Next in Tech | Ep. 232: The ONUG Tech Community

View Full Transcript

Eric Hanselman: Welcome to Next in Tech, an S&P Global Market Intelligence podcast where the world of emerging tech lives. I'm your host, Eric Hanselman, Chief Analyst for Industry and Company Data at S&P Global Market Intelligence. And today, we're going to be talking about community-driven efforts to tackle some of the complexity of technology and tech implementation. And with me to discuss it is Nick Lippis, the founder of the Open Networking Users Group of Nick, welcome to the podcast. 
 

Nick Lippis: Awesome, Eric, thanks so much for having me. So psyched to be here with you. 
 

Eric Hanselman: Well, it's great to have you on. We talk a lot about the challenges of really getting effective technology implementations out there. But you actually took the initiative now some years ago to really put together a community who could come together to understand some of those challenges and actually start to do something about it. 
 

Nick Lippis: Yes. I think that doing something about it is like was kind of the hard part to figure out like it's always great to get people together to talk and there's a lot of value in that. And that's the networking, and that is just the ability for people to exchange ideas in places where they're not being judged or they're not being -- there isn't a salesperson like lurking behind. So those are really important. But I think what we learned early on with this community is that its real power is within 2 spaces. It's their engineering prowess because obviously, when you're a consumer of IT in a large enterprise, you get to see how things connect and work or don't. And so, it's the engineering to kind of figure all that out and make it work. So, it's like the design, build and run part that that really, we -- in our industry, we don't really talk at all about, which is like too bad. It's almost like that's the thing that we should talk more about in the industry. It's almost like we don't talk about manufacturing technologies and the skill set that goes along in manufacturing. We don't really talk about what it takes to kind of stitch all this stuff together. 
 

Eric Hanselman: Well, its technology, right? It just -- you fire it up and it all bolts together and your ponies and rainbows and all that good stuff. 
 

Nick Lippis: Pies and rainbows. 
 

Eric Hanselman: Well, but this is something the origins were back in the go-go days of software-defined networking when all sorts of great innovations were taking place, but you had all these initiatives from different vendors heading off in different directions. And you had technology teams now trying to bolt all this stuff together. You granted the potential was great. And yet actually getting it all to play right together was complex, risky, error prone. And I compare it to a lot of what we're seeing with AI initiatives today in that we're back in this world of complex environments, a lot of risk, you're investing in what is a very rapidly changing environment. The ecosystem is basically sort of moving under our feet. Which model you're going to use today, probably not the same one you're using next week. It's -- a lot of these challenges in terms of making this stuff actually really work, right? 
 

Nick Lippis: Yes. Yes. I kind of view like where we are today is in terms of technology and job adoption, especially with AI is very different from when we were building the Internet or when cloud came on, right? So, like on the Internet, we had gravity in the open protocols of TCPIP. It took us a little time to figure that out. We're going to do OSI or TCPIP, but we -- but the industry decided, and we all rallied. And that was a really great rallying for the industry. It opened up this amazing economic activity, and it was really bottoms up. And then the cloud stuff was all top down. Those are run really by 5 companies with massively large marketing budgets that we were trying to convince executive management that it's faster, easier, and cheaper to kind of do cloud. But there's no standards there that kind of -- that allow a broad ecosystem to develop, not like we had like in the Internet days. And so, the AI space, that's also kind of moving -- it's almost like a hybrid between the 2, right? Because there are like about 8 companies that are doing the bulk of the consumption of AI infrastructure equipment, right? And then there isn't really open standards in that space. We're starting to see that maybe in the Agentic space, there's a whole rewriting of new protocols like MCP and agent to agent as well as directory services for agents and so forth and many other protocols there, too. And hopefully, we'll see -- there seems to be a lot of gravity around MCP. But anyway, so all I'm saying is that there seems to be like in the AI space, there's clearly the large suppliers, both cloud providers as well as XAI and OpenAI, and it's -- I'm sorry, parent... 
 

Eric Hanselman: The whole collection of all these folks, and they've all got their own approach and granted, okay, maybe we've got things like MCP but now you've also got folks who are actually starting to integrate those capabilities. And now we've got AI control, and I'll put that in quotes of a lot of infrastructure as well, which gets just as complicated. 
 

