WEBVTT 1 00:00:01.839 --> 00:00:03.839 - Join us as we gather around the hedge, 2 00:00:03.839 --> 00:00:05.219 - where we dig into technology, 3 00:00:05.519 --> 00:00:08.320 - business, and culture with the finest minds in 4 00:00:08.320 --> 00:00:09.220 - computer networking. 5 00:00:20.894 --> 00:00:23.794 - Well, good morning, Tom, and good morning, Audrey. 6 00:00:23.855 --> 00:00:24.835 - Good morning, Audrey. 7 00:00:26.094 --> 00:00:29.019 - Audrey says good morning. Yes. That's good. That's 8 00:00:29.339 --> 00:00:32.479 - yeah. That's important that Audrey has 9 00:00:32.939 --> 00:00:34.960 - her say, I suppose, it is. 10 00:00:35.260 --> 00:00:37.979 - Right. Right. Happy plant happy plant, happy life. 11 00:00:37.979 --> 00:00:38.479 - Right? 12 00:00:40.395 --> 00:00:42.395 - Something something like that. You're gonna have to 13 00:00:42.395 --> 00:00:44.335 - come up with something other because, you know, 14 00:00:44.594 --> 00:00:47.454 - the the, the rhyme isn't working, Tom. 15 00:00:47.835 --> 00:00:49.695 - But okay. It's fine. 16 00:00:50.075 --> 00:00:53.009 - It's fine. Alright. And today, we are joined 17 00:00:53.009 --> 00:00:55.090 - by Akshay, and I'm not even gonna try 18 00:00:55.090 --> 00:00:56.549 - your last names, and Shushanth 19 00:00:57.170 --> 00:00:58.070 - from Linkai. 20 00:00:58.609 --> 00:01:00.390 - And we are talking about 21 00:01:00.929 --> 00:01:03.429 - AI ops. Well, let's let's start here. Akshay, 22 00:01:03.489 --> 00:01:05.864 - are you are you in San Jose, or 23 00:01:05.864 --> 00:01:07.965 - are you West Coast, or where are you? 24 00:01:08.825 --> 00:01:10.525 - I traveled to San Jose. 25 00:01:10.984 --> 00:01:13.165 - My base location is Bangalore, India. 26 00:01:13.784 --> 00:01:14.284 - Okay. 27 00:01:14.665 --> 00:01:17.805 - Alright. Cool. And, Sushant, same thing? Are you 28 00:01:18.150 --> 00:01:19.990 - the same thing? We'll be silent once in 29 00:01:19.990 --> 00:01:21.689 - a while when we are in Bangalore. 30 00:01:22.469 --> 00:01:24.549 - Okay. Alright. So you're both in Bangalore right 31 00:01:24.549 --> 00:01:27.109 - now, which means the time differential is pretty 32 00:01:27.109 --> 00:01:27.609 - radical, 33 00:01:28.070 --> 00:01:30.314 - which means it's like nighttime there. Right? 34 00:01:30.635 --> 00:01:32.234 - Because it's What do you mean? Yeah. Yeah. 35 00:01:32.234 --> 00:01:34.174 - It's mid morning here. Yeah. 36 00:01:34.875 --> 00:01:37.855 - Alright. Cool. Okay. So let's begin with this. 37 00:01:37.995 --> 00:01:41.034 - So explain to me what AIOps is. I'm 38 00:01:41.034 --> 00:01:42.795 - sure people have in the back of their 39 00:01:42.795 --> 00:01:44.555 - head something they would think of when they 40 00:01:44.555 --> 00:01:47.159 - think of the term AI ops. 41 00:01:47.540 --> 00:01:50.100 - But what what is it really? Like, what 42 00:01:50.100 --> 00:01:52.520 - what are y'all what would be your definition? 43 00:01:52.579 --> 00:01:54.200 - I don't know who wants to take that. 44 00:01:55.460 --> 00:01:55.960 - Sure. 45 00:01:56.659 --> 00:01:58.740 - Thanks, Russ. First of all, thanks for having 46 00:01:58.740 --> 00:01:59.880 - us on your podcast. 47 00:02:00.965 --> 00:02:02.025 - It's a true privilege. 48 00:02:03.844 --> 00:02:05.305 - AIOps, the terminology 49 00:02:05.685 --> 00:02:08.264 - has been around for quite some time now. 50 00:02:08.805 --> 00:02:11.444 - Well, almost for the last half of the 51 00:02:11.444 --> 00:02:11.944 - decade. 52 00:02:12.805 --> 00:02:14.185 - And the most 53 00:02:14.900 --> 00:02:17.939 - simplest and most agreed upon definition of AIOps 54 00:02:17.939 --> 00:02:18.439 - is, 55 00:02:19.860 --> 00:02:22.599 - application of AI in IT operations. 56 00:02:25.060 --> 00:02:27.780 - Right? That's that's the most simplest and the 57 00:02:27.780 --> 00:02:28.599 - most broad 58 00:02:28.979 --> 00:02:29.479 - definition 59 00:02:30.224 --> 00:02:31.685 - of AI ops. And, 60 00:02:32.305 --> 00:02:34.885 - well, of course, every vendor and every product 61 00:02:35.344 --> 00:02:37.365 - come up with their own niche definition 62 00:02:37.825 --> 00:02:40.385 - of of that, and they derive their own 63 00:02:40.385 --> 00:02:41.764 - meaning out out of that. 64 00:02:42.305 --> 00:02:42.805 - But, 65 00:02:43.889 --> 00:02:46.049 - what is vendor what what most vendors are 66 00:02:46.049 --> 00:02:47.669 - going to do is build a product, 67 00:02:47.969 --> 00:02:50.209 - and then they are going to say, this 68 00:02:50.209 --> 00:02:51.189 - is AIOps. 69 00:02:52.530 --> 00:02:54.449 - That's that's what most vendors are going to 70 00:02:54.449 --> 00:02:56.449 - do. So I'm sorry. That's 71 00:02:57.489 --> 00:02:58.870 - That's true. That's true. 72 00:03:00.435 --> 00:03:03.155 - Well, yeah, that's the option. What me and 73 00:03:03.155 --> 00:03:04.215 - Shushant have 74 00:03:04.594 --> 00:03:07.175 - been building for last couple of years and, 75 00:03:08.915 --> 00:03:09.415 - what, 76 00:03:10.275 --> 00:03:12.215 - what is the correct terminology, 77 00:03:13.635 --> 00:03:14.855 - as per the latest, 78 00:03:16.729 --> 00:03:19.069 - state of the tech that we have built 79 00:03:19.370 --> 00:03:19.870 - is 80 00:03:20.250 --> 00:03:21.310 - agentic ops, 81 00:03:22.489 --> 00:03:23.150 - which is 82 00:03:23.449 --> 00:03:23.949 - bringing 83 00:03:24.330 --> 00:03:25.229 - AI agents 84 00:03:25.769 --> 00:03:26.830 - into operations, 85 00:03:27.530 --> 00:03:29.389 - and more specifically, the network operations. 86 00:03:31.444 --> 00:03:31.944 - Okay. 87 00:03:32.485 --> 00:03:34.025 - And that's pretty broad. 88 00:03:34.485 --> 00:03:37.784 - And in what way are you thinking about 89 00:03:38.245 --> 00:03:39.465 - this of bringing 90 00:03:40.245 --> 00:03:43.950 - AI into operations? Are we thinking like I 91 00:03:43.950 --> 00:03:45.170 - know I've seen workflows 92 00:03:45.629 --> 00:03:46.689 - where, for instance, 93 00:03:46.990 --> 00:03:49.490 - you build a digital twin of the network, 94 00:03:49.790 --> 00:03:52.110 - basically using container lab or some such like 95 00:03:52.110 --> 00:03:52.610 - that, 96 00:03:52.909 --> 00:03:53.889 - and then you 97 00:03:54.510 --> 00:03:56.849 - throw an AI at it like Claude or 98 00:03:58.324 --> 00:04:00.324 - Quinn or something, whatever it happens to be 99 00:04:00.324 --> 00:04:02.585 - chat GPT or whatever. And you say, 100 00:04:02.885 --> 00:04:05.525 - now, you know, does this look correct? Is 101 00:04:05.525 --> 00:04:06.905 - this design workable? 102 00:04:07.284 --> 00:04:09.064 - Where are the flaws? Like, 103 00:04:09.780 --> 00:04:11.379 - is that what you consider to be AI 104 00:04:11.379 --> 00:04:13.219 - ops, or are you more on the other 105 00:04:13.219 --> 00:04:14.199 - side of, 106 00:04:14.580 --> 00:04:15.080 - okay, 107 00:04:15.379 --> 00:04:17.080 - I want to troubleshoot something? 108 00:04:17.620 --> 00:04:20.180 - Like, which which side of the AI ops 109 00:04:20.180 --> 00:04:22.259 - game are you thinking of when you think 110 00:04:22.259 --> 00:04:23.264 - of AI ops? 111 00:04:25.185 --> 00:04:28.004 - We are very focused on day two operations 112 00:04:28.545 --> 00:04:29.845 - of a network interest. 