(So you can skip ahead, if you want.)

00:00 Welcome from AutoCon 4
03:38 FlowAI: Bringing AI Safely Into Infrastructure
04:31 Governance as the Prerequisite for AI
06:51 How MCP Limits Blast Radius & Adds Predictability
07:44 Real-World Example: Agents vs. Brittle Workflows
11:32 The Unprecedented Speed of AI Innovation
12:49 MCP Server, MCP Gateway, & Platform Architecture
18:11 Bring-Your-Own-Model Flexibility Across Public & Private LLMs
23:59 AI for Troubleshooting, State Analysis & Network Understanding
30:33 Why CLI Still Dominates & Why That Must Change
35:04 Philosophy: Make AI Simple, Usable & Operational

Peter Sprygada • 00:00 

And you know, we should be using the right tool for the right job. Now, to your point about governance, absolutely, you know, governance is is no question the key here. Um, you know, we’ve been able to leverage Idential platform, which has got you know 10 plus years of uh layers built in it just to handle governance when we were just simply doing static workflows. And that’s what we’ve been able to tie our AI strategy to. So we’ve got a hardened platform that really allows an organization to get the traceability, to get the R back, to get the uh secrets management that they need to start to trust AI. 

William Collins • 00:44 

Welcome to another episode of the Cloud Gambit. We are still here in Austin. So the vibe is incredible. There’s tons of engineers. There’s tons of leaders. Um you wouldn’t think the energy could get any better from Prague, but wow. Uh it is electric. 

William Collins • 01:04 

So we’re really enjoying it. We’re here on site and um I’m here with my co-host Eyvonne. 

Eyvonne Sharp • 01:09 

Hello, William. How are you doing? I’m a little squeaky. I’ve been doing a lot of talking, some late night dinner last night. So we’re gonna try to keep the voice available for the rest of this conversation. We’ll see how it goes. 

William Collins • 01:21 

There we go. And uh with us, we’re excited to have um a sponsor that’s really been uh if you think about the origins of autocon and the question that autocon’s trying to answer, why didn’t network automation take off? You know, we’re still here, 2025, a lot of things are CLI driven. And one of the big sponsors and the big advocates behind this movement, Itential, we’re happy to have them as a sponsor. We have their chief architect, Peter Sprogata. How are you doing today, Peter? 

Peter Sprygada • 01:51 

I’m doing fantastic. Thanks uh for having me on the program. Yeah. So what do you, you know, how has the conference been for you so far? The conference has been absolutely insane. I’ll be honest with you. Um, you know, I’ve been I’ve been doing this for so long. 

Peter Sprygada • 02:04 

It’s like I can’t walk from you know one side of the room to the other without running into 10, 15, 20 people. Um, but it’s it’s great. You know, it’s great catching up with with old friends. It’s great reminiscing about uh, you know, how we’ve gotten here. And the conversations have been tremendous. Uh, you know, the the fact that that um you know we are finally coming together as a as a um you know, as a group and and being able to really just talk about network automation and we’re not just a bolt on somewhere else. So it’s it’s really been a great conference. 

William Collins • 02:34 

Yeah, absolutely. And you, you know, you’ve been like you said, you’ve been in this space for so long. And uh, you know, looking back at even the the architect days at Ansville, Ansible for networking, you know, and Moving the CLI and adopting something besides just typing things in a terminal, hitting enter, looking at outputs, and really changing that model to not just support systems, but to support network devices as well. Looking at where we are now, have we really changed a ton? You know, how has it evolved? What do you think? 

William Collins • 03:08 

Wow. 

Peter Sprygada • 03:08 

Um, we have changed a ton. Let me start there. There is no question about it. Now, are we still dealing with a lot of the same problems that we were dealing with 10 years ago? Absolutely. We absolutely are. And, you know, in some ways it’s, it’s, it can be frustrating, but we have evolved so much. 

Peter Sprygada • 03:24 

Um, just in a recognition as a as a group and being able to now share information and kind of talk through some of these things and share best practices. You know, I think that’s where a lot of the the new stuff is really coming into play. Um, you know, for once we’re we’re finally um, you know, kind of going through this as a collective group and and really sharing information, sharing best practices, I said. Um I think that’s really the big change that I’ve seen. 

William Collins • 03:48 

So there’s been some exciting news. So you were on stage and just like in Prague last year, Itential launched the MCP server. And now we have Flow AI. Um, great announcement, great presentation. So, do you want to just unpack just a high level? Like what was this announcement about and how is it relevant right now? 

Peter Sprygada • 04:09 

Yeah, absolutely. So, you know, Flow AI is a culmination of about the last 12 to 15 months worth of work. And as I said in my presentation, you know, I’m I’m a network engineer by trade. That’s where I got my start. And when this whole AI thing came about, I was like, no way. Uh-uh. AI is not touching infrastructure ever. 

