups. We partner with all three of the big vendors ourselves, since we help connect their customers' systems together — that's where we sit. I'm surprised at the volume, mostly because manufacturing and engineering companies tend to be pretty risk-averse in my experience — "no one gets fired for buying the thing everybody already knows," same joke we make in ERP about SAP. I heard someone at an event describe it as FOMU — fear of messing up, instead of FOMO. So I'm glad to see it, but genuinely surprised.

Michael: I'm more surprised there are so many than surprised they're succeeding. If you start building software today, in 2026, with a solid foundation and clearly explain to your tooling exactly what you want built, it's genuinely not that hard to produce production-ready software. Most of the founders I talk to came out of places like SpaceX, Tesla, or Formula 1, and saw real dysfunction firsthand — tools that don't do what engineers need, workflows shaped entirely by technical debt and legacy systems. Michael Rosam of Quix, ex-Formula 1, told me their multi-million-dollar cars, with tens of millions in R&D behind them, had test data sitting as literal piles of paper on desks — none of it digitized. He built a system to digitize and speed that up, and that became his startup. Adam Keating, similarly, built AutoReview because design reviews are a mess when one team is on CATIA and another's on Creo — the big three aren't exactly eager to make cross-tool collaboration easy, since they'd rather keep your whole supply chain on their own tool.

The other issue — I won't name names — is that some vendors have been raising prices without delivering meaningfully new value, and even conservative companies are starting to look for alternatives. Some of them call me directly asking what else is out there, since paying to heavily customize software and keep it updated through that customization is already a burden, without an arbitrary 20% price hike layered on top with nothing new to show for it.

Real-World Agentic AI Use Cases
Michael: What's promising on the agentic AI side specifically is when you put in real guardrails and deploy agents intelligently — for example, teams combining CAD and simulation data in intensive workflows, where engineers can still see where the data's going but automate the tedious parts, instead of spending their day copying files between drives, which has genuinely eaten a disappointing amount of engineers' time. The real promise of AI in engineering is removing the mundane so people get to the interesting, innovative work.

Jeff: And there was a real architecture shift too — around 2010 to 2012 every vendor moved to the cloud, which made collaboration and system connections much easier, since you didn't have to fight through firewalls constantly.

Michael: Right, though we might see a bit of retreat from that model given tariffs and other politically driven pressures around data sovereignty. AI actually helps there too — with something like MCP, the AI can handle all those connection protocols itself; you just hand it the keys it needs. So the architecture shift going forward is really toward agentic workflows using MCP or similar agent protocols to handle interfacing. Have you talked to Michael Cor of Duro? He just exited — Arena acquired Duro about three or four weeks ago. Michael had originally spent four years writing the first version of Duro, and after the ChatGPT wave hit, told his team they were starting over, building it AI-native — specifically having AI write all the interfaces to other software, since there's no reason to have a developer manually figuring out inputs and outputs when AI handles that well. What took four years the first time took about six months to rewrite. That was apparently convincing enough for Arena to acquire the company for an undisclosed sum.

Jeff: I wanted to ask — with any new technology, there's a hype cycle, and I've heard some pushback that agentic AI investments haven't lived up to what boards and investors expected. Where have you actually seen it work, and what are real use cases these startups are shipping today?

Michael: Starting broadly with AI in general: it makes writing product requirement documents incredibly simple — you describe what you want to build, it generates the documentation and walks you through step by step to make sure nothing's missing. Nearly every startup I talk to uses AI throughout their own development process too — developers using Claude Code or Cursor to build the entire application. From there it's a decision point: train your own models, or use the public ones augmented with RAG or similar techniques — most go with the public models. What really moves the needle is layering in domain knowledge — training the AI to actually understand mechanical engineering, how manufacturing works, blending knowledge across disciplines so overall engineering efficiency improves. That's where I think we're seeing the biggest breakthroughs, and it applies to manufacturing too — there's a French startup, Neuron (or similarly named), whose idea was letting machines coordinate directly with each other without a human in the loop when two machines need to sync up; I think Flexxbotics does something similar.

There's a genuinely huge range of use cases, and it's exciting to watch, provided the guardrails stay in place — the concern everyone has is AI hallucinating and things going sideways. I sat in on a conference in 2025 where an Oxford professor made a great point: airplanes have black boxes, so if a plane crashes you can always roll back and find the cause. Robots have no equivalent standard yet, and eventually a robot will hurt or kill someone — she was pushing for something like a black box for robots. My extension of that: manufacturing and engineering need the same thing for AI-assisted decisions — some standard ensuring that if I use AI to make a decision, that's logged somewhere, so if something goes wrong later, you can trace whether it was me, the engineer, or the AI hallucinating.