113 00:04:30.465 --> 00:04:32.324 - So once the network is in place, 114 00:04:33.025 --> 00:04:35.285 - there is a gamut of, 115 00:04:36.384 --> 00:04:37.925 - job to be done under 116 00:04:38.490 --> 00:04:41.310 - network operations, right, which which could be troubleshooting 117 00:04:41.529 --> 00:04:43.610 - our CA. That is one of the jobs 118 00:04:43.610 --> 00:04:44.430 - to be done. 119 00:04:45.290 --> 00:04:45.930 - You know, 120 00:04:46.410 --> 00:04:49.149 - managing the hygiene, the security compliance, 121 00:04:50.250 --> 00:04:51.550 - managing the intent, 122 00:04:52.824 --> 00:04:55.305 - you know, managing the sanity around the intent 123 00:04:55.305 --> 00:04:57.305 - is one of the things. So all of 124 00:04:57.305 --> 00:04:58.845 - these things are, 125 00:04:59.785 --> 00:05:00.605 - coming under 126 00:05:00.985 --> 00:05:01.964 - network operations, 127 00:05:02.824 --> 00:05:04.125 - day two network operations. 128 00:05:04.939 --> 00:05:08.560 - And our role has been bringing AI 129 00:05:09.259 --> 00:05:10.959 - to assist, augment, 130 00:05:11.740 --> 00:05:12.560 - and now, 131 00:05:12.939 --> 00:05:14.620 - do a lot of heavy lifting of these 132 00:05:14.620 --> 00:05:15.120 - operations. 133 00:05:15.899 --> 00:05:17.980 - So what kind of when you talk about 134 00:05:17.980 --> 00:05:18.720 - heavy lifting, 135 00:05:19.475 --> 00:05:21.335 - are you is it that, 136 00:05:23.074 --> 00:05:24.455 - AI will build the tools, 137 00:05:24.835 --> 00:05:26.455 - or is it more of 138 00:05:26.835 --> 00:05:29.634 - a sort of agentic approach where you have 139 00:05:29.634 --> 00:05:31.475 - you have an agent doing all the things 140 00:05:31.475 --> 00:05:34.269 - that you would classically have done by writing 141 00:05:34.269 --> 00:05:36.370 - some Python or building out an automation orchestration 142 00:05:36.509 --> 00:05:38.589 - solution? How does it how does it fit 143 00:05:38.589 --> 00:05:41.229 - with existing tooling or tooling that you might 144 00:05:41.229 --> 00:05:43.250 - consider building had AI never existed? 145 00:05:44.750 --> 00:05:46.449 - Great question. That's a great question. 146 00:05:47.354 --> 00:05:49.375 - So agentic network operations 147 00:05:50.394 --> 00:05:51.615 - builds on top 148 00:05:52.555 --> 00:05:53.134 - of existing 149 00:05:53.595 --> 00:05:54.095 - instrumentation 150 00:05:55.194 --> 00:05:58.654 - that already exists in our customer's network infrastructure. 151 00:05:59.729 --> 00:06:01.349 - So your existing monitoring, 152 00:06:01.970 --> 00:06:02.470 - observability, 153 00:06:03.729 --> 00:06:04.789 - source of truth, 154 00:06:05.089 --> 00:06:05.589 - CMDB, 155 00:06:06.449 --> 00:06:07.189 - your Ansible 156 00:06:08.289 --> 00:06:09.750 - scripting, all of that 157 00:06:10.209 --> 00:06:12.389 - becomes the foundational layer 158 00:06:12.884 --> 00:06:15.464 - on top of which agentic network operational 159 00:06:16.084 --> 00:06:17.305 - stack can be built. 160 00:06:19.365 --> 00:06:20.264 - So, primarily, 161 00:06:20.884 --> 00:06:21.625 - the agentic 162 00:06:22.644 --> 00:06:23.544 - network operation 163 00:06:24.164 --> 00:06:24.664 - architecture 164 00:06:26.004 --> 00:06:26.504 - involves 165 00:06:27.419 --> 00:06:28.079 - two distinct stacks. 166 00:06:30.379 --> 00:06:32.639 - Number one is a data stack, 167 00:06:33.660 --> 00:06:36.639 - and number two is the agentic stack. 168 00:06:38.220 --> 00:06:38.720 - The 169 00:06:39.754 --> 00:06:40.654 - data stack 170 00:06:41.035 --> 00:06:41.535 - involves 171 00:06:42.795 --> 00:06:43.295 - preparing 172 00:06:44.074 --> 00:06:44.975 - the existing 173 00:06:46.555 --> 00:06:47.774 - data, telemetry, 174 00:06:48.074 --> 00:06:48.574 - configurations, 175 00:06:49.754 --> 00:06:51.615 - you know, whatnot, the intent, 176 00:06:52.269 --> 00:06:53.009 - the operational history, 177 00:06:53.949 --> 00:06:54.930 - all of this 178 00:06:56.029 --> 00:06:56.689 - to be 179 00:06:57.069 --> 00:06:58.610 - consumed by agents. 180 00:07:00.029 --> 00:07:00.529 - Now 181 00:07:01.149 --> 00:07:03.949 - all the existing tooling that we have built 182 00:07:03.949 --> 00:07:06.289 - in last three decades for network operations 183 00:07:07.595 --> 00:07:09.935 - were built primarily for human consumption. 184 00:07:10.875 --> 00:07:13.835 - Right? So the dashboards and the logs and 185 00:07:13.835 --> 00:07:15.134 - all whatnot. Right? 186 00:07:15.754 --> 00:07:16.254 - Now 187 00:07:17.835 --> 00:07:19.375 - the agents are 188 00:07:19.979 --> 00:07:22.379 - going to sit in parallel to a human 189 00:07:22.379 --> 00:07:22.879 - engineer. 190 00:07:23.259 --> 00:07:25.759 - So they would be able to do 191 00:07:26.139 --> 00:07:29.099 - complex workflows that a human engineer can do 192 00:07:29.099 --> 00:07:29.599 - today. 193 00:07:30.779 --> 00:07:32.160 - And for the agents 194 00:07:32.620 --> 00:07:34.319 - to be able to do that, 195 00:07:34.975 --> 00:07:37.535 - we have to build an abstraction layer on 196 00:07:37.535 --> 00:07:39.394 - top of the existing instrumentation 197 00:07:40.254 --> 00:07:42.995 - to convert the data, to build a context, 198 00:07:43.455 --> 00:07:45.875 - and make it available for AI to consume. 199 00:07:49.420 --> 00:07:50.959 - Right? Yeah. Yeah. 200 00:07:51.259 --> 00:07:53.120 - So the so how does the, like, 201 00:07:53.580 --> 00:07:55.740 - how does that actually happen? Do do we 202 00:07:55.819 --> 00:07:58.300 - do you say to customers, hey. You're gonna 203 00:07:58.300 --> 00:07:59.899 - need an API and everything before we can 204 00:07:59.899 --> 00:08:01.524 - get started, or do you just say, hey. 205 00:08:01.524 --> 00:08:02.884 - Give us your logins and we'll build it 206 00:08:02.884 --> 00:08:05.125 - all. Just give us your, credentials in AWS 207 00:08:05.125 --> 00:08:06.324 - and you're good to go. We'll just do 208 00:08:06.324 --> 00:08:08.324 - it all for you. Like, what's the how 209 00:08:08.324 --> 00:08:10.245 - do you get to because if we don't 210 00:08:10.245 --> 00:08:12.504 - have those interfaces, the machine interfaces, 211 00:08:13.509 --> 00:08:15.110 - what's the process to get? And I guess 212 00:08:15.110 --> 00:08:17.189 - I'm not just talking about your product, but 213 00:08:17.189 --> 00:08:19.270 - but kind of the approach in general. Like, 214 00:08:19.270 --> 00:08:20.970 - how do you how do you go from, 215 00:08:21.750 --> 00:08:23.750 - interfaces that are made for humans to full 216 00:08:23.750 --> 00:08:24.730 - agentic operations? 217 00:08:25.430 --> 00:08:26.649 - Yeah. Do you need 218 00:08:26.949 --> 00:08:28.170 - do we need to everything? 219 00:08:29.615 --> 00:08:30.754 - Is can we, 220 00:08:31.134 --> 00:08:33.934 - like, count on the AI to write the 221 00:08:33.934 --> 00:08:36.095 - converters or the thunk layers to make it 222 00:08:36.095 --> 00:08:38.674 - work so that the data becomes readable 223 00:08:39.054 --> 00:08:40.754 - in a rational way for 224 00:08:41.134 --> 00:08:41.795 - the system? 225 00:08:42.690 --> 00:08:44.370 - There are other questions around this, but go 226 00:08:44.370 --> 00:08:46.070 - ahead and see if there's yeah. 227 00:08:46.690 --> 00:08:47.190 - Sure. 228 00:08:47.809 --> 00:08:50.470 - So I'll start with the most simplest 229 00:08:51.090 --> 00:08:51.910 - form of, 230 00:08:52.610 --> 00:08:55.429 - agentic network. Network operations is 231 00:08:56.195 --> 00:08:56.695 - you 232 00:08:57.075 --> 00:08:58.214 - you can deploy, 233 00:08:58.914 --> 00:08:59.815 - open source 234 00:09:00.754 --> 00:09:01.254 - agent, 235 00:09:02.115 --> 00:09:03.414 - something like NetFlow, 236 00:09:04.115 --> 00:09:07.075 - which is a fork from OpenFlow, but designed 237 00:09:07.075 --> 00:09:08.695 - specifically for network engineering. 238 00:09:09.154 --> 00:09:10.294 - It's got the skills, 239 00:09:11.519 --> 00:09:12.740 - of a network engineer. 240 00:09:13.200 --> 00:09:15.759 - So that's your agent. So let's say we 241 00:09:15.759 --> 00:09:17.539 - put one agent in our environment, 242 00:09:18.399 --> 00:09:20.799 - and now the agent needs to talk to 243 00:09:20.799 --> 00:09:21.460 - the instrumentation 244 00:09:21.919 --> 00:09:22.419 - layer. 