Peter Sprygada • 04:29 

And the more I got comfortable with the technology, um, the more I really started to see the things that we could start to do with AI as we started to attach it to infrastructure. So, as an outcome of that, we built a uh new framework called the Flow AI framework, which is composed of a number of different products that are going to become part of the Itential platform. And it’s really all about how we can start to securely, safely, with governance, with traceability, start to attach AI to infrastructure in a way that’s going to allow us to sleep at night. 

Eyvonne Sharp • 05:02 

You said governance. Can you talk a little bit more about that? Because I know that folks really get concerned either if they’re in a regulated industry, or even if they just want to be sure that the AI is not gonna, you know, take over their network, like some sci-fi movie that we’ve they’ve watched for the trust network automation that much yet. 

William Collins • 05:22 

There’s a lot of folks that don’t trust network automation. How or why would they let AI touch anything? 

Peter Sprygada • 05:27 

Well, I think it’s a great point. I think there’s a couple of things here. You know, 1st and foremost, you know, kind of what I talked about in in my presentation is we have to recognize that AI is just a tool. Let’s let’s start there. I mean, AI is a transformative tool, it’s a revolutionary tool, but it’s just a tool. And, you know, we should be using the right tool for the right job. Now, to your point about governance, absolutely, you know, governance is is no question the key here. 

Peter Sprygada • 05:51 

Um, you know, we’ve been able to leverage Idential platform, which has got you know 10 plus years of uh layers built in it just to handle governance when we were just simply doing static workflows. And that’s what we’ve been able to tie our AI strategy to. So we’ve got a hardened platform that really allows an organization to get the traceability, to get the R back, to get the uh secrets management that they need to start to trust AI. But as I tell everybody, right? This is not, this is not uh one size fits all. And and just like we use any other tool, we need to think through what are the right use cases to use AI for. And that’s how we we start to move this forward. 

William Collins • 06:30 

So you mentioned this pedigree of like having governance and you know, you’ve built up all this great technology over the years. Can you use that with the AI that you’re building? Like, can you recoup that investment and leverage it with Flow AI or with the new things you’re building? 

Peter Sprygada • 06:46 

Oh, without question. So, and that that’s really kind of the key to what we did. You know, we’ve been in the platform business for a very long time. And so we built Flow AI on top of platform so that we can leverage all that because it would be silly for us not to. It’s already been battle-tested, battle-hardened, and some of the biggest infrastructures in the world. And so we are absolutely reusing all of that as part of the Flow AI framework. 

William Collins • 07:08 

So, one question I have is this this trust thing, hallucinations, getting towards a more deterministic outcome and not having to just, I mean, I’ll put it this way. Why couldn’t we just use RESTful APIs for everything? 

Peter Sprygada • 07:23 

We can. There’s absolutely that’s fine. So the big turning point for me was really with the introduction of MCP. That changed everything. Once I saw what MCP was capable of doing, and the fact that we could use MCP to attach tools to agents and therefore put guardrails in place so that agents, even if they hallucinate, we’ve at least got some level of control over the blast radius. Now you couple that with some of the RBOC and some of the security controls that were already native to platform. And that’s actually how we can create a very safe environment for a lot of these AI agents to ultimately run in. 

William Collins • 07:58 

So I mean, give us I’m trying to visualize this in my head. So would you want to uh walk us through just an example of like 1st of all, how would a customer use ital or build a workflow or automate uh maybe configuration for a router or back up the config and then How does that change when you throw an MCP? 

Peter Sprygada • 08:16 

Yeah, I’ll use my favorite, my absolutely favorite use case. And it’s 2025, and it’s sad that this is probably the most common use case still for network automation, and that is software upgrades. And we still haven’t quite figured it out. But that being said, okay, so let’s kind of talk through this here a little bit. Uh I run an infrastructure and I’ve got, like most people have, a multi-vendor approach. I’ve got Arista, I’ve got some Cisco iOS, I’ve got some Cisco Nexus, and I want to be able to do software upgrades. So I build workflows, and each one of those workflows is going to have a specific branch, if you will, that’s for each vendor, because each vendor, of course, knows it a little bit differently. 

Peter Sprygada • 08:52 

And the problem here, though, is that my parent workflow, my controlling workflow, is very statically defined. Everything works great, no problems until that day when we go out and we acquire a new company. And of course, what does that new company have? A network full of Juniper devices. And so the only way I can start to apply my software upgrade workflow is now I got to go build all of the Juno specific workflow stuff. And I got to go touch my parent workflow. That has the uh risk of introducing errors, not quite working. 