The Need for a Skeptical "Doubting" AI Agent
Michael: One more thing — the everyday agents we use, Claude, ChatGPT, Gemini, are trained largely on data that rewards being friendly and telling you how brilliant you are. I think we actually need the opposite — a "doubting" agent.

Jeff: Wait, say that again — a doubting agent?

Michael: Yeah, an agent whose whole job is to push back and question you, rather than just affirm whatever you propose.

Jeff: [laughs] We'll have to train that one on engineering forums specifically.

Software Architecture Shifts: Graph Databases and Cognitive Digital Threads
Jeff: Let's talk about the architectural changes in software and how you think they'll help with these AI workflows.

Michael: Modern coding — separating control, data, and view — lets us build interfaces far better adapted to how humans actually think, without worrying about implementation details. That's essentially what chat interfaces have done: given us a plain text interface and hidden away all the "click here, then click there" complexity. On the architecture side, graph databases matter a lot too — being able to build powerful graphs that surface the implicit connections between data, so I can see why a change on this part has a downstream impact on manufacturing, and therefore needs a different design approach, or a reordered set of steps in the work instructions so I don't put a worker in a dangerous position or a part somewhere it can no longer be machined without being removed and reset. That whole space — graph databases, cognitive digital threads — is what's letting engineers become genuinely more multidisciplinary.

Advice for Engineering Students in the AI Era
Michael: On my podcast I get a lot of younger listeners, and I always ask founders what advice they'd give students right now, since a lot of them are anxious — "why am I studying all this if Grok or ChatGPT is going to have my job." My honest answer: they still need the fundamentals, and the actual problem is that a lot of them don't have the fundamentals and just trust whatever ChatGPT tells them. A customer told me this week about a simple problem — three unknowns in a fraction, just cross-multiply and divide — that a junior engineer got wrong because he plugged it into ChatGPT and never checked whether the answer was even in the right order of magnitude. In that case, you're not just replaceable by AI, you're actively worse than it, because at least you should've caught an obvious error it made. You'll always need the fundamentals — the intuition to look at an answer and know something's wrong, even before you've done the math yourself.

Jeff: Fundamentals stay fundamentals — these are just tools that make the fundamentals more effective. I came from a completely different background, foreign language, and having lived in Paris, you'd know this too — learning a language, there are tools that help, but the fundamentals still matter. You still have to learn the words, still have to do the work — you can't lean entirely on the tools.

Michael: Exactly — you still have to be able to think for yourself, at least a little. Hopefully that never goes away.

Assessing Your Organization's AI Readiness: Data Governance and Architecture
Jeff: We touched on open APIs, OData, REST — the market's clearly moving toward AI adoption with all these startups. So what's next — for someone listening or watching, with 440-plus startups, 12.6 billion in VC, 50 billion in market cap, obviously you can't evaluate them all — where do people even start?

Michael: I'm tempted to just say "better call Fino," but beyond that — start with an honest, clear-eyed assessment of where you actually stand: where the gaps are, where you've got too much dependency on customization that makes your code brittle and hard to upgrade. Look at your architecture for scalability and openness, and make sure cybersecurity is genuinely foundational to your thinking, not an afterthought — that's non-negotiable.

Organizationally, I'd separate the data function from the IT function — they should be different reporting lines. Data should be the oil that makes everything run, distinct from the IT gears.

Jeff: When you say "data," you mean data governance specifically — quality of the data, what you're collecting?

Michael: Right, and that typically gets treated as an IT decision, which I think is unfortunate — it should sit with a chief data officer distinct from the chief IT officer. You need a matrixed structure: business units on one side, and horizontal functions bringing things together on the other. Most data-quality problems aren't actually IT problems or something an app can fix — they're about how people work and think about the data. A chief data officer defines the databases, the formats, what gets used; the business units then act as data owners and stewards, ensuring their data aligns with corporate strategy so cross-division analytics actually works. Data governance is something that gets ignored a lot.

Jeff: That's interesting — hadn't quite framed it that way, but it tracks, since in an AI-driven world your proprietary data really is the fuel, the currency, more than the specific product. I actually had our founder on the podcast making a similar point — that the data ends up mattering more than the product itself in an AI context. So the quality of that data, how it's structured, how it's stored and accessed, becomes more critical than the IT infrastructure it sits on, which obviously still matters too.