245 00:09:23.115 --> 00:09:25.674 - Right? The let's say NetBox is one of 246 00:09:25.674 --> 00:09:26.414 - the instrumentations 247 00:09:26.794 --> 00:09:27.695 - that we have. 248 00:09:28.075 --> 00:09:30.174 - The agent will talk to the NetBox 249 00:09:30.634 --> 00:09:32.174 - or an MCP protocol. 250 00:09:33.355 --> 00:09:35.375 - Right? So as long as the instrumentation 251 00:09:35.835 --> 00:09:36.335 - layer, 252 00:09:36.794 --> 00:09:37.294 - instrumentation 253 00:09:37.754 --> 00:09:38.254 - toolkit 254 00:09:38.730 --> 00:09:39.870 - supports MCP. 255 00:09:40.330 --> 00:09:42.490 - That means an agent can talk to it 256 00:09:42.490 --> 00:09:42.990 - now. 257 00:09:44.330 --> 00:09:44.830 - MCP, 258 00:09:45.529 --> 00:09:47.470 - the most simplest definition 259 00:09:48.009 --> 00:09:49.470 - that we can have is 260 00:09:50.330 --> 00:09:52.110 - MCP is the API 261 00:09:52.649 --> 00:09:53.709 - for an agent 262 00:09:54.065 --> 00:09:56.404 - or a model to talk to a traditional 263 00:09:56.544 --> 00:09:57.044 - software. 264 00:09:57.904 --> 00:10:00.544 - Just like an API was the channel for 265 00:10:00.544 --> 00:10:02.004 - software to software communication. 266 00:10:03.184 --> 00:10:03.684 - Right? 267 00:10:05.985 --> 00:10:08.245 - One agent one NetLaw agent 268 00:10:08.940 --> 00:10:12.480 - talking to your existing tool set or MCP 269 00:10:13.259 --> 00:10:15.679 - is the most simplest form 270 00:10:16.139 --> 00:10:16.720 - of agentic 271 00:10:17.419 --> 00:10:18.480 - network operations 272 00:10:19.340 --> 00:10:19.840 - stack. 273 00:10:21.195 --> 00:10:23.295 - However, we can get started there. However, 274 00:10:24.154 --> 00:10:25.134 - this setup 275 00:10:25.835 --> 00:10:26.335 - allows 276 00:10:26.875 --> 00:10:28.975 - allows the user to do, 277 00:10:30.394 --> 00:10:33.514 - you know, allows the user to automate and, 278 00:10:34.075 --> 00:10:35.295 - do certain tasks, 279 00:10:36.600 --> 00:10:38.840 - very, very easily with, with the assistant of 280 00:10:38.840 --> 00:10:40.620 - that agent. For example, 281 00:10:41.399 --> 00:10:44.360 - we have a NetLaw agent deployed in our, 282 00:10:44.360 --> 00:10:47.080 - in our own office network. It has access 283 00:10:47.080 --> 00:10:49.625 - to our office Wi Fi, office firewall, 284 00:10:50.164 --> 00:10:51.865 - the ServiceNow and all of that. 285 00:10:52.404 --> 00:10:53.705 - And it's available 286 00:10:54.245 --> 00:10:56.264 - to me in my team's chat. 287 00:10:57.445 --> 00:11:00.004 - Okay. So I can just talk to the 288 00:11:00.004 --> 00:11:00.504 - agents 289 00:11:00.850 --> 00:11:03.029 - like I would talk to my IT guy 290 00:11:03.169 --> 00:11:03.909 - and say, 291 00:11:04.210 --> 00:11:05.990 - what is happening with 292 00:11:06.450 --> 00:11:08.549 - the Wi Fi in 2nd Floor? 293 00:11:09.649 --> 00:11:12.129 - Right? And the the brain of the agent 294 00:11:12.129 --> 00:11:13.029 - is the model. 295 00:11:14.014 --> 00:11:15.315 - It could be any model. 296 00:11:15.774 --> 00:11:17.934 - If we hook the agent to a frontier 297 00:11:17.934 --> 00:11:19.554 - model like a plot or OpenAI, 298 00:11:20.014 --> 00:11:22.495 - obviously, it's got a lot of training and 299 00:11:22.495 --> 00:11:23.134 - a lot of, 300 00:11:23.774 --> 00:11:24.754 - knowledge base. 301 00:11:25.215 --> 00:11:26.754 - It understands my question. 302 00:11:27.110 --> 00:11:29.350 - It can go to the MCP call. It 303 00:11:29.350 --> 00:11:32.170 - can fetch the Wi Fi statistics. It can 304 00:11:32.310 --> 00:11:34.710 - analyze it and then summarize it saying that 305 00:11:34.710 --> 00:11:36.009 - this is what is happening. 306 00:11:36.710 --> 00:11:37.210 - Now 307 00:11:37.590 --> 00:11:38.570 - in this case, 308 00:11:38.870 --> 00:11:41.509 - the agent is acting like an assistant for 309 00:11:41.509 --> 00:11:42.009 - me. 310 00:11:42.345 --> 00:11:42.584 - Do, 311 00:11:43.225 --> 00:11:44.764 - wherein I'm able to delegate, 312 00:11:46.024 --> 00:11:46.684 - a very 313 00:11:47.225 --> 00:11:49.625 - defined task for it, which it can go 314 00:11:49.625 --> 00:11:51.784 - and fetch some data, analyze and, 315 00:11:52.424 --> 00:11:53.964 - you know, summarize for me. 316 00:11:55.049 --> 00:11:56.269 - For more complex 317 00:11:57.049 --> 00:11:58.509 - workflows, for example, 318 00:12:00.809 --> 00:12:02.269 - we want an agent 319 00:12:03.450 --> 00:12:04.269 - to be proactive, 320 00:12:05.129 --> 00:12:06.110 - or we want 321 00:12:06.730 --> 00:12:07.870 - a team of agents 322 00:12:08.294 --> 00:12:09.595 - running in our environment 323 00:12:10.054 --> 00:12:11.115 - which are proactively 324 00:12:11.735 --> 00:12:13.434 - picking up a ServiceNow ticket 325 00:12:14.134 --> 00:12:17.254 - and understanding what the ticket means, then going 326 00:12:17.254 --> 00:12:18.475 - into different systems, 327 00:12:19.414 --> 00:12:20.214 - doing the, 328 00:12:20.695 --> 00:12:23.355 - diagnosis, coming back, stitching it all together 329 00:12:23.799 --> 00:12:27.100 - would require a team of agents to operate. 330 00:12:28.679 --> 00:12:31.259 - And for this team of agents to operate, 331 00:12:31.960 --> 00:12:35.080 - it would need a proper data stack to 332 00:12:35.080 --> 00:12:35.820 - be built 333 00:12:36.519 --> 00:12:38.945 - before these agents can touch the instrumentation. 334 00:12:41.965 --> 00:12:44.365 - And that's precisely what we are building at 335 00:12:44.365 --> 00:12:44.865 - LinkI 336 00:12:45.485 --> 00:12:47.504 - is we are building an, 337 00:12:49.004 --> 00:12:49.504 - universal 338 00:12:50.125 --> 00:12:52.144 - abstraction layer and a universal 339 00:12:52.925 --> 00:12:53.425 - middleware 340 00:12:54.259 --> 00:12:55.720 - on top of the existing 341 00:12:56.100 --> 00:12:57.240 - network infrastructure 342 00:12:58.019 --> 00:12:58.759 - and instrumentation 343 00:12:59.139 --> 00:12:59.639 - layer 344 00:13:01.220 --> 00:13:03.139 - on top of which the agents can run 345 00:13:03.139 --> 00:13:03.799 - and operate. 346 00:13:06.514 --> 00:13:09.154 - So then so then, presumably, that means that 347 00:13:09.154 --> 00:13:10.295 - you are taking responsibility 348 00:13:10.674 --> 00:13:13.475 - to talk to anything people would deploy. Like, 349 00:13:13.475 --> 00:13:15.315 - if it's net but if it's NetBox, if 350 00:13:15.315 --> 00:13:17.875 - it's ServiceNow, if it's whatever it is. So 351 00:13:17.875 --> 00:13:19.554 - so you're saying we'll build all the south 352 00:13:19.795 --> 00:13:20.855 - the southbound interfaces, 353 00:13:21.379 --> 00:13:23.379 - and then here's here's your abstraction. And since 354 00:13:23.379 --> 00:13:24.440 - we've done all the work, 355 00:13:24.899 --> 00:13:27.539 - now now agents can consume your tool set. 356 00:13:27.539 --> 00:13:28.039 - Right? 357 00:13:29.459 --> 00:13:31.940 - Yeah. That's that's very accurate. So we have 358 00:13:31.940 --> 00:13:34.279 - the connectors for all the observability, 359 00:13:35.059 --> 00:13:36.919 - platform major observability platforms, 360 00:13:37.514 --> 00:13:39.295 - ITSMs, source of truth, 361 00:13:40.075 --> 00:13:42.495 - and whatnot. And we become the middleware 362 00:13:42.795 --> 00:13:45.915 - for the customer to connect to his existing 363 00:13:45.915 --> 00:13:46.415 - infrastructure. 364 00:13:46.955 --> 00:13:49.195 - And on top of that, he can build 365 00:13:49.195 --> 00:13:50.095 - his own agents, 366 00:13:50.970 --> 00:13:53.149 - buy agents from vendor or marketplace, 367 00:13:54.330 --> 00:13:57.129 - or deploy his own agents and, you know, 368 00:13:57.129 --> 00:13:59.230 - basically get into agent network operations. 