Peter Sprygada • 09:27 

Sometimes it’s very brittle. That’s where the big change comes into play. We can now focus on building very bespoke workflows that do one thing and do it well. The old Unix philosophy, right? I mean, it’s time tested um way to build out uh pretty much anything. Um so, but anyway, so so now instead of having that overarching brittle workflow, now I can let an AI agent start to reason through this a little bit. And so now when I have that new Juno’s workflow, it can simply understand oh, that’s a Juno’s device. 

Peter Sprygada • 09:58 

I know how to deal with this now. And I didn’t have to go and change all of my workflows to support that use case. 

Eyvonne Sharp • 10:04 

So, what I’m hearing is really as much about toil reduction as I am anything. It’s just incredibly reduces the amount of toil that we have to take. And it instead of building it by hand, we have a tool that helps us do that. 

Peter Sprygada • 10:20 

You’re exactly right. You’re exactly right. And not only is it help in the reduction of toil, but it but it also is making it safer to use in some cases. Um, you know, agents, the the beautiful thing about agents is when they don’t hallucinate, they tend to do the right thing. And especially when we can control the set of tools that they have access to, it becomes a very powerful way for the agent to be able to go off and do stuff. And therefore, now we can not worry about, not worry as much anymore about breaking our workflows. But more importantly, it also gives us the ability to focus on more important things than network engineering, right? 

Peter Sprygada • 10:53 

Software upgrade is an interesting case, but I really don’t want, you know, my high-priced network engineering talent worrying about that problem. I need them working on network engineering problems, not software upgrade problems. 

William Collins • 11:03 

So you launched the MCP server in Prague last year, and uh it’s on GitHub. Anybody can go out and look at it. And um, you know, one of the things that we were talking about recently is uh you were kind of going through this, you know. peak stability of Itential’s platform. It’s stable and it scales. And we all know that you can’t have this massive scale. And, you know, when you think about adding intelligence and everything, they’re two different things. 

William Collins • 11:32 

So I could locally create things on my machine and I could do some really cool stuff. And that’s great. But there’s a difference between connecting to five things in container lab and then having an enterprise fleet, federated systems, all these things. So one thing that you were telling me is you had actually integrated Tune, a new tool into the MCP server. So you you seem to be balancing this rapid innovation on the MCP side because I mean everybody that’s listening knows this stuff moves fast. It moves really fast. 

Peter Sprygada • 12:04 

That might be the the biggest understatement I’ve heard in a long time. You’re absolutely right. This stuff is moving. I have never in my I might I’ve been doing this for 30 years. I’ve never seen technology move this fast. And when I say technology move this fast, innovation has always moved fast. But the technology and in its its push to move it into production is going at just crazy speeds right now. 

Peter Sprygada • 12:25 

And I find that just fascinating. It’s just amazing to watch how fast this is moving. But yeah, you’re absolutely right. You know, all of the pieces that we’re putting together whether it’s the open source MCP or the new flow AI stuff, um, the ability to add net new feature and functionality has to come and it has to meet the market. And this is the speed the markets work at right now. 

William Collins • 12:51 

So whenever I, you know, I worked in enterprise for a while. And whenever I would look or evaluate a product, you know, I’m always thinking of like, what what goes where? Is part of this in the cloud? Is part of this on-prem? Like, where does stuff live? How do you launch it? Like there’s all these questions. 

William Collins • 13:06 

So I think we all know that, you know, for those that have paid attention to Itential, you have platform. Um, so you have the whole platform component, but as far as MCP, like how would you how would you launch it? How would you use it? You know, what does that look like? 

Peter Sprygada • 13:21 

Yeah. So so there’s a couple of components here. So we’ve got the MCP server northbounded platform. And that’s what allows an organization to integrate their external agents into through platform so they can attach their infrastructure to it. We also have the MCP gateway functionality. And that one is truly exciting for me because that opens up platform’s ability to work with any MCP server out there. And, you know, at last count, we’re at what 649 billion MCP servers or whatever it is this time. 

Peter Sprygada • 13:50 

I don’t know. I’m making that up, but you might be close. And the point is that there’s an MCP server out there for everything these days. And being able to tap into that ecosystem with platform and still have all of the enterprise controls that platform brings is really a big win for our customers. 

William Collins • 14:06 

Did you say MCP gateway? Is that like an API gateway? Like is there a difference? Talk talk to us here. 

Peter Sprygada • 14:12 

Yeah. No. So, you know, I think I think that is, but I’m glad you asked that question because I think that is a big problem. I think a lot of people make that that very that make that correlation. And MCP gateways are very different than API gateways. Um, you know, and I think that an MCP gateway um really is focused on its ability to connect into your reasoning layer, right? An API gateway is really just there. 

