Wired In: Building an AI Strategy for Product Development
00:09 Welcome to Wired In, the show that connects you to the latest advancements in electronics and high-tech innovation. I'm your host, Stephen Humphrey. AI is on everyone's mind, and the transition from pilots to enterprise scale is happening at an incredible pace. Today, we're joined by Eora Berry, PTC's Vice President of AI Product Management. He brings deep expertise in artificial intelligence and spends much of his time helping our customers reinvent their product life cycles through advanced
00:39 intelligence systems. I'm looking forward to our discussion, particularly around how companies can frame an effective AI strategy to balance rapid innovation with elevated product quality. We'll also explore best practices for enterprise adoption, the importance of a strong data foundation, and the practical use cases proving most valuable for engineering teams today. Let's dive in. Eora, thank you so much for joining me on Wired In today. Yeah, great to be here. Yeah, we're very excited to have you. So, you know, we can just dive right in. I'm excited for our conversation today.
01:09 So let's start, you know, from your perspective. Right? How should electronics and high-tech companies frame their overall AI strategy, right? You know, to effectively balance the rapid innovation that we see so predominantly in electronics and high-tech, but also to maintain that high product quality. Yeah. I like to think about it i- in terms of, like, maturity levels. So it's a three-part view where you have, one is more around, like, turbocharging workers. Sure. Two is around the enterprise.
01:39 I, I, you know, call it elevate enterprise IQ. And three is around how we even rethink product development, reinventing product life cycles. So let's kind of walk through each three and see what that means from a, a maturity model and how do you manage your, your strategy. So right now, all the action, it's in the turbocharging workers. Right. It's like 80% of the use cases are oriented towards, like, AI advisors and assistants enabling individuals to be faster, be more productive. So think, like, asking questions around how to use a piece of software.
02:10 That's very basic, but let's move beyond that to, like, hey, I need help in decision support, in how I change a product, or how I modify a product, or I get ready for an inventory plan. And then there's an additional level, which is how can you actually assist me in some of that work? Can you review my requirements for quality? Can you go update geometry based on this mass property analysis? So that's where all the action is right now. Sure. And where we see things kind of also leading to, merging to happening today is around this AI working across products.
02:40 This how can we elevate the enterprise's IQ in terms of doing business? So then you get into, like, more business process-based value delivery where you see things like how can we improve enterprise change management? How can we improve design collaboration? How can we improve, service life cycle management? And in that framing, what we're seeing is AI, and in particular agents, are becoming a vehicle to accelerate customers' digital transformation projects. Right. And the reason for that is that we now have this unit of work, an agent.
03:13 Think of it as, like, labor embedded in your software. It can go do jobs for you, and those jobs not only include, like, individual pro- you know, like, use cases in a product, but they can go over to other products. They can talk in language like MCP. It can gather that information, bringing it back. But also it can do more than that. It can prepare for that information that comes over. So let me go get some information from the ERP system to see what the impact is on inventory on a product change, for example. So that's the second stage of a kind of maturity. And then the third one is, like, how can we actually rethink how products are built?
03:46 How can we reinvent how engineers, designers, technicians do their job using AI, where you have these, like, super agents working across products and redefining the process itself? Right. So instead of, maybe going through a typical kind of waterfall process of, like, let's design, build, test, let's start and do design and test right away. Let's have agents that are your requirements builders, your design builders, but also let's have agents that are your validators, your test case creators, and then go run those test cases working together.
04:18 So that upfront in the beginning of the product development process, you have that, what I would say collapsing the design test iteration loop that we see in all of our processes. So that's an important framing because when you think about your strategy, you gotta, you gotta identify where am I gonna invest first? So invest first in turbocharging your workers. Sure. That's a, that's a, a, a well-proven use case that you can get value today. You need to start working on that enterprise IQ because there's a lot of work that goes into connecting the dots.
04:48 Like, you're gonna need to have access controls managed. You're gonna need to have agents be able to speak across different data systems. You may require an ontology to support that fa- that framing. And on the third one, you need to start thinking about, okay, hey, in terms of being relevant the next one, two, three years, and if you asked me, by the way, like, three months ago what the timing was on this- Very different answer maybe I would've said maybe three to five years. Yeah, right. But, like, now we're getting to a point where this reinventing product life cycles, it's clearly happening in the software engineering space in terms of how we write software, how we create software, but it's gonna be happening much faster, I think, in how we
05:19 build hardware, how we design requirements for hardware, how we test those requirements through, you know, model-based system, engineering system. All of that is gonna get rapidly upended sooner than I think. You know, in terms of, like, two, three years, it's gonna be more one to three years. Yeah, you know, the maturity model kind of framing you make makes a lot of sense to me, right? Mm-hmm. And like you said, like, if you would've asked that question a year ago, the answer would've been different, timelines would've been different. And even just more broadly, like, when I think about electronics and high tech and, like, how fast that industry moves and new innovative technologies, right?