369 00:14:01.769 --> 00:14:02.269 - Okay. 370 00:14:03.209 --> 00:14:03.709 - So 371 00:14:05.245 --> 00:14:07.884 - it sounds like we need a middleware layer 372 00:14:07.884 --> 00:14:08.785 - like you're saying 373 00:14:09.165 --> 00:14:10.304 - to go between 374 00:14:12.205 --> 00:14:15.024 - systems that don't have an MCP interface 375 00:14:15.725 --> 00:14:19.105 - and having the MCP interface for Agintiq AI 376 00:14:19.799 --> 00:14:21.179 - on the operation side. 377 00:14:21.639 --> 00:14:24.059 - That seems like there's a there's a disconnect 378 00:14:24.200 --> 00:14:24.700 - there 379 00:14:25.080 --> 00:14:27.480 - in the operation side of the market is 380 00:14:27.480 --> 00:14:30.440 - what you're saying as a first step towards 381 00:14:30.440 --> 00:14:31.340 - getting to 382 00:14:31.879 --> 00:14:34.434 - AI ops. Like, there's gotta be some way 383 00:14:34.434 --> 00:14:35.495 - beyond just 384 00:14:36.274 --> 00:14:38.134 - saying, okay. Read these logs, 385 00:14:38.514 --> 00:14:41.154 - right, and figure this out or whatever. There's 386 00:14:41.154 --> 00:14:42.375 - gotta be a structure. 387 00:14:42.834 --> 00:14:44.914 - And, normally, that structure for us would look 388 00:14:44.914 --> 00:14:45.735 - like yang, 389 00:14:46.194 --> 00:14:46.694 - SNMP, 390 00:14:47.240 --> 00:14:48.699 - god forbid, mibs, 391 00:14:50.360 --> 00:14:52.860 - but please let's not do mibs. But okay. 392 00:14:53.000 --> 00:14:53.500 - Whatever. 393 00:14:55.159 --> 00:14:58.440 - Or structured text or even unstructured text in 394 00:14:58.440 --> 00:14:59.174 - many cases 395 00:14:59.654 --> 00:15:01.975 - or log like log files. Right? Structured text 396 00:15:01.975 --> 00:15:03.595 - would be something like a log file. 397 00:15:04.534 --> 00:15:06.214 - And what you're trying to say is, okay. 398 00:15:06.214 --> 00:15:08.455 - I have this stuff, but it but it 399 00:15:08.455 --> 00:15:09.274 - has to be 400 00:15:09.575 --> 00:15:11.975 - unified in some way to get me to 401 00:15:11.975 --> 00:15:14.370 - the point where I can attach an agentic 402 00:15:14.429 --> 00:15:17.149 - AI system to it, whatever that agentic AI 403 00:15:17.149 --> 00:15:19.649 - system might happen to be to do something 404 00:15:19.789 --> 00:15:20.929 - with this data. 405 00:15:21.470 --> 00:15:23.549 - Now another thing that's interesting to me in 406 00:15:23.549 --> 00:15:24.289 - that area 407 00:15:24.669 --> 00:15:27.009 - is how are you cleaning the data? 408 00:15:27.629 --> 00:15:28.850 - I we assume 409 00:15:29.574 --> 00:15:30.394 - all the time, 410 00:15:31.014 --> 00:15:32.534 - and this is not going back to the 411 00:15:32.534 --> 00:15:35.195 - AI hallucination problem. That's an entirely different 412 00:15:35.574 --> 00:15:37.034 - problem and set of situations. 413 00:15:37.975 --> 00:15:38.475 - But 414 00:15:38.934 --> 00:15:39.674 - we assume 415 00:15:40.214 --> 00:15:42.634 - that when I pull data off the network, 416 00:15:42.774 --> 00:15:44.875 - that it's clean and it's good. 417 00:15:45.949 --> 00:15:49.069 - But the fact is, it's not for many, 418 00:15:49.069 --> 00:15:49.809 - many reasons. 419 00:15:50.269 --> 00:15:52.190 - There are many places where I will drop 420 00:15:52.190 --> 00:15:52.690 - data. 421 00:15:53.230 --> 00:15:53.970 - For instance, 422 00:15:54.350 --> 00:15:55.250 - I am doing 423 00:15:55.789 --> 00:15:57.409 - net flow, and I have 424 00:15:58.134 --> 00:16:00.475 - sampling rate of one packet in 10,000, 425 00:16:00.695 --> 00:16:03.095 - and yet somehow or another log system gets 426 00:16:03.095 --> 00:16:05.254 - clogged up and I drop one of the 427 00:16:05.254 --> 00:16:08.375 - samples, or I drop a I drop a 428 00:16:08.375 --> 00:16:11.450 - log a log that says the Erica or 429 00:16:11.450 --> 00:16:12.190 - the BGP 430 00:16:12.570 --> 00:16:13.070 - notification 431 00:16:13.450 --> 00:16:16.570 - code on a BGP neighbor down. Like, is 432 00:16:16.570 --> 00:16:18.110 - there anything in this 433 00:16:18.649 --> 00:16:20.589 - is there any concern about that? 434 00:16:21.129 --> 00:16:23.549 - Or or is that, like, something that seems 435 00:16:23.610 --> 00:16:24.110 - unsolvable? 436 00:16:24.725 --> 00:16:26.404 - Or are you trying to infer that data? 437 00:16:26.404 --> 00:16:28.485 - Like, what what's going on with that side 438 00:16:28.485 --> 00:16:29.304 - of the situation? 439 00:16:31.285 --> 00:16:32.504 - Great question, Russ. 440 00:16:34.325 --> 00:16:35.065 - That can 441 00:16:35.684 --> 00:16:37.644 - you know, that that question has a swing 442 00:16:37.644 --> 00:16:40.110 - in in fact to double click and explain 443 00:16:40.110 --> 00:16:42.049 - how we prepare the data 444 00:16:42.590 --> 00:16:44.929 - before it it's ready for agentic consumption. 445 00:16:45.870 --> 00:16:46.370 - Right? 446 00:16:46.750 --> 00:16:48.610 - Now we have three 447 00:16:49.149 --> 00:16:51.330 - categories or three main buckets, 448 00:16:52.750 --> 00:16:54.450 - of source of information. 449 00:16:55.745 --> 00:16:57.445 - We always start with 450 00:16:58.304 --> 00:17:00.565 - capturing the intent of the network. 451 00:17:01.745 --> 00:17:03.904 - Now how do we capture intent of the 452 00:17:03.904 --> 00:17:04.404 - network? 453 00:17:05.265 --> 00:17:08.644 - Our own novel way of doing that is 454 00:17:09.105 --> 00:17:11.045 - we allow the architect 455 00:17:11.640 --> 00:17:13.019 - to talk to the system 456 00:17:13.559 --> 00:17:14.859 - in a natural language 457 00:17:15.799 --> 00:17:16.299 - and 458 00:17:17.000 --> 00:17:19.660 - convey the architecture in a natural language. 459 00:17:20.839 --> 00:17:22.920 - For example, if I'm the architect, I could 460 00:17:22.920 --> 00:17:23.420 - say 461 00:17:24.025 --> 00:17:25.804 - my infrastructure spans over 462 00:17:26.105 --> 00:17:27.244 - 200 sites. 463 00:17:29.304 --> 00:17:30.825 - All of 180 464 00:17:30.825 --> 00:17:32.765 - of them are retail stores 465 00:17:33.144 --> 00:17:36.345 - running SD WAN with primarily being MPLS, secondly 466 00:17:36.345 --> 00:17:37.565 - being IPsec VPN. 467 00:17:38.240 --> 00:17:39.779 - Hub and spoke topology. 468 00:17:40.480 --> 00:17:41.940 - 20 of them are warehouses. 469 00:17:43.279 --> 00:17:46.019 - They're running on, Cisco as a core 470 00:17:46.320 --> 00:17:46.720 - and, 471 00:17:48.320 --> 00:17:49.779 - Aruba as the Wi Fi 472 00:17:50.224 --> 00:17:52.705 - and whatnot. Right? So I can as an 473 00:17:52.705 --> 00:17:54.404 - architect, I can very 474 00:17:54.705 --> 00:17:55.205 - clearly 475 00:17:55.825 --> 00:17:58.884 - articulate my network architecture in plain 476 00:17:59.424 --> 00:18:00.805 - language, natural language, 477 00:18:01.424 --> 00:18:02.805 - and that the AI 478 00:18:03.430 --> 00:18:04.250 - can capture 479 00:18:04.710 --> 00:18:06.170 - and build the first 480 00:18:06.549 --> 00:18:08.950 - set of graph, which is which becomes the 481 00:18:08.950 --> 00:18:10.650 - base or which becomes the foundation, 482 00:18:11.509 --> 00:18:13.830 - of the next set of data that we 483 00:18:13.830 --> 00:18:14.809 - are going to build. 484 00:18:15.744 --> 00:18:18.005 - We can also capture intent 485 00:18:18.384 --> 00:18:18.884 - by 486 00:18:19.744 --> 00:18:22.144 - having an upload of the HLDs and the 487 00:18:22.144 --> 00:18:24.304 - LLDs and the SOPs and the runbooks, which 488 00:18:24.304 --> 00:18:25.924 - are already built for the network. 489 00:18:26.464 --> 00:18:28.644 - Right? The the foundation 490 00:18:29.105 --> 00:18:29.765 - of the 491 00:18:30.349 --> 00:18:31.650 - data is the 492 00:18:32.430 --> 00:18:34.769 - knowledge graph, which is built on intent. 