Peter Sprygada • 14:39 

Dare I say it’s more as a I shouldn’t say it this way, but it’s more of a proxy type of a situation in reality, right? So they’re they are very different purposes um and they serve different purposes and they have different needs and they have different features. 

William Collins • 14:53 

Yeah, so I guess what I’m hearing is, you know, we going way, way, way back, you had like SOAP, uh, suds, all these, you know, frameworks that folks would use. And then REST came along. I mean, I don’t know. Well, actually, REST was kind of already there, but I don’t think it was adopted. I can’t remember exactly, but um, you had this proliferation of APIs everywhere. They became the hot stuff. And then out of nowhere, it’s like, okay, we’ve got this stuff everywhere. 

William Collins • 15:19 

We we need a gateway to manage, we need some sort of centralization. We need a little bit of control, you know. So that’s kind of where API gateways came along. And I guess looking back at history, do you think it’s like a repeat of history in a sense? Because you have all these MCP servers, and now you have this idea of an MCP gateway. 

Peter Sprygada • 15:38 

I mean, yes. The reality is this whole AI journey is repetitive to what we saw with automation. So let’s let’s think about this, right? Um when automation started. Now I started my automation journey back in 2012. Um and when we started automation then, and especially in the network space, the 1st thing we would do is we would build a, you know, we use automation to build a network config, a device config. The problem is we didn’t push it to the device. 

Peter Sprygada • 16:02 

What did we do? We all sat there and we stared and looked at it and went, okay, yep, that looks right. Now I’ll go ahead and push it. And as we got more and more comfortable with the technology, we started to let automation do a little bit more and a little bit more and a little bit more. I think AI is going to take the exact same trajectory. It’s just going to do it much, much faster. 

Eyvonne Sharp • 16:18 

Oh no, that’s I was just gonna say that’s really insightful. You know, that that that trajectory of, hey, we’re gonna play around with it, we’re gonna get comfortable, and then we’re gonna, you know, loosen the reins and give it more control over time. I think that’s really insightful. 

William Collins • 16:32 

So this this Flow AI agent builder thing. So we had we talked about the MCP. We have the platform. So if I’m an existing customer today and I I want to, you know, check this out. Like, how do you how do you use it? What do you how do you access it? 

Peter Sprygada • 16:47 

Where is it? So so it’s a new application that gets loaded onto Attention platform and that makes it available to the customer. But one of the things that we’re doing in this particular with with Flow AI is we’re working in concert with some of our biggest customers and we’re leading leading, wow, we’re letting them help drive the roadmap. Because the reality is that again, we’ve already talked about the speed at which AI is is evolving. The use cases are evolving just as fast. And it’s naive of me to sit here and think that I understand everything that that AI is gonna be applied to in infrastructure. I don’t. 

Peter Sprygada • 17:19 

And so we’re really leaning into with some of our biggest customers and how they’re helping us to really define how Flow AI will continue to evolve so that we’re delivering features that organizations can find benefit from day one. 

William Collins • 17:31 

I love that. So I mean, and basically instead of taking the approach of, hey, I think we know what we’re doing. We’re going to build this out. We’re going to roll out some product and you know, push it on customers. You’re doing that quick iteration, you know, that feedback loop. You have that feedback loop going, you’re taking in that feedback and it’s growing with the way your customers are growing and also the way the industry is growing. 

Peter Sprygada • 17:53 

I mean, yes. And the reality is we should always build products that this way. We don’t, we should. Um, but you know, I think that what’s interesting about AI is so much of the company ethos of any organization is in AI now, right? As they start to use AI. So it’s it’s um absolutely crazy for me to think that I’m gonna understand best what features they need because everyone’s a little bit different. Their ethos is a little bit different, their organic knowledge is gonna be different. 

Peter Sprygada • 18:19 

So that’s why we’ve really taken this approach, and it really needs to be a partnership and and will be a partnership going forward. 

William Collins • 18:26 

What about so I know that you know with MCPs in general, you can kind of you you use your own models, you can pick, it’s pretty flexible. So this uh, you know, FlowAI, like is that gonna be the same case? Can you bring your own models? Are you hosting a model? Uh what does that look like? 

Peter Sprygada • 18:41 

So, yes, yes, and yes. Um again, this was another decision that we did make very early on when we we put together the Flow AI framework. So, because platform and ITential platform has always been a federated type system. We tend to not be the source of truth or anything. 

William Collins • 18:58 

I shouldn’t have used the term source of truth, but I did. Here we go. This is the MCP conference now, not the source of truth conference. 