Nick Lippis: It's very complicated. And I think in kind of in the owner community, there's really kind of a couple of vectors, right? There's like there are those who are doing their own kind of GPU clusters. And that's a smaller group, but that tends to be aligned around health care, pharmaceutical, financials. 
 

Eric Hanselman: Heavily regulated Yes. 
 

Nick Lippis: Yes. And -- but they have the use cases, right, like in pharmaceutical, it's drug discovery. In health care, it's identifying cures and treatments for various different diseases, really great projects. In the financials, it's around fraud mitigation. And clearly, in manufacturing, especially in personal vehicles, it's around self-driving. So, there's a lot of -- so those are really kind of very vertical oriented, but there's orders of magnitude differences like a bank might have a GPU cluster of like maybe 500 GPUs maybe the most like 1,000 GPUs -- where you look at colossal is like 250,000 GPUs right now. So, they're on really different scales. So, there's that part of building the infrastructure. But I think we're most in the owner community are really focused on is around operations and how do you utilize AI to streamline operations, both NetOps and SecOps. That's a really -- that's actually extremely promising and a lot of really good work is being done there as well as how do you integrate LLMs and RAs and there's a whole bunch of kind of projects. So that's what I was saying before, it's a little bit of a hybrid. So, there's a lot of bottoms up. People want to use it, and they're going out and grabbing these technologies and trying to integrate them. And it's also top-down because a lot of the kind of the -- well, the LLMs anyway are coming from the large players and also agents are coming from them as well. But the agent marketplace is a little -- obviously, it's really early, and there's going to be such a great opportunity to write your own agents. And so not quite sure how that's going to shake out yet. It's like a large supplier versus like kind of roll your own, probably clearly a combination of both. 
 

Eric Hanselman: Well, there's so much just in terms of figuring out architectures and exactly how much work is an individual agent going to do? How do you split out those tasks between multiple agents. There's a lot that's there. 
 

Nick Lippis: A lot. 
 

Eric Hanselman: That's one of the things that I think the collaborative nature of OAG is also useful is that you get a chance to contrast and compare what approaches others are taking. Some are going to work, some will work here, some will work there. And actually, I guess, before we get too far down the road, -- maybe you can characterize what the ONOG community is really made up of. And that's one of the things that I always find wonderful is just the breadth that exists. Back in the early days, it was focused on a lot of leading-edge technology adopters. But I mean, if you look at what the ONUG membership looks like today, it's a pretty broad swath. -- you're talking about financial services, pharmaceuticals, manufacturing, it really ranges far and wide. 
 

Nick Lippis: Yes. Yes, we're really fortunate. We do have a large cross-section of the economy. So, I think there's every major industry sector represented, there's locus of concentration in the financial services, pharmaceutical companies, health care, the 3 that I had mentioned earlier on. So, there's a large group of that. And maybe that's because we're in New York coming up in about 10 weeks or so from now. And so -- and you have -- you tend to have a lot of financials there and pharmaceuticals over in New Jersey and clearly, a very vibrant and large health care community. So, I think those -- so like I said, there will be people from media, ABC, CBS will be sending people their people over. There'll be folks from manufacturing, folks from local government. So, there's a wide swath that attend. 
 

Eric Hanselman: Well, and I was just contrasting that because we just come out of -- so just a little bit of background for listeners. ONUG has gatherings that happen spring and fall and spring this year was in Dallas and fall is going to be in New York. those that are interested, I'll put the links in the show notes, so you actually take a look at what's on the agenda. But the other thing that I think is also powerful about ONUG is the working group idea because this is -- it's a group that's really built for collaboration. So, you're tackling some very specific use cases. And it's community-driven and community staffed. And it really does run this whole game. I mean you've got -- you're mentioning some of the big 3 for verticals, but you've also got logistics companies in there. You got all sorts of players. 
 

Nick Lippis: Yes. Well, yes, if we even just think about the Executive Board, there's Pfizer on there, there's Raytheon on there. FedEx is on there. HSBC is on there. Cigna is on there. So yes, you're right. There's a Memorial Sloan Kettering Center is on that Executive Board. And then we have a larger Board and that Executive Board kind of oversees both the collaborative as well as the conferences. And yes, I think -- so yes, I think there's a lot of people that come from different industry sectors, and I think that makes us richer because it used to be that the vendor community wanted to kind of lump all industry sectors in together because they thought they had the same problems. And by and large, I can understand that. But I think that as you start to really dive into it as we have, there are more similarities across industry sectors than there are differences. So, I think that it works. There's like everyone is trying to figure out agentic AI. As an example, their use cases might be somewhat different, but the core technologies are going to be the same. And so that actually -- you mentioned the collaborative and the working groups, Eric. So maybe -- so how they're kind of structured and then maybe I can talk about one of the projects that are going on in there now. Does that -- would that work? 
 