493 00:18:36.590 --> 00:18:37.650 - The next stage 494 00:18:38.190 --> 00:18:38.690 - is 495 00:18:39.309 --> 00:18:42.049 - data collection or collection of the live network 496 00:18:42.109 --> 00:18:42.609 - state. 497 00:18:43.765 --> 00:18:44.265 - To 498 00:18:44.804 --> 00:18:47.444 - and this happens through plugins of all the 499 00:18:47.444 --> 00:18:49.224 - auxiliary stack and the existing 500 00:18:49.684 --> 00:18:50.184 - orchestrators, 501 00:18:51.365 --> 00:18:52.744 - that the customer has. 502 00:18:54.005 --> 00:18:54.505 - Here, 503 00:18:55.204 --> 00:18:56.744 - because we have captured 504 00:18:57.125 --> 00:18:59.384 - the intent of the network first, 505 00:19:00.380 --> 00:19:02.559 - the AI can recommend 506 00:19:02.940 --> 00:19:04.399 - the gaps in the visibility. 507 00:19:05.740 --> 00:19:07.599 - And this is one of the, 508 00:19:09.099 --> 00:19:10.319 - novel applications 509 00:19:10.700 --> 00:19:11.440 - of AI 510 00:19:11.740 --> 00:19:12.240 - is 511 00:19:12.940 --> 00:19:13.440 - now 512 00:19:14.275 --> 00:19:17.315 - once we have an architecture laid out, we 513 00:19:17.315 --> 00:19:17.815 - can 514 00:19:18.515 --> 00:19:19.414 - have the system 515 00:19:19.795 --> 00:19:22.194 - tell us the gaps in the visibility or 516 00:19:22.194 --> 00:19:24.275 - gaps in the monitoring or the gaps in 517 00:19:24.275 --> 00:19:24.855 - the logging. 518 00:19:25.394 --> 00:19:28.035 - The AI can tell us and given the 519 00:19:28.035 --> 00:19:28.535 - autonomy, 520 00:19:29.389 --> 00:19:31.809 - the agent can go into the device 521 00:19:32.190 --> 00:19:34.349 - and set that up, set up the net 522 00:19:34.349 --> 00:19:35.649 - flow, set up the syslog, 523 00:19:36.349 --> 00:19:39.069 - or given the autonomy, the agent can log 524 00:19:39.069 --> 00:19:41.730 - into the Cisco Meraki or the Cisco Dnack 525 00:19:42.190 --> 00:19:45.474 - or whatnot. Right? And set up the required 526 00:19:45.534 --> 00:19:45.934 - flows. 527 00:19:47.615 --> 00:19:49.855 - This is the second stage wherein we collect 528 00:19:49.855 --> 00:19:51.394 - the live network state. 529 00:19:52.414 --> 00:19:53.554 - The third stage 530 00:19:54.174 --> 00:19:55.554 - is we collect 531 00:19:55.934 --> 00:19:57.154 - operational history 532 00:19:57.759 --> 00:19:58.500 - through ServiceNow 533 00:19:59.200 --> 00:19:59.679 - or, 534 00:20:00.160 --> 00:20:02.339 - the Cheetah project manager or whatnot. 535 00:20:02.799 --> 00:20:04.019 - Right? The entire 536 00:20:04.400 --> 00:20:07.460 - history of what has happened on the network, 537 00:20:08.960 --> 00:20:09.859 - which is captured 538 00:20:10.555 --> 00:20:12.335 - through change management requests 539 00:20:12.715 --> 00:20:13.615 - and problem, 540 00:20:13.914 --> 00:20:15.535 - resolution tickets and whatnot. 541 00:20:16.555 --> 00:20:19.455 - This becomes the third layer of the graph. 542 00:20:19.835 --> 00:20:21.375 - This is operational history. 543 00:20:22.410 --> 00:20:23.070 - All of 544 00:20:23.450 --> 00:20:25.150 - these are stored 545 00:20:25.609 --> 00:20:26.750 - as contextual 546 00:20:27.529 --> 00:20:28.670 - knowledge graphs 547 00:20:30.250 --> 00:20:31.150 - in our system. 548 00:20:32.410 --> 00:20:33.789 - We do not store 549 00:20:34.424 --> 00:20:34.924 - the 550 00:20:35.384 --> 00:20:37.404 - the the full breadth 551 00:20:38.105 --> 00:20:40.825 - of monitoring data in our system. That still 552 00:20:40.825 --> 00:20:42.684 - resides with the monitoring tool. 553 00:20:43.144 --> 00:20:44.924 - What we are building essentially 554 00:20:45.705 --> 00:20:46.924 - is a traversable 555 00:20:47.910 --> 00:20:50.330 - knowledge graph for the agent. 556 00:20:51.509 --> 00:20:52.950 - So I wanted to if if you don't 557 00:20:52.950 --> 00:20:54.869 - mind, I kinda wanna, I thought of something 558 00:20:54.869 --> 00:20:56.549 - when you're talking about intent I wanted to 559 00:20:56.549 --> 00:20:58.330 - get get your thoughts on. So, 560 00:20:58.789 --> 00:21:00.710 - you mentioned a bunch of things like Aruba 561 00:21:00.710 --> 00:21:02.309 - has the wireless over here, and we have 562 00:21:02.309 --> 00:21:03.644 - an SD WAN over there. And, 563 00:21:04.125 --> 00:21:05.184 - when I think about 564 00:21:05.724 --> 00:21:08.045 - when I think about intent, I don't think 565 00:21:08.045 --> 00:21:09.105 - about what is. 566 00:21:09.484 --> 00:21:11.244 - I think about what's supposed to be. Like, 567 00:21:11.244 --> 00:21:13.244 - what is what is the thing that like, 568 00:21:13.244 --> 00:21:14.605 - you can go out and and and all 569 00:21:14.605 --> 00:21:16.205 - that most of the stuff you said is, 570 00:21:16.205 --> 00:21:18.224 - like, here's what's on the network. Okay. 571 00:21:18.539 --> 00:21:20.700 - But that's and and I and I understand. 572 00:21:20.700 --> 00:21:22.220 - Like, you do have to do that. Obviously, 573 00:21:22.220 --> 00:21:23.820 - you have to discover the environment and see 574 00:21:23.820 --> 00:21:26.460 - what it is. But but to me, if 575 00:21:26.460 --> 00:21:29.019 - we start if that's our starting point, just 576 00:21:29.019 --> 00:21:29.840 - just philosophically, 577 00:21:30.394 --> 00:21:32.634 - all we're ever going to do is make 578 00:21:32.634 --> 00:21:35.194 - tweaks to what what ended up getting deployed, 579 00:21:35.194 --> 00:21:36.894 - which we all know is very imperfect. 580 00:21:37.194 --> 00:21:37.694 - Like, 581 00:21:38.394 --> 00:21:40.794 - the real intent is how is this supposed 582 00:21:40.794 --> 00:21:43.355 - to behave? What's what's supposed to happen? Not 583 00:21:43.355 --> 00:21:45.579 - what is, not what what wireless controllers we 584 00:21:45.579 --> 00:21:47.579 - have as in control, you know, and which 585 00:21:47.579 --> 00:21:49.740 - vendors are running the SDWAN. Like, to me, 586 00:21:49.740 --> 00:21:51.279 - intent should be a higher level, 587 00:21:51.900 --> 00:21:53.660 - thought than that. And I realized, like, you 588 00:21:53.660 --> 00:21:55.099 - gotta take high level things and put them 589 00:21:55.099 --> 00:21:57.164 - into concrete. That's what software does. Right? But 590 00:21:57.325 --> 00:21:59.005 - but I guess I don't know. Just kind 591 00:21:59.005 --> 00:22:00.765 - of at a higher level, what are your 592 00:22:00.765 --> 00:22:02.605 - thoughts about that? I I think I'll give 593 00:22:02.605 --> 00:22:03.984 - you mine, and I wanna hear yours. 594 00:22:04.684 --> 00:22:06.225 - AI should give us the possibility 595 00:22:06.765 --> 00:22:07.424 - to actually 596 00:22:08.125 --> 00:22:10.445 - make our intent actually happen on the network, 597 00:22:10.445 --> 00:22:12.684 - which really has been very imperfect to this 598 00:22:12.684 --> 00:22:14.839 - point. We've had limited tools. We had limited 599 00:22:14.839 --> 00:22:15.579 - opt automation. 600 00:22:16.279 --> 00:22:18.119 - But if we just say, here's the existing 601 00:22:18.119 --> 00:22:20.039 - environment, go tweak that and make it good, 602 00:22:20.039 --> 00:22:21.720 - we're not we're gonna miss the boat. We're 603 00:22:21.720 --> 00:22:23.240 - not gonna get the improvements we thought we 604 00:22:23.240 --> 00:22:24.519 - were gonna get is my view. I don't 605 00:22:24.519 --> 00:22:25.579 - know. What do you think? 