Peter Sprygada • 19:03 

I know it, I know it. As soon as I said it. Um, but uh, you know, I we’ve always been really good at federating systems. That’s one of the key, that’s one of the reasons why customers love what we do. We took that exact same approach for LLM. So a customer can connect any or multiple LLMs to platform. And again, they get all of the security and the control to be able to decide who can do what with which LLMs. 

Peter Sprygada • 19:27 

So whether it’s they’re using public services or they’ve got their private LLM, they can control not only who gets access to it, but what agents get access to it. And therefore, that again, it’s all about keeping the customer in control. 

Eyvonne Sharp • 19:40 

Well, and in a world where we weekly hear new models, like this week it’s Jim and I, right? Next week it’ll be open AI. They’re their models are being delivered so quickly and built. It’s really important to have the flexibility in the platform to be able to innovate and grow along with the new models as a release, I’d imagine. 

Peter Sprygada • 19:57 

I could not agree more. And it what’s fascinating to me is um, so we spend a lot of time building prompts and context. And what fascinates me is how we can take some of the same prompts and run them through different engines and get wildly different results. Um, so yeah, giving the customer the ability to, you know, choose the right model for the job is really a big key to I think Floyer’s success. 

William Collins • 20:22 

This seems pretty big from a just an industry perspective. Like it seems kind of like uh, you know, ahead of the pack because if I’m a network engineer and I’m listening to this right now, it’s like, oh, I don’t want to, okay, I didn’t even want to touch automation. I don’t want to touch AI. You know, what is what is the trust look like? You know, are customers actively asking for this? You know, what do you think the proliferation to to the enterprises is gonna look like? 

Peter Sprygada • 20:48 

Yeah, I think, you know. Each customer’s journey is different. We we all recognize that. No one’s going to take the same same path here. But you know, we actually have a number of customers today that have invested in building their own agentic layers. Um, it’s not true of every customer, and I agree 100% with what you just said. There are going to be those out there that will say, This is not for me. 

Peter Sprygada • 21:09 

And that’s okay. That’s all right. Um, I do think, though, that The transformative possibilities, though, with AI, we’re only starting to scratch the surface. Because here’s the thing. Right now today, we’re doing, and again, this is that same playbook from excuse the pun, the same playbook from the automation world, right? Is we’re starting, you know, AI is going down this path of trying to solve the same use cases over and over and over, right? 

Peter Sprygada • 21:34 

The same operational use cases. Here’s what we haven’t started to get into yet is how do we leverage AI to make troubleshooting a network better and faster? How do we leverage AI in our network design, right? These are all areas that I think AI can have a very profound effect on really how we as network engineers go about our job. So it doesn’t always have to be just about touching the device. Sometimes maybe it’s just for troubleshooting purposes. Sometimes maybe it’s for doing design. 

Peter Sprygada • 22:01 

So just different ways we can start to think about using this technology. 

Eyvonne Sharp • 22:04 

Well, and I said I think about, so let’s think about the mobile era, right? The 1st thing we got were these mobile devices. But then over time, as mobile grew in popularity, then we got 2nd and 3rd order technologies to go with that. There would never been an Uber without mobile, right? But we are in this Steve Jobs standing on the stage holding an iPhone phase of AI innovation, right? And so there are 2nd and 3rd and 4th order evolutions that are gonna happen. But if you didn’t have an iPhone, 

Eyvonne Sharp • 22:42 

You were locked out of that. And now you can hardly file your taxes without a mobile phone. Right. And so I believe in the next few years, and it may even be sooner than that, if you there’s going to be tons of capability that you just can’t unlock if you’re not, don’t have the capa capability to do that in your environment. 

Peter Sprygada • 23:02 

Oh, I could not agree more. I, you know, I think you’re absolutely right on that. Um, and and again, it goes back to this idea that AI, it’s AI’s ability to. Contextualize and understand what infrastructure is doing and do it at speed that we can’t do as humans. You know, I think about something as simple as, and this is something I’m playing with right now in my lab. Um, you know, what happens if I feed the entire OSPF database into my AI model, right? Now all of a sudden I can troubleshoot my network a whole lot faster, and AI can immediately tell me where my hotspots are. 

Peter Sprygada • 23:36 

It can tell me what happened in the case of an outage, it can tell me how to fix it. 

Eyvonne Sharp • 23:40 

Ultimately, it can just fix it itself. And then what happens when the model can also understand current running state, right? Oh, from your network, right? Oh, really? I have config and now I can understand state. All of a sudden it can draw some really great conclusions. 

Peter Sprygada • 23:54 

It really can, you know. And I think that, yeah, thinking back to to, you know, when I was 1st getting started in my network engineering career and, you know, managing, you know, I had a short stint where I was actually, I was the customer. And at that time, I was part of a network engineering team that run a global infrastructure. And, you know, invariably, you know, and and I I hope a lot of people can can um relate to this, but there was always that one or two or three routers, you’re right, that had been running for the last 10 years. No one has any idea why it’s there, but they know if they turn it off, everything breaks, right? Now we can let AI kind of help us through those types of scenarios. And so that really is, again, another area where where it becomes that game changer. 