Eric Hanselman: Yes, that would be great. 
 

Nick Lippis: Okay. So, what we found out kind of early on is that you need to kind of separate the vendor community from the user community in the very beginning of these projects. So -- and the beginning is really. 
 

Eric Hanselman: We keep the salespeople out of the room. 
 

Nick Lippis: It's -- well, you're absolutely right, but it's -- there's part of that. And the other part, though, is that even though they may not be salespeople, they might be project managers from the vendor community, what happens is that IT executives aren't -- they don't have the freedom to express themselves because they feel like anything they say that could be relayed back to salespeople or what have you. And then there is a sales kind of opportunity that gets created or might create some issues for them back in their office. So -- so we realized early on, we have to really separate them. And so, what works is we have a group whenever we start a project, that's exclusively IT execs. And the amount of dynamicism and the exchange and the discussion are amazing. And also, the collegiate nature of it. Everyone wants to help each other because they're struggling with the same problem. So, the one project we have is called the Agentic AI overlay reference architecture. So, it's putting together a reference architecture. And so, we have -- Bank of America is on that. eBay is on that. Cigna is on that, Ernst & Young is on there, a few other companies. And we've been meeting -- and so I think we finally have the beginnings of a really good reference architecture. And it kind of shows -- if you think about it, it's like if you have agents within your data center where you kind of control the access that those agents have to data and the tools, that's a pretty self-contained problem. You can probably more than -- you can move on that. You have total control over it. The problem is that as you move out of like your private infrastructure domain and then -- so what happens if you have an agent that's like in a cloud. And that agent now is accessing certain tools and certain inference machines and also a certain amount of data. then that agent needs to get authenticated. You need to control it from a security point of view. You got to make sure that if it's rogue, that it can be isolated, you need to have governance over it. You need to make sure that if there's any PII, it gets scrubbed before it leaves your facility. So, once you take the agent that's in your facility and then you extend it into other essentially domains, the problem gets really complicated. And it's not just there, then you add another domain like maybe the public cloud or the public AI services like Cloud and ChatGPT, right? That's another set of inference machines, another set of public data -- or private and public data and tools that it needs. And then there's kind of Internet services like that companies rely upon like there could be public domain kind of Internet service and you can have agents there. So anyway, all of that gets really complicated even when you're just thinking about a few agents. And then when you think that this is going to be like hundreds, thousands, tens of thousands of agents, how do they -- what is the kind of the monitoring and control structure look like? And then also what is the means of communications between them looks like. So anyway, it's been fascinating to work with that group putting together a reference architecture that people can now hopefully be able to look at and utilize and they'll be able to get insights into how do you now think about this entire Agentic piece. So that's one project. So, what we'll do there is that probably I think it's like September 9, we're going to -- we'll do a big reveal. So, we have various different use cases, and the use cases will be used as thought experiments like, okay, well, how does that use case map into this reference architecture? And then that will modify, I'm sure, the reference architecture. So, we'll have a bunch of iterations on that. And then we'll reveal that to the vendor community. And then there will be the collaboration between the vendor community and the kind of user community on how do we address this? What's missing in the industry, what needs to be done. Some companies will clearly focus on one piece, others on another piece, but how do you stitch it all together? 
 

Eric Hanselman: I mean this is really the challenge, right? We're in environments where these are new architectures. And if we think about some of the capabilities that they've got to bring to bear, these are areas where organizations are in uncharted territory. They haven't had a lot of experience defining use cases, understanding not only what they are looking to achieve from a business perspective, but also looking at the scale of some of this. And this is a very different area in terms of what's required for infrastructure, where this needs to head, a lot of the capabilities that are there. 
 