606 00:22:27.075 --> 00:22:28.194 - You're 100 607 00:22:28.194 --> 00:22:29.015 - right, Tom. 608 00:22:29.554 --> 00:22:30.054 - And 609 00:22:30.355 --> 00:22:30.855 - intent 610 00:22:31.154 --> 00:22:31.654 - drift 611 00:22:32.194 --> 00:22:34.835 - happens to be one of the most common 612 00:22:34.835 --> 00:22:37.095 - use cases that we get from customers, 613 00:22:37.954 --> 00:22:40.679 - especially the large enterprise custom, the super large 614 00:22:40.679 --> 00:22:43.880 - enterprise customers where the network spans, you know, 615 00:22:43.880 --> 00:22:45.899 - a few tens and thousands of devices. 616 00:22:47.240 --> 00:22:48.220 - The intent 617 00:22:48.599 --> 00:22:52.119 - can be literally anything that the architect defines. 618 00:22:52.119 --> 00:22:53.019 - It could be, 619 00:22:54.684 --> 00:22:55.825 - it could be that, 620 00:22:57.005 --> 00:22:59.265 - you know, it it could be 621 00:22:59.964 --> 00:23:03.424 - not necessarily what is, but what is his 622 00:23:03.484 --> 00:23:04.705 - intent with the network? 623 00:23:05.404 --> 00:23:07.884 - He can describe that. That gets formed as 624 00:23:07.884 --> 00:23:08.464 - the intent. 625 00:23:08.789 --> 00:23:10.809 - And then when you get to the agent, 626 00:23:11.110 --> 00:23:13.670 - which has been designed to detect the drift 627 00:23:13.670 --> 00:23:14.970 - from the intent, 628 00:23:15.509 --> 00:23:17.910 - it can go and crawl the network, pick 629 00:23:17.910 --> 00:23:20.250 - up all the data, and then bring 630 00:23:20.630 --> 00:23:22.250 - a bring a report of the 631 00:23:22.625 --> 00:23:23.125 - the 632 00:23:23.505 --> 00:23:24.484 - desired intent 633 00:23:25.105 --> 00:23:27.684 - versus what is there on the network today. 634 00:23:28.785 --> 00:23:30.465 - And this happens to be one of the 635 00:23:30.465 --> 00:23:33.045 - most loud and most used use cases 636 00:23:33.424 --> 00:23:34.565 - for us today. 637 00:23:35.279 --> 00:23:36.799 - So I would I would actually I wanna 638 00:23:36.799 --> 00:23:38.820 - glom onto that for a second time because 639 00:23:38.960 --> 00:23:40.720 - I think part of the problem that I've 640 00:23:40.720 --> 00:23:42.480 - always seen with intent driven systems 641 00:23:42.960 --> 00:23:45.359 - there are others. There's many problems, but one 642 00:23:45.359 --> 00:23:48.400 - of them is is that in order for 643 00:23:48.400 --> 00:23:48.900 - me 644 00:23:49.200 --> 00:23:49.700 - to 645 00:23:50.424 --> 00:23:52.505 - have an intent driven system, I have to 646 00:23:52.505 --> 00:23:54.045 - have basically a language 647 00:23:54.424 --> 00:23:56.765 - in which I express my intent. 648 00:23:57.625 --> 00:23:59.484 - Right? Because because 649 00:23:59.785 --> 00:24:02.205 - and what the intent based system often does 650 00:24:02.265 --> 00:24:03.325 - is just say, 651 00:24:04.109 --> 00:24:06.130 - operator said create VLAN. 652 00:24:06.589 --> 00:24:08.609 - I know that to create VLAN, 653 00:24:09.230 --> 00:24:11.710 - I create these commands on this type of 654 00:24:11.710 --> 00:24:14.049 - box and that on that type of box. 655 00:24:14.589 --> 00:24:16.849 - To me, where agentic should really 656 00:24:17.150 --> 00:24:18.289 - start to be 657 00:24:18.755 --> 00:24:19.255 - interesting 658 00:24:19.875 --> 00:24:20.375 - is 659 00:24:20.995 --> 00:24:21.495 - saying, 660 00:24:22.115 --> 00:24:24.134 - I want to separate traffic, 661 00:24:24.515 --> 00:24:27.255 - my HTTP traffic from my voice traffic. 662 00:24:27.795 --> 00:24:28.535 - And now 663 00:24:30.115 --> 00:24:32.035 - the system should be able to say, if 664 00:24:32.035 --> 00:24:34.410 - it's truly agentic, should be able to say, 665 00:24:34.890 --> 00:24:37.609 - well, how do I create traffic separation? Well, 666 00:24:37.609 --> 00:24:39.930 - there are many ways. I can do an 667 00:24:39.930 --> 00:24:42.029 - access list. I can use 668 00:24:42.330 --> 00:24:43.150 - IP tables. 669 00:24:43.529 --> 00:24:44.269 - I can 670 00:24:44.650 --> 00:24:47.049 - find different paths, and I can use traffic 671 00:24:47.049 --> 00:24:50.269 - engineering or traffic steering to make that happen, 672 00:24:50.464 --> 00:24:52.244 - or I can create a VLAN. 673 00:24:53.585 --> 00:24:55.684 - And, right, at that point, 674 00:24:56.224 --> 00:24:58.384 - then the agentic system starts to become a 675 00:24:58.384 --> 00:25:00.704 - little more interesting. Now it also becomes a 676 00:25:00.704 --> 00:25:03.825 - little more problematic. Right? Because now it's chosen 677 00:25:03.825 --> 00:25:04.404 - a solution 678 00:25:05.660 --> 00:25:07.359 - that may or may not fit 679 00:25:07.660 --> 00:25:10.700 - with your overall architecture goals for the next 680 00:25:10.700 --> 00:25:13.579 - ten years because it doesn't know. Right? It 681 00:25:13.579 --> 00:25:15.200 - doesn't know that in general, 682 00:25:15.500 --> 00:25:18.859 - you don't want 5,868,000 683 00:25:18.859 --> 00:25:20.319 - VLANs on your network. 684 00:25:21.034 --> 00:25:23.194 - Right? And so it doesn't have that kind 685 00:25:23.194 --> 00:25:25.355 - of knowledge, and that's where it has to 686 00:25:25.355 --> 00:25:26.095 - be interactive 687 00:25:26.634 --> 00:25:28.414 - and where you have to have very 688 00:25:28.714 --> 00:25:29.214 - clear 689 00:25:29.835 --> 00:25:31.134 - bounds and goals 690 00:25:31.515 --> 00:25:33.115 - to make it work. So I don't know. 691 00:25:33.115 --> 00:25:34.315 - What do you think, Tom? Is that that's 692 00:25:34.315 --> 00:25:35.134 - kind of my 693 00:25:35.539 --> 00:25:37.380 - my sense of where I think it would 694 00:25:37.380 --> 00:25:38.039 - be useful. 695 00:25:38.659 --> 00:25:40.200 - Yeah. Yeah. Well, I I think, 696 00:25:40.659 --> 00:25:42.500 - because the you could get a ton of 697 00:25:42.500 --> 00:25:44.659 - value from saying it it goes out and 698 00:25:44.659 --> 00:25:47.299 - discovers everything, figures out what's out there. And 699 00:25:47.299 --> 00:25:48.740 - you say, I I would like it to 700 00:25:48.740 --> 00:25:50.505 - be this. What's the difference? Like, what what 701 00:25:50.505 --> 00:25:51.945 - do I have to do to get to 702 00:25:51.945 --> 00:25:53.404 - my to my goal? 703 00:25:54.345 --> 00:25:56.105 - And you could it you know, I could 704 00:25:56.105 --> 00:25:57.785 - see it being very useful to say, well, 705 00:25:57.785 --> 00:25:58.904 - you could do that, but you have to 706 00:25:58.904 --> 00:26:00.045 - do x, y, and z. 707 00:26:00.664 --> 00:26:03.305 - And then if your if if your stated 708 00:26:03.305 --> 00:26:05.404 - intent is far enough from the reality, 709 00:26:06.029 --> 00:26:07.090 - then, you know, 710 00:26:07.549 --> 00:26:09.230 - you this is the sort of thing where, 711 00:26:09.230 --> 00:26:10.910 - you know, we would sit and build slide 712 00:26:10.910 --> 00:26:12.990 - decks and spreadsheets and budgets, and we would 713 00:26:12.990 --> 00:26:14.509 - take we would take months to come up 714 00:26:14.509 --> 00:26:15.950 - with these, and we could only ever do 715 00:26:15.950 --> 00:26:17.309 - one or two scenarios, at least in my 716 00:26:17.309 --> 00:26:19.585 - experience. Like, here's what we have. If we 717 00:26:19.664 --> 00:26:21.825 - if we're trying to accomplish this this business 718 00:26:21.825 --> 00:26:24.065 - objective, here's where we gotta go. And you 719 00:26:24.065 --> 00:26:25.204 - could you could never 720 00:26:26.065 --> 00:26:27.825 - look at all the possibilities. It had to 721 00:26:27.825 --> 00:26:29.265 - only be a couple because it was too 722 00:26:29.265 --> 00:26:31.744 - expensive, right, to go to go see how 723 00:26:31.744 --> 00:26:33.929 - do we change the infrastructure to be to 724 00:26:33.929 --> 00:26:35.549 - separate traffic. Right? Whatever. 725 00:26:36.169 --> 00:26:38.490 - That to me is the real possibility that, 726 00:26:38.889 --> 00:26:39.869 - AI gives us. 727 00:26:40.569 --> 00:26:42.750 - Yeah. That's that's what I would hope 728 00:26:43.129 --> 00:26:43.629 - myself. 