Peter Sprygada • 24:34 

And I think you’re absolutely right in that journey in that when we start to leverage this technology, it will get to that point where it’s going to be an absolute must. 

Eyvonne Sharp • 24:43 

Well, and I just want to shout out a 15 year up time on your router is not the flex you think it is. I think it’s important that we say that. 

William Collins • 24:51 

Agreed. So the, you know, one of the things that I’ve heard a ton just walking around this conference, it’s actually been uh the highlight of many of the conversations I’ve been in is uh where are the big vendors? You know, what’s happening with the big vendors right now? Uh do you think that this is going to be a thing where you’re gonna have platforms like Itential that are almost like a gentic net op Switzerland that connect everything? Um, and when you think about, you know, some of the things I have seen, and and you’ve probably seen a lot of this too, is um one thing that some vendors are doing that have platforms is they’re just taking that MCP, they’re not open sourcing it, they’re not putting it anywhere. They’re they’re building it into their platform and they’re, you know, exposing an endpoint, and that’s it. Um, what what do you think? 

William Collins • 25:38 

Um, is there anything you can tell us about like that pattern versus what you’re trying to do? 

Peter Sprygada • 25:44 

Well, to to to kind of hit the 1st part, yes, obviously I’m I’m hoping that’s the case uh with Itentia, that we become, yeah, that the democratized uh you know integration for AI. Um, you know, and and you know, we do know, I mean, look at what just happened at Cisco Live with their announcements around some of the AI work that they’re doing. But unfortunately, you’re right. There a lot of that’s being done behind closed doors, you know, um, which is is unfortunate, but it is what it is, and and and we’ll we’ll kind of work ourselves through it. But I do think it’s really important, and you kind of brought up the MCP case. You know, I think that it’s really important. I I talk a lot about the fact that I think mostly the world’s doing MCP wrong right now today. 

Peter Sprygada • 26:20 

Um, you know, MCP is a protocol that we can use to bring value to the LLM. And unfortunately, what many organizations are doing, they’re just simply taking their APIs and they’re rebuilding them in MCP. Well, we haven’t done anything then, right? It’s still just the same API. Um, one of the things that we did very differently in our MCP implementation is we recognize that LLMs allow us to work with things in English language, obviously. Um, APIs aren’t built that way. APIs are built with unique IDs and whatnot. 

Peter Sprygada • 26:50 

Well, I as a human don’t remember that when I want to open up a document, I gotta go get 643521594, right? I don’t remember those things. Um, why should I make my LLM have to do that as well? Through MCP, we can make those translations so we still get a robust machine driven API, but now we can start to work with our things in English language. And really, isn’t that kind of what the whole point of this is all about? 

William Collins • 27:12 

Love that. I love that answer. One thing I’ve just got to ask. So, I mean, if you talk to most network engineers on the street right now, they probably heard um about Ansible or what a playbook was because of you. You know, a lot of them know who you are. Like that’s how they they figured out what Ansible was and that they even wanted to, you know, automate a network. And then one thing that we’ve learned and why this conference exists is the the culture and there’s a stigma to just keep doing things the same way. 

William Collins • 27:40 

How did you, as someone that you know, really brought a lot of innovation to just automation in general. You’ve been a gigantic part of the community. Um, you’ve been very giving with your time for many. How did you go over this um hurdle into this to embracing MCP, to embracing AI? 

Peter Sprygada • 28:01 

Well, you know, for me, and and this was true even for Ansible, it’s the same answer. Um, it starts with curiosity, right? It starts with I mean, I mean, I’m a technologist 1st , right? And I’m curious as to different ways we can play with technology. Yeah, and what I found in the automation case and what I also believe in in the AI/MCP case is that just start doing stuff. Don’t try and make it perfect. Don’t worry about you know what your end goal is right now, just play with it, just like you would play with any other technology, right? 

Peter Sprygada • 28:29 

And what happens is then you start to see the pattern start to merge. You start to see what you can do. And the more that you start to see that, the more you want to do it. And now you get that snowball rolling. It was the exact thing I used to tell network engineers all the time that wanted to start with Ansible. I’m like, if if all you ever do is automate putting a description on an interface, that’s still automation. But what we found is that once you did that, you realized, hey, this is kind of cool, right? 

Peter Sprygada • 28:53 

Now it’s that’s when the ball gets rolling. And I think that we’re gonna see the exact same thing here. 