Michael: Right — and honestly, different skill sets entirely; IT skills and data skills probably deserve different hats, not the same person wearing both.

Evaluating AI Startups: What to Look For
Michael: To answer the second part — evaluating startups depends a lot on your budget and risk appetite. Series C or D startups are already established, in growth or expansion stage, and pricier. Seed or Series A companies tend to be far more willing to listen to customers and shape the product around your needs. Look at the backing they have, the pedigree of the founders, and who's already bought in — you might be surprised by some of the customers already using startup software.

For evaluating AI claims specifically: check whether their reference customers are actually referenceable, and talk to them directly about real-world performance. Watch for buzzword abuse — someone who actually knows what they're doing isn't leaning entirely on buzzwords, there'll be concrete use cases. Honestly, having talked to nearly a hundred of these companies, I haven't run into many that were pure PowerPoint — I've been genuinely impressed with what I've found. Not much vaporware out there.

Jeff: Interesting.

Michael: I don't think vaporware survives long in this space — if you want to get picked up by some of the money currently flowing toward defense contractors, you need real software, or you're out within minutes if you're not actually solving a real problem. There might be more room for vaporware in other industries, but ours moves too fast and matters too much for that. Adoption really comes down to giving people interfaces they can relate to — something with a hundred confusing clicks gets ignored in favor of something natural that solves an actual problem. You've also got to identify a genuine user pain point and solve it well enough that people trust you with the next one too, and it snowballs from there.

Most organizations will have some technical debt with one of the big three, especially in brownfield accounts — look at whichever parts of their portfolio you haven't fully evaluated yet, see if they solve your problem and at what cost, and compare that against a handful of startups solving the same problem at a lower price point. Most startups already offer some REST or OData integration, and hopefully more and more will offer MCP interfaces over time, so we can skip a lot of the custom integration work entirely. If I had to pick one thing to look at this quarter, I'd look at your workflows and find wherever you're getting the most complaints through your IT ticketing system — the one system everyone hates — then go evaluate three or four startups against that specific pain point. With 440 startups out there, there's likely at least one in the right ballpark for what you need.

Where the Big Three End and Startups Begin
Jeff: Last question — I imagine you don't think the big three are going anywhere?

Michael: Well, Autodesk just announced a round of layoffs, so who knows what's happening there.

Jeff: Right, and there was that speculation for a while about them trying to buy PTC.

Michael: They actually leaked something they shouldn't have — a genuinely botched, failed attempt.

Jeff: If we assume the big three, or big four counting Autodesk, aren't going anywhere — what are the startups actually there to do? Chip away at market share from the big players, or address entirely new markets the big three don't touch?

Michael: I think the honest answer is both. There's real white space in newer technology areas the big three simply don't cover well yet — newer physics solvers, especially in simulation, or in CAM, where outside of NX CAM there are real gaps connecting to tools like GibbsCAM or Mastercam. There's also the mid-market problem — the big three are built for the Fortune 500, and their PLM platforms genuinely aren't great fits for mid-market companies. That's created real opportunity for companies like OpenBOM, Duro, Propel, even Aras to some degree, though Aras has skewed more Fortune 500 historically.

You're also seeing a lot of new hardware startups in that mid-market space — energy, defense, drones — and now that DJI's been banned here, I'd expect the drone industry specifically to explode with new development. Those fast-moving smaller companies are going to be eager to adopt startup tools rather than commit fully to one of the big three, since starting with one of them usually means they'll try to sell you the whole pie even if you only wanted one slice.

Wrap-Up and Where to Find Michael
Jeff: We're about at time, Fino — thank you for this, I genuinely learned a lot. I'll plug your podcast and your LinkedIn — I believe you've got an episode on bills of materials coming up?

Michael: Yeah, that's next week.

Jeff: Perfect — and of course your newsletter, PLM Demystified. Anything else you want to mention?

Michael: I'm also starting a conference series called Threaded, for startups — the first one's in Warwick, UK on March 25th, the next in Miami on April 13th. It's a forum for startup founders to meet each other and talk about how AI is transforming engineering and manufacturing, and how to survive as a startup in this ecosystem. The Miami event will be tied to the ARS event, so a lot of larger customers will be around too, letting the startups mix with them directly. Those are the first two, but I've got at least three more planned this year.

Jeff: That's great — looking forward to more conversations. Folks can find us at cadtalk.com and on LinkedIn — I'm Jeff Brickler. Fino, this was a great conversation — I think it's about beer o'clock, or wine o'clock, whichever applies. Great talking with you, we'll talk soon.

Michael: Take care — thank you very much, everybody.