729 00:26:44.649 --> 00:26:46.569 - I think we're far from that right now, 730 00:26:46.569 --> 00:26:48.329 - but I think that's that that would be 731 00:26:48.329 --> 00:26:50.174 - what I would hope for in the long 732 00:26:50.174 --> 00:26:50.674 - run. 733 00:26:51.375 --> 00:26:52.595 - Absolutely. I think 734 00:26:53.054 --> 00:26:54.734 - we may we may be four to five 735 00:26:54.734 --> 00:26:56.034 - years away from that. 736 00:26:57.454 --> 00:26:58.355 - Given the 737 00:26:58.734 --> 00:27:00.835 - given the state of the tech today, 738 00:27:03.809 --> 00:27:06.789 - we you would be very pleasantly surprised 739 00:27:07.490 --> 00:27:08.549 - how good 740 00:27:09.329 --> 00:27:11.910 - AI can be in network operations 741 00:27:12.849 --> 00:27:14.470 - given the right context 742 00:27:15.009 --> 00:27:15.750 - to it. 743 00:27:16.450 --> 00:27:16.950 - Now 744 00:27:17.855 --> 00:27:18.355 - the 745 00:27:19.615 --> 00:27:23.634 - most of the AI exposure that the masses 746 00:27:23.775 --> 00:27:26.975 - had or had so far is interacting with 747 00:27:26.975 --> 00:27:29.475 - the chat GPT or interacting with the cloud. 748 00:27:29.855 --> 00:27:30.355 - Right? 749 00:27:30.975 --> 00:27:32.515 - Now you take 750 00:27:33.419 --> 00:27:34.480 - the brain 751 00:27:35.019 --> 00:27:35.519 - behind 752 00:27:36.139 --> 00:27:37.519 - Claude or Chad GPT, 753 00:27:38.700 --> 00:27:41.279 - and that becomes the brain of an agent. 754 00:27:42.460 --> 00:27:42.960 - But 755 00:27:43.339 --> 00:27:46.379 - now you have a super intelligent, super trained 756 00:27:46.379 --> 00:27:46.879 - PhD, 757 00:27:47.740 --> 00:27:48.240 - but 758 00:27:49.315 --> 00:27:52.535 - the agent doesn't know anything about the network 759 00:27:52.835 --> 00:27:55.494 - or the network history or the architecture 760 00:27:56.115 --> 00:27:56.615 - or 761 00:27:56.994 --> 00:27:57.494 - whatnot. 762 00:27:57.795 --> 00:27:58.295 - Right? 763 00:28:01.470 --> 00:28:04.670 - In a traditional workflow where a human led 764 00:28:04.670 --> 00:28:06.929 - operator is operating a network infrastructure, 765 00:28:08.349 --> 00:28:11.009 - most of the context about the network 766 00:28:11.950 --> 00:28:15.069 - stays as a mental model with the human 767 00:28:15.069 --> 00:28:15.569 - engineer. 768 00:28:16.315 --> 00:28:16.815 - Right? 769 00:28:17.195 --> 00:28:19.535 - Why we set up this network this way? 770 00:28:19.994 --> 00:28:22.654 - What changes we did last three months ago? 771 00:28:23.355 --> 00:28:25.535 - Why did we choose not to 772 00:28:26.315 --> 00:28:27.615 - replace this box? 773 00:28:28.075 --> 00:28:30.095 - Why is the traffic being routed 774 00:28:30.580 --> 00:28:31.880 - this particular way? 775 00:28:32.660 --> 00:28:36.019 - Whatnot. Right? All of these are context which 776 00:28:36.019 --> 00:28:38.200 - currently reside with the human engineer. 777 00:28:39.380 --> 00:28:40.759 - Now for an agent, 778 00:28:41.460 --> 00:28:43.400 - which is super intelligent 779 00:28:44.144 --> 00:28:47.025 - by by by its own nature of the 780 00:28:47.025 --> 00:28:47.525 - LLM, 781 00:28:48.224 --> 00:28:50.805 - but it doesn't know anything about the network. 782 00:28:52.144 --> 00:28:55.265 - To bridge that gap is we need an 783 00:28:55.265 --> 00:28:55.765 - intermediary 784 00:28:56.144 --> 00:28:56.644 - layer 785 00:28:57.089 --> 00:28:59.029 - which brings all these elements 786 00:28:59.569 --> 00:29:00.470 - as context. 787 00:29:01.569 --> 00:29:04.609 - So when an agent finally gets to touch 788 00:29:04.609 --> 00:29:05.269 - the network, 789 00:29:06.049 --> 00:29:08.309 - it goes with all this context, 790 00:29:09.490 --> 00:29:10.549 - built for it. 791 00:29:13.974 --> 00:29:15.755 - Okay. Interesting. Yeah. 792 00:29:16.134 --> 00:29:16.634 - So, 793 00:29:21.414 --> 00:29:22.855 - I was just thinking so okay. So we've 794 00:29:22.855 --> 00:29:24.855 - talked a little bit about data cleaning. We've 795 00:29:24.855 --> 00:29:27.115 - talked a little bit about what the process 796 00:29:27.174 --> 00:29:29.970 - looks like and what you're trying to achieve. 797 00:29:31.470 --> 00:29:33.309 - When it cut now there's gonna be a 798 00:29:33.309 --> 00:29:35.549 - cost trade off here as well. Right? Something 799 00:29:35.549 --> 00:29:37.250 - that has just come up recently 800 00:29:37.789 --> 00:29:39.869 - is that some companies are starting to cut 801 00:29:39.869 --> 00:29:42.429 - back on the amount of AgenTic stuff they 802 00:29:42.429 --> 00:29:45.695 - allow their provide their coders and other people 803 00:29:45.695 --> 00:29:48.095 - working in the company to do, because they're 804 00:29:48.095 --> 00:29:50.654 - worried about the trade off cost of what 805 00:29:50.654 --> 00:29:53.215 - you're gaining versus, like, how much it's costing 806 00:29:53.215 --> 00:29:55.295 - you to get there. Have you done any 807 00:29:55.295 --> 00:29:58.255 - work thinking around what that trade off looks 808 00:29:58.255 --> 00:30:00.720 - like in AI ops? The answer might be 809 00:30:00.720 --> 00:30:02.640 - no, because this is like a totally new 810 00:30:02.640 --> 00:30:04.720 - realm where people haven't really spent a lot 811 00:30:04.720 --> 00:30:06.880 - of time. They're just thinking, oh, you know, 812 00:30:06.880 --> 00:30:07.380 - AI 813 00:30:07.759 --> 00:30:09.700 - is really cool. I can do these things. 814 00:30:09.840 --> 00:30:11.680 - But we haven't really spent a lot of 815 00:30:11.680 --> 00:30:14.100 - time thinking about the cost of things, 816 00:30:14.884 --> 00:30:15.944 - to this point. 817 00:30:18.884 --> 00:30:19.865 - That's true. It's 818 00:30:20.565 --> 00:30:22.264 - it's early days of AI, 819 00:30:22.724 --> 00:30:23.784 - and the exact 820 00:30:24.325 --> 00:30:25.224 - unit economics 821 00:30:25.845 --> 00:30:26.345 - and 822 00:30:26.724 --> 00:30:28.825 - the ROI and the TCO is 823 00:30:29.230 --> 00:30:31.009 - still something to be figured out. 824 00:30:31.470 --> 00:30:33.649 - Okay. Are are do you see a trend 825 00:30:34.190 --> 00:30:36.289 - in your work? Like, are you seeing 826 00:30:36.589 --> 00:30:39.230 - anything that would imply to you the TCO 827 00:30:39.230 --> 00:30:41.710 - is really good or, like, needs to be 828 00:30:41.710 --> 00:30:43.149 - worked out? Like, what are you seeing in 829 00:30:43.149 --> 00:30:43.889 - that area? 830 00:30:45.284 --> 00:30:45.784 - Yeah. 831 00:30:46.085 --> 00:30:50.085 - The the the first set of exposure that 832 00:30:50.085 --> 00:30:50.825 - we had 833 00:30:51.204 --> 00:30:52.345 - with deploying 834 00:30:53.044 --> 00:30:54.424 - agent decay for 835 00:30:55.044 --> 00:30:55.784 - use cases. 836 00:30:56.609 --> 00:30:59.190 - I I'll I'll call out few use cases. 837 00:31:00.049 --> 00:31:00.549 - Right? 838 00:31:01.089 --> 00:31:01.589 - Now 839 00:31:02.049 --> 00:31:04.069 - we have an agent built, 840 00:31:05.809 --> 00:31:06.549 - to do 841 00:31:07.250 --> 00:31:07.750 - security 842 00:31:08.049 --> 00:31:08.549 - compliance 843 00:31:09.009 --> 00:31:09.509 - audit 844 00:31:10.065 --> 00:31:12.784 - for, let's say, SOC two or an IS 845 00:31:12.784 --> 00:31:15.764 - two or Dora or ISO, anything 846 00:31:16.384 --> 00:31:17.365 - on the network. 847 00:31:17.984 --> 00:31:18.484 - Right? 848 00:31:19.505 --> 00:31:20.565 - Agent to 849 00:31:21.184 --> 00:31:23.365 - check the actual compliance 850 00:31:24.409 --> 00:31:25.710 - posture of the network 851 00:31:26.329 --> 00:31:27.630 - for a given framework. 