William Collins • 28:58 

So one thing I have to ask, and something that has been just so painful with networking in general. So if we think about everything being CLI driven and it took so long for when we still aren’t there for network devices to have, you know, APIs and such, but integrations are so hard between things at a software level. So if you have a network product that needs to integrate with a visibility product, you know, or you know, all these AI 1st uh telemetry products coming to market. Um everything you’re talking about sounds so good, but does it help? with integrations with other vendors, other things, maybe things that you build. Uh what does that look like? Because if you think about I I just know historically, like you could put something in the ear of a vendor and maybe it gets to their roadmap at some point. 

William Collins • 29:55 

Maybe it doesn’t. Maybe they it just gets lost. Maybe you’re waiting a year, maybe you’re waiting five months. It’s just different. And you know that’s that can be um operationally impacting in some circumstances. 

Peter Sprygada • 30:08 

So there’s a couple of things here, you know, when I think about this. Um, you know, 1st and foremost is this idea that um AI is gonna blur a lot of lines. And I don’t think we know where those lines are gonna blur just yet. Um, I think we’re gonna rapidly move into a world where products aren’t quite so rigidly classified. Now, this is just a guess on my part, but the reality is this AI is opening up a lot of doors right now. You know, do we need dedicated observability? Do we need dedicated orchestration? 

Peter Sprygada • 30:38 

Do we need dedicated source of truth? I don’t know. Um, you know, I think that AI can really start to, like I said, tear down some of those those barriers. Um, and we were going to start to see hybrid products show up on the market. The other part of that answer though, and and this is the part that saddens me very much is that here in 2025, the de facto software interface for a network device is still the CLI. Let’s be honest. And and that part is unfortunate. 

Peter Sprygada • 31:02 

Um, it is what it is, and you know, we’ll continue to to work our way through it. I don’t think it’s gonna change anytime soon. But I also think that we have to start to recognize that network as network engineers, and this is a path, a personal journey I went on, is that back when I ran networks, my value was not typing out CLI commands. It my value was understanding how networks work and being able to engineer those networks. And I think that that as we can hopefully get there and more and more individuals be more become more comfortable with API constructs with software engineering paradigms. It doesn’t mean they have to be developers, but they do we do need to recognize that the CLI cannot continue to be our de facto software interface. 

Eyvonne Sharp • 31:44 

Well, and the other emerging challenge that we’re going to continue to have is scale. When a lot of us 20, 30 years ago started out in this industry, we may have had five, 10, 15, 20 devices. All of that’s still manual. Possible to do manually. But as The connectivity of our world has grown. There’s just so much scale. 

Eyvonne Sharp • 32:10 

We’re not going to be able to do it any other way. So it’s not a matter of do you have value? Do you not have value? It’s can you even get the job done without it? And I think uh the further we go, we’re gonna see that the winners are those who know how to get that job done and get it done efficiently. I absolutely love that point. 

Peter Sprygada • 32:29 

I I did not thought about it in those terms before, but you’re absolutely correct. At some point in time, these infrastructures have gotten so big, I physically do not have enough time in the day to log into a thousand or 2000 or 10000 machines to make a change. You’re absolutely right. 

William Collins • 32:45 

So I just have to ask, and this all sounds like it would be really expensive. Like, is there gonna be a, you know, just a trail to this big money pot that I Tensal is gonna want to snatch up? Like what does that look like? How do you even begin thinking about calculating cost for this? Because it all goes back to tokens at some point, in a sense, on the model side. But if your customers are bringing their own models, you know, that cost would be on them. So, how do you even differentiate, you know, where the cost responsibility lies? 

Peter Sprygada • 33:18 

So, so let’s be honest, tokens are the new currency, right? That’s the reality. That’s the new world we’re gonna live in. Tokens are the new currency. Um But kind of when I talked about the fact that we are working very closely with some of our biggest customers and some of our smallest ones too to take this AI journey together, that’s actually gonna be part of the conversation. Let’s talk about what is the right consumption model, what is the right commercial model. 

Peter Sprygada • 33:43 

The reality is, and I’ve experienced this, you know, everywhere I’ve ever been, customers don’t mind paying for things that they get value from. They really don’t. Um, as long as they see the value, they see the return. So we’re gonna go through this journey. We’re gonna talk to our customers about it. And ultimately, you know, as we look to GA, going to GA with this product next year, that’s when we’ll announce all the commercials. I don’t know what it’s gonna be. 

Peter Sprygada • 34:05 

I don’t know what it’s gonna be yet. We’re gonna work through that together. 

William Collins • 34:08 

Yeah, and I didn’t I don’t think that’s just titential. I think a lot of uh, I mean, hey, this is so new. I mean, when so MCP is gonna be a one-year-old uh next week, right? One year old. They grow up so quick, don’t they, Peter? Yeah, they do. 