05:49 Like, the technology adoption curve. Like, I think we're still pretty, you know, early in some of those phases, especially in the enterprise. Yeah. So what you talk about, about, like, AI becoming kind of this, like, supercharging, going into, like, then this orchestration layer- Yeah across other business applications, and then where does it e- where does it go from there, right? Right. There's probably things on the horizon we haven't even thought about or that have been developed yet. Mm-hmm. But it is, you know, it's a very exciting time and, you know, we're obviously diving in headfirst here at PTC, right? Yep, 100%. Awesome. So moving on to our next topic on, as we're talking about these
06:19 enterprise companies, right? How, how, you know, what are some of the things you're seeing that help them become ready to really step into this AI space, right? What are some of the best practices for, for them to get started? And what questions should they be asking first? Yeah, I love that. Um- It's essential to get your data house in order, right? At PTC, it's, you know, your product data foundation. Sure. What is that? Your product data foundation is not only the parts and requirements and tests and designs and service work orders, it's also how that
06:51 data is constructed in our products. There's data tables. There's, within those data tables, access rules. Within those access rules, there are workflows with control points. All of that constitutes your, your data foundation. And so why is that important? It's important because, one, AI needs to be able to have the context of that information. Sure. It sets it up to kind of interpret, make sense of it, go take action with that data. But what's really important in the enterprise too with that product data foundation is around that governance. Like, the AI needs to operate within the
07:24 governance model of the access rules that either the, that the user invoked it or some business process invoked the agent. It needs to follow that system. So when you think about that product data foundation, what's also really important from, like, getting started is, like, okay, so I, I've got this data now, and now I'm putting some AI into it. I think an important question everyone should ask themselves when they're getting into AI is, like, what's different? Like why, what's the big value add with this AI technology? Sure. How is it changing what I've done in the past? The reason why I state that is that we've been on digital transformation journeys for decades.
07:58 Right. We've talked about digital thread, we've talked about digital tweed, and it's being, it's, it's constituted in our products. We have processes around it, and now we're saying AI is, like, turbocharging that process, right? So how is it doing that? Why is it doing that? So when you look at your use cases, when you look at your opportunities, say, what's different? So let's take an example. Like, in the, that middle maturity model of, okay, let's elevate enterprise IQ, what's different with AI? Well, one of the things I mentioned is, like, well, AI is really good at interpreting information quickly. It can go over to another system.
08:29 Mm-hmm. It can use model context protocol. What's that? It's just a wrapper around APIs that helps the AI process that data, those API tool calls. Right. And it says, "Hmm, you've got these tools. I'm an AI system. I can process that with natural language, and now I see what you have. I know what's in the toolbox, and I can get to work now." That simplification of, like, dynamically reading what tools are available in the system and then coming back and delivering an output from that is different than before, where you had to do hard-coded integrations. Sure. So that's just, like, one example. So when you think about AI and you're building out your system, you really wanna focus on
09:03 what's different, what's it changing. I know that seems very simple, but when you start peeling back the layers and actually, like, think about that technically, think about that operationally, think about that how it impacts the business, there's all these interesting things that come out of it. So another example, like, on the business side, this one I, I find very fascinating. So, like, if you think about AI and what it does for us, like, we now have these agents. What, what is an agent? Right. Think of an agent, and the, the best way to do is, like, anthropomorphically. Like, an agent is like a persona. It's a specialist. There's general agents, like the, call them,
09:36 like, the, the, the, the construction site kind of planner, and they know the plan. They know all the agents, what they do, and they orchestrate them. But there's also specialists too- Mm-hmm like the plumber, the electrician, the person that's out there- Sure and they know how to do certain jobs. And so these agents are working together, cool, and they have a job, okay, and they have certain data they work with. But what's interesting now from, like, a what's different perspective is that those are now being delivered with software. And I know we all know that, but just take a look back at, like, the software journey of,
10:07 like, originally we would sell software, and you'd get it, and you'd pay a, a, a, a price point for it, right? Like, just like buying a hammer. I'll pay five bucks for a hammer. I'll put it in my toolbox. When I need it, I can pull it out. Same thing with software. I can pay for that software, put it in my toolbox. I have a perpetual license. I can use it when I want it. And then we moved over to, like, a subscription model, where, like- Right now the software has more tooling. It's getting updated more regularly. It has a kind of maintenance plan around it. We've kind of, like, upgraded from hammer to, like, nail gun or something like that, right? Sure. Good analogy. And then now we're like, all right, not only do you get the hammer or the nail gun, but now when we deliver software,
10:40 we're gonna give you the labor for that too, the agents. So they're actually gonna know how to use the hammer and go do that job. So, like, that's what's different, another angle. Like, from a business perspective, software now comes with this digital labor, and it does work. So there's all these different angles to kind of think about when you're making your strategic decisions on how to use AI, whether it's technical, operational, or business. And the question I always ask myself to kind of think about is, like, what's different? And then I peel back the layers and kind of say, "Okay, that's interesting. What's the ramifications for me as either a buyer or producer of AI?" On this one, the
11:10 business side, it was like, as a consumer, I now gotta think about different price points. Right. As a builder, on the other one, as an agent builder, I gotta think about what MCP endpoints are there. Like, that's the example of, like, if you peel it back, you kind of understand where to go next. You know, I love that analogy with the hammer and- you know, then the agents and, and then the plumber and, you know, the different specialties. And, you know, I see this a lot in my day-to-day in the companies we work with, right, is that, you know, agents fundamentally operate different than people. Yeah. Right? The way they interact with the software. Yeah. Right? The way that they interact with the data. Mm-hmm.