Nick Lippis: Yes. It's like -- it's basically -- it's having the outcome in mind, right? And okay, well, what value are we going to be able to create by putting this system in place. And then -- so identifying that, and then you have to go through your whole ROI kind of calculations and so forth. But yes, I think this is -- that project, in particular, has been extremely -- actually, all of the projects that we do are extremely fun, but this one in particular, because it's something that everybody is really kind of concerned about, thinking about, want some solutions about. And everyone gets like little snippets from kind of a broad ecosystem out there. So, I thought that was -- that's a really great kind of project that everyone is working on. So, we'll do a big reveal with the vendor community on the 9th, and then we're going to see what can we do to maybe bring it to life for the fall. And it may just be more of, okay, we're going to -- here's the reference architecture, here's a white paper. And then for 2026, we start to really just like one vision that we have, I'm not sure if it's going to work out or not, but have the reference architecture all color-coded and then companies can opt in as to which parts of the ecosystem do they support and then we can give them plaques on their booth at the AI Networking Summit so that people know how they map into this. It would be really great if we just build a large Agentic system and then kind of run some use cases on it. That would actually be a lot of fun if we could do that. 
 

Eric Hanselman: Agents running the entire show, what could go wrong? 
 

Nick Lippis: Is okay. It's like that's what it's all about. 
 

Eric Hanselman: Well, I think in terms of really being able to build an environment in which you can start to provide some clarity in terms of what the capabilities are, where they all fit. I mean that's one of the big benefits. I mean the challenge, of course, is that this is still early days. Organizations are still figuring this out. Yes, a lot of proof of concept. 
 

Nick Lippis: It's early, but I think, Eric, it's coming on so fast. Like the other project that we have, I won't go into the same amount of detail, is a WAN communications API. And this is all about doing procurement for WAN services, which is amazingly laborious, and toil ridden for everyone. We have we've developed -- so first, we started off with, okay, requirements. Those requirements then were mapped into what are the sections of the API. And then someone on the team took it with DeepMind and all their other AI tools coded for it. So, we have -- it's both JSON and our rest API now. And then there's -- and then somebody else on the team used that and also to create a portal. So like in the matter of just a few weeks, we went from requirements to coding the API and now into coding a portal where you can put in the address, the points of the address that you want and then it basically spits out all the options are available. You put in, okay, how much bandwidth, latency, -- and then you get this map that also layers out if there is redundancy in the link, where the overlaps are. It's -- I'm blown away by just how fast it is, and I look at what we did there, and that's like a little glimmer into, I think, how fast maybe the agentic stuff is going to come on. It's like -- it will start in places that aren't as complicated, but it will just -- it will get busy pretty quick, I think. 
 

Eric Hanselman: Well, I mean, there's so much that can be done if you architect it properly. If you're giving agents task-specific capabilities and really being able to target what's actually being taking place. Is something like this that's so procurement heavy. And this is -- it's such a challenge. I mean these are areas where you've got organizations that are spending huge amounts of time just figuring out what's available going company to company to company. It's one of these roll-up things that agents could potentially be particularly good at throughout this process. 
 

Nick Lippis: RFI, RFP, it's like -- just to do it like if you're a large company is probably at least a 6- to like 9-month process. That's just to choose the vendor and then you got to do the implementation. 
 

Eric Hanselman: And there's just so much that has to get integrated with this and going to all the different vendors, figuring out all the different capabilities. 
 

Nick Lippis: Yes. It's like this is -- I've been working with Tony Fernaci, who is now part of the team and said, Tony is like exJPMC. He is such a pleasure to kind of work with. So, we have -- for the first time on our day 0, we have an extensive workshop program that's available. So, all around kind of AI skills and tools. So that's really brand new. So, like a full day, there's probably going to be like 12 or so workshops occurring during day 0. Also, Executive Board meeting happens on day 0 also. And then really for the next couple of days, we have just some really great keynoters, Berta Rodriguez from over at -- she's the Chief AI and Data Scientist over at Pfizer. We have Tom Gillis coming in from Cisco. Jay, the CEO over at Zscaler is coming in. It looks like David Reilly, he's the ex-CIO over at Bank of America, just recently kind of retired from that role. Daniel Conroy from Raytheon is coming in. Actually, Daniel's talk is going to be really, really interesting. He's doing a talk with DXC, and they've been working on one of the hardest problems, I think, within IT, and that is dealing with tech debt. It's hard to change technical debt because it's so expensive. The thing works, right? And it's chugging along and maybe its revenue generating. 
 

Eric Hanselman: It's in place. It's a monolith. 
 