Peter Sprygada • 34:23 

I didn’t even think about that. Do you want to have a party or what? 

Eyvonne Sharp • 34:25 

Wow, we do need to have a party. Happy birthday. Are we gonna have MCP with cake all over it? Like that’s like the you know, that’s that’s what I’m going to. It’s like we’re gonna put it in a high chair and let it, yeah. I love it. 

William Collins • 34:39 

Yeah. And I I just, you know, you think about so companies like new companies coming to market, they get a lot of funding, and yeah, they can move quick, but it just seems like uh it’s taken on a new, just an incredible velocity. They seem, you know, very ahead right now. And this type of stuff is really hard to rethink. Like, okay, we did one thing, but now like this is the new hot thing. But it does seem like a natural evolution from platform automation, orchestration, really, federated systems, like having all the mechanics in place. And it just seems like a natural fit. 

William Collins • 35:12 

It doesn’t seem like there’s a lot of friction to really layer that on to what the existing offering does. Whereas with other products, it might be a whole refactor. You have to do so much work because you have to rebuild the product to fit the new narrative and the new value proposition uh for the market. 

Peter Sprygada • 35:30 

Yeah, you know, I’ve got I’ve got two philosophies when it comes to building products. Philosophy number one is get rid of the magic, right? Make it something that people can consume, that they can understand that they can use, that they can get value from, right? That’s philosophy number one. Philosophy number two is make it simple. I can remember some years ago when I was at Aristotle. I remember sitting down, uh, we were we were sitting, I was sitting down with Jay Shree and we were just talking. 

Peter Sprygada • 35:59 

And she said something to me I’d never thought about before. And she she’d asked me the question, she’s like, you know, if we look back on the networking industry and old people like me, we remember the router wars, right? And there was a new router startup, seemed like every other day. And by and large, we can say Cisco won the router wars, right? They are the dominant networking company. And you start to ask yourself, why? What made Cisco so special? 

Peter Sprygada • 36:21 

Because they were not the most technically advanced router out there, not by a long shot. So what made them, why is it that they won? And it came down to two very simple things. One, that they worked with everything out there. It didn’t matter if you were IP, IPX, SNA, uh, DeckNet, Apollo for people who remember Apollo. Um, they worked with everything. And two, the operational experience was well thought out and it was consistent. 

Peter Sprygada • 36:49 

Right. And that’s really what she had surmised is the reason that Cisco was um, you know, had won those rudder rows. I really took that to heart and I really tried to model a lot of, you know, when I build products and when I build solutions with that kind of mindset. It’s got to work with everything. People have to have choice, and it’s got to be easy to use. 

William Collins • 37:08 

Yeah, I love that. User experience is so key these days because if I go out and I try to use a product and I have to take a whole business stage just to get it set up and start working with it, that’s just uh it’s hard. It’s a deal breaker. 

Peter Sprygada • 37:20 

Now here’s the sad part of my whole commentary here. Um I still write code every day, and my preferred ID is still VI. 

William Collins • 37:27 

Hey, me too. Hey, that’s not bad. It’s not a bad thing. I don’t know why people hate on VI so much. I don’t either. 

Eyvonne Sharp • 37:33 

I cut my teeth on VI uh in college and Vim forever. This is the perfect panel right here. 

William Collins • 37:39 

I love it. Fantastic. Absolutely. Um, yeah, so thank you for that. So if anybody’s curious about like where to learn more, or even if maybe do they want to find you on LinkedIn or they want to talk to you, they want to ask the chief architect some questions and get in his brain. Um, where can folks find you and where can they figure out more about what really is flow AI as updates continue to come out? 

Peter Sprygada • 38:02 

We have a you can always go to attention.com slash flowai to learn more about the product for sure. Uh I love talking with people. I absolutely love it if you can’t sell. Um, so you know, I love it when people reach out to me. And in I love to have the conversations. More importantly, I love it when when people kind of challenge me and challenge my assumptions and and really kind of because that’s how we get better. That’s how we all get better. 

Peter Sprygada • 38:24 

So you can definitely find me on LinkedIn. I’m active there. You can find me on Twitter and Mastodon. I’m at private IP. Um, you know, obviously I’m at all the big conferences. Um, so yeah, absolutely. Reach out. 

Peter Sprygada • 38:35 

Let’s have the conversation because that’s how we all get better. 

William Collins • 38:38 

Thank you so much. This has been a blast. Thank you for joining us in this magnificent studio in Austin at Autocon 4. And enjoy the rest of the conference. 

Peter Sprygada • 38:49 

Thanks for having me. This has this has really been a lot of fun. Thank you. All right.