11:40 Right? The, the user interface. So agents, some agents don't even need user interfaces- Correct right? They use command line, or they're just using API, APIs and headless applications, and there's all these new approaches and, and, and, and modalities of how these agents can work. And so we have to think in the context of an agent, maybe less in the context of a normal day-to-day- individual contributor. Yeah. Right? And, and so that idea of, like, one agent governing other agents and how they're interoperating with different applications and, you know, what they're doing and why they're doing it, it's, it's an exciting time to
12:10 explore- how that can help these enterprises, and that product data foundation is fundamental, right? Yep. Great. So I think we know AI at this point, it's really only as good as the data that it consumes, as we're talking about this product data foundation element, right? So how critical is it to establish a single source of truth, like a PLM system- for example, a system of record, before deploying these advanced tools in the enterprise? Yeah. It's essential. Sure. And, and the reason for that is, um- I'm smiling right because I was, I was in a meeting yesterday, and I started talking about this framing I have, which
12:42 is like context is king. And I was I had my glasses on at that moment, and I was like, there's actually two lenses when you're thinking about data access. There's, like, context and there's control. Sure. And so the reason why I bring up this point of context and control is that these systems of record, a PLM system, getting that set up and established as your product data foundation for your PLM operations and the AI associated to it, is it is delivering both lenses. So on the context side, AI is only good as the context.
13:15 We talked about that earlier, and the reason why is it, with context, it delivers reliable and relevant results. Sure. So when you think about an agent going into a system to, like you said, whether it's headless or through a UI, it needs to know what that information is, how to use it, what access rules are, are my privileges. But in terms of, like, the context of just, hey, is this relevant for the question that's b- just been asked? Like, hey- Yeah I, I've just queried this, and I've got a bunch of tools in my MCP server. Is this one relevant or not? It needs to be able to interpret that. So the PLM system, for example, delivers that intent.
13:47 It says, "This object is related to this object in the data schema, and this person right now is working on this specific part and is part of this change request." That thread, that digital thread of data is context that the AI just soaks up. Like, that's- Yeah how it can deliver a relevant and reliable s- result. So that's the context side. But then there's the control side. We hear stories like you need to have AI constrained. It's important. Like, if you let it loose, it's gonna hallucinate at the minimum. At the- Right major- It could go do something wrong.
14:17 It could delete a data point by mistake. It- It needs guardrails. It needs guardrails. Sure. Right? And so you design software to have guardrails, but what's interesting, not only, like, the agent has guardrails, but what you get this added layer of guardrails through the governance system that's already established in your PLM system. So you have all these business workflows, and they say, hey, if I'm gonna be doing a change process or a configuration management or I have some kind of quality process, you design that in Windchill with the business objects and those workflows, and you say, these users have access to this stage and that stage.
14:47 And you say, "Hey, Mr. And Mrs. Agent, plumber, contractor, whatever, this is where you get to work, and that's where you get to work." Right. And when you're in w- in that stage or that stage, you have access to this data. So you're getting added layers of, as you call, the governance in the PLM system. So that's why it's so essential. Like, you need to have both the context and the controls, and the PLM system is that governance system to deliver both. Yeah, and I couldn't agree more, and especially, you know, when we think about PLM as And I think sometimes companies will even view it maybe as a bit rigid because- it needs to be, right? It is the tool that governs your, your product
15:18 data, your product development process, the controls that go into that, who can access what data, how you can share that, how a product moves through the product life cycle- Yeah and those different stages and gates. And so when you add in an agent into that and the context, if they don't have that context and they say, "Well, I'm gonna, you know, push this thing from prototype to production and go tell ERP to order X thousands of units-" Yep and it's actually not ready or didn't go through a design review, right? So those are things that, you know, the human in the loop, I think, is still important. Yeah. But at the end of the day, like, AI can really
15:49 supercharge, you know, your business processes at the end of the, you know, to go faster. Yeah, I was just thinking, like, in your s- you know, in your space of, like, where you focus in EHC, like, I think that's incredibly important 'cause, like, the velocity of delivery of product to, to market is very fast. Absolutely. So if you make, if an agent makes a commit on a change, you're gonna have that go out to market quickly, and your ability to remediate that is much harder in the EHC market. So those controls have even added more added value in terms of meeting compliance and quality delivery, delivery in the market.