Nick Lippis: Yes, no one wants to touch it, right? But it's a burden to its life cycle management is burdensome. So anyway, what they've done is developed a kind of a plan where you can actually utilize AI to accelerate the refactoring, rewriting of legacy applications. So, you extract a lot of the customer information, which is extremely important in those applications. You now are free to add maybe other kind of records or other services on top of that to create extensions, better experiences for users who are used to using that particular application. And they're having amazing success in that. So anyway, he's going to talk about that. I think that's something that no one's really talked about before. And so, he has a real -- they've done it, they're doing it over at Raytheon with the help of DXC. So, they're going to talk about that. 
 

Eric Hanselman: And this is one of those areas where AI scale can really help. because it can digest where a complex set of interactions and reduce risk in all that process of app modernization. And it's something where if you think about some of the challenges and all this. So often, there's so much business logic that gets built into an application. 
 

Nick Lippis: Yes. Well, you're absolutely right. Like business logic is kind of baked into them and being able to extract that and then manipulate it, add value to it, put it into a better form factor where it's much less costly to maintain. 
 

Eric Hanselman: And if you think about what's necessary, it's a challenge. But again, something that AI can do really well in. 
 

Nick Lippis: So that will be really great. And then there's going to be a focus around Sonic. There's like a Sonic Pavilion. A lot of folks Walmart will be there, eBay will be there who have big Sonic implementations kind of talking about it. We actually are going to have like a quantum pavilion. So, we're going to have like, I think, between 5 and 10 companies that are participating in the quantum space, focused on the large enterprise, and that's very kind of security oriented, as you would imagine, right? So there's that -- but I think most of the kind of the sessions and the discussions are going to be focused around how do you build -- how do you get your infrastructure ready, not only to support the kind of AI workloads that we're already seeing right now, but also what do you need to do around getting ready for agentic. So, there will be a lot of discussion about that. Andy Brown is going to have a really great session. it seems that only about 4% of AI projects like corporate AI projects succeed. 
 

Eric Hanselman: That few 4%. And it's something that we see tremendous challenges. And if you think about the movement for proof of concept to production, it's tough. 
 

Nick Lippis: Yes. Well, yes, you guys -- you have a lot of that insight. The is going to pull together a really great panel. And okay, well, what goes wrong? Why does it go wrong? -- why to look out for it? And then There'll be a session right after that about where the rubber hits the road. So, this is going to be from the 4 presenters, those who have actually... 
 

Eric Hanselman: Those precious few. 
 

Nick Lippis: Yes. So anyway, I think it's -- it will be a really great -- it's really 3 days if you include the workshops. 
 

Eric Hanselman: It's great to see the kind of work that you're doing to raise the overall level of understanding because as you come into this -- into the conference, being able to have workshops beforehand that are going to give people a grounding means that you come into this with a common understanding and you can go at these problems with a common language. And that's a huge benefit. 
 

Nick Lippis: Eric, like yesterday's meeting, Monday's meeting, it was Monday, on the WAN API working group, I have never felt skill inadequate until that meeting. and seeing what these folks have done with Claude in such a short amount of time, I'm like, oh my God, I am like Jan son is the CISO of FedEx, who said this at one of our conferences that AI will not replace you your job, but people with AI skills will. That came so crashing down real on Monday. It's like, wow, that's a necessary skill. We've got to be there. Yes. 
 

Eric Hanselman: And this has been great. If you -- especially as we start looking at that, maybe we should do is a wrap up after the event itself. So, get you back on and we can talk about everything that went down and get things set up for the spring. 
 

Nick Lippis: Happy to do that, yes. 
 

Eric Hanselman: Well, it was great having you on, especially an important topic like this. I really appreciate it, Nick. 
 

Nick Lippis: Eric, it's always a pleasure. I love talking to you. I love the practice that you've built like over all these years. You're one of the smartest, if not the smartest analysts that's out there. So, thanks so much. It's always a pleasure. 
 

Eric Hanselman: And that is it for this episode of Next in Tech. Thanks for staying with us. And thanks to our production team, including Sophie Carr, Feranmi Adeoshun, Kyra Smith, and Adam Kovalsky on our marketing and events teams. Please keep in mind that statements made by persons who are not S&P Global Market Intelligence employees represent their own views and are not necessarily the views of S&P Global Market Intelligence. Join us for our next episode where we're going to be looking at data integration concerns. I hope you'll join us then because there is always something Next in Tech.