852 00:31:28.409 --> 00:31:29.470 - And the agent 853 00:31:29.929 --> 00:31:31.069 - can do this 854 00:31:31.369 --> 00:31:32.589 - at 100 855 00:31:32.650 --> 00:31:33.390 - the cost 856 00:31:34.409 --> 00:31:37.625 - at 100 x the speed of a traditional 857 00:31:37.924 --> 00:31:39.384 - workflow, which involves 858 00:31:39.845 --> 00:31:40.984 - having an SME 859 00:31:41.684 --> 00:31:44.184 - or having a consulting firm come in, 860 00:31:44.724 --> 00:31:46.025 - getting all the configurations, 861 00:31:46.325 --> 00:31:48.164 - and then going on doing the audit for 862 00:31:48.164 --> 00:31:50.505 - next two weeks and then submitting a report. 863 00:31:51.029 --> 00:31:53.350 - This an agent can do this in run 864 00:31:53.350 --> 00:31:53.850 - time 865 00:31:54.309 --> 00:31:58.070 - at, like, 100 the cost, but, 100 extra 866 00:31:58.070 --> 00:31:58.570 - speed. 867 00:31:59.350 --> 00:31:59.850 - However, 868 00:32:00.950 --> 00:32:04.164 - the cost that it takes the the initial 869 00:32:04.164 --> 00:32:06.424 - investment that it takes to build, 870 00:32:08.005 --> 00:32:10.244 - build a system in place so that an 871 00:32:10.244 --> 00:32:11.465 - agent can run, 872 00:32:13.125 --> 00:32:15.045 - right, that is questionable. I think that's what 873 00:32:15.045 --> 00:32:18.670 - you're referring to. Yeah. Right. Exactly. Yeah. Yeah. 874 00:32:18.670 --> 00:32:20.769 - That that that is still an unknown territory. 875 00:32:21.309 --> 00:32:23.170 - They're all trying to figure out that. 876 00:32:24.830 --> 00:32:27.549 - Interesting. Yeah. Okay. Because that that is an 877 00:32:27.549 --> 00:32:28.769 - issue that I worry. 878 00:32:29.070 --> 00:32:31.490 - I kind of I shouldn't say worry about, 879 00:32:31.549 --> 00:32:33.490 - but it is an interesting area 880 00:32:34.024 --> 00:32:36.664 - where I'm not certain that we've spent a 881 00:32:36.664 --> 00:32:38.365 - lot of time thinking about, 882 00:32:39.304 --> 00:32:39.964 - you know, 883 00:32:40.265 --> 00:32:42.105 - what is the what is the trade off 884 00:32:42.105 --> 00:32:44.904 - in terms of cost. And then going back 885 00:32:44.904 --> 00:32:45.724 - to hallucinations, 886 00:32:46.740 --> 00:32:49.859 - have you found that it's impacting your work 887 00:32:49.859 --> 00:32:50.519 - a lot, 888 00:32:50.980 --> 00:32:52.980 - or is it like something that seems to 889 00:32:52.980 --> 00:32:55.299 - be not so applicable? Because really when we 890 00:32:55.299 --> 00:32:56.359 - talk about hallucinations, 891 00:32:56.900 --> 00:32:59.220 - a lot of the hallucination problems come down 892 00:32:59.220 --> 00:32:59.720 - to, 893 00:33:00.099 --> 00:33:02.119 - are we giving it open at a task 894 00:33:02.335 --> 00:33:03.715 - or closed in a task? 895 00:33:04.095 --> 00:33:07.535 - Right? And how specific can we be and 896 00:33:07.535 --> 00:33:08.994 - how much can you tie 897 00:33:09.295 --> 00:33:09.955 - the LLM 898 00:33:10.414 --> 00:33:10.914 - to 899 00:33:11.535 --> 00:33:13.634 - a particular data set? Right? 900 00:33:14.015 --> 00:33:16.349 - Those seem to me to be questions that 901 00:33:16.750 --> 00:33:18.509 - we don't seem to think about a lot 902 00:33:18.509 --> 00:33:19.569 - in terms of hallucinization. 903 00:33:20.990 --> 00:33:21.490 - So 904 00:33:21.950 --> 00:33:24.269 - have you seen a lot? Are you is 905 00:33:24.269 --> 00:33:26.589 - there something you're thinking about to address that? 906 00:33:26.589 --> 00:33:29.730 - Is it not something you've been concerned about? 907 00:33:30.029 --> 00:33:31.730 - Like, where does that lie? 908 00:33:32.884 --> 00:33:33.784 - Great question. 909 00:33:34.164 --> 00:33:36.825 - By the way, that is that is the, 910 00:33:37.284 --> 00:33:40.325 - that is the crux of the problem statement 911 00:33:40.325 --> 00:33:41.784 - that we're trying to solve. 912 00:33:42.244 --> 00:33:43.224 - Okay. Now 913 00:33:43.605 --> 00:33:45.464 - given if if you give 914 00:33:46.430 --> 00:33:47.650 - a frontier model, 915 00:33:48.829 --> 00:33:49.970 - access to the network 916 00:33:50.750 --> 00:33:52.049 - without the context, 917 00:33:52.430 --> 00:33:53.569 - it's gonna hallucinate 918 00:33:53.869 --> 00:33:54.690 - for sure. 919 00:33:55.390 --> 00:33:57.730 - Now how we bring the hallucination 920 00:33:58.109 --> 00:33:58.609 - down 921 00:33:59.174 --> 00:33:59.914 - is through 922 00:34:00.295 --> 00:34:02.715 - a process called context engineering. 923 00:34:03.735 --> 00:34:04.235 - Meaning, 924 00:34:05.095 --> 00:34:08.315 - before we let the AI touch the infrastructure 925 00:34:08.695 --> 00:34:11.114 - or do any decision making 926 00:34:11.829 --> 00:34:14.010 - on the on any of the tasks, 927 00:34:14.469 --> 00:34:16.730 - we want to build a context 928 00:34:17.190 --> 00:34:18.250 - for the LLM. 929 00:34:19.190 --> 00:34:22.089 - And what does this context look like? Right? 930 00:34:23.190 --> 00:34:23.690 - Now 931 00:34:24.230 --> 00:34:25.929 - let's say there is an agent 932 00:34:26.605 --> 00:34:27.265 - which has 933 00:34:27.565 --> 00:34:28.465 - which has given, 934 00:34:29.085 --> 00:34:30.704 - direct access to 935 00:34:31.244 --> 00:34:32.704 - the router's SSH 936 00:34:33.164 --> 00:34:35.505 - or the router's log. Right? 937 00:34:36.605 --> 00:34:39.744 - It can understand the logs. It can understand 938 00:34:39.805 --> 00:34:42.065 - the SSH output, and it can 939 00:34:42.400 --> 00:34:44.260 - use its LLM as a brain 940 00:34:44.639 --> 00:34:45.880 - and, you know, 941 00:34:46.880 --> 00:34:48.820 - deduce something and come to a conclusion. 942 00:34:50.320 --> 00:34:50.820 - However, 943 00:34:51.440 --> 00:34:54.585 - that conclusion may not be relevant at all 944 00:34:54.824 --> 00:34:55.885 - for for, 945 00:34:57.144 --> 00:34:59.144 - for the from the ops point of view. 946 00:34:59.144 --> 00:34:59.885 - For example, 947 00:35:02.505 --> 00:35:03.484 - given the 948 00:35:04.025 --> 00:35:07.005 - direct access to, let's say, the logs, 949 00:35:07.304 --> 00:35:09.244 - the agent looks at the logs 950 00:35:09.619 --> 00:35:10.440 - and the config, 951 00:35:10.739 --> 00:35:12.440 - and it deduces that 952 00:35:13.140 --> 00:35:15.619 - based on the logs that I see, this 953 00:35:15.619 --> 00:35:16.599 - is the RCA, 954 00:35:17.140 --> 00:35:19.300 - and this looks like a bug, and we're 955 00:35:19.300 --> 00:35:20.920 - gonna up upgrade this, 956 00:35:21.539 --> 00:35:22.760 - router or a firewall 957 00:35:23.059 --> 00:35:25.514 - to the next version. That that is the 958 00:35:25.674 --> 00:35:28.255 - conclusion that the agent has come up to. 959 00:35:28.875 --> 00:35:32.074 - Now the missing context here, which an engineer 960 00:35:32.074 --> 00:35:33.914 - knows, which an agent doesn't know as of 961 00:35:33.914 --> 00:35:34.414 - today, 962 00:35:34.875 --> 00:35:35.375 - is 963 00:35:36.394 --> 00:35:38.894 - what is the change management process involved 964 00:35:39.194 --> 00:35:40.734 - in going through the upgrade? 965 00:35:42.420 --> 00:35:45.960 - Why is that is that device covered under 966 00:35:46.260 --> 00:35:46.760 - the, 967 00:35:47.380 --> 00:35:47.880 - OEM's, 968 00:35:48.820 --> 00:35:49.480 - you know, 969 00:35:50.019 --> 00:35:52.659 - license subscription for an upgrade? Does it qualify 970 00:35:52.659 --> 00:35:53.480 - for an upgrade? 971 00:35:54.179 --> 00:35:56.579 - What happens if I upgrade? What else will 972 00:35:56.579 --> 00:35:57.079 - break? 973 00:35:57.675 --> 00:35:59.914 - These are all the context that a human 974 00:35:59.914 --> 00:36:01.215 - engineer has today. 975 00:36:02.074 --> 00:36:03.375 - We have to bring 976 00:36:03.914 --> 00:36:05.215 - or it's a effort 977 00:36:05.515 --> 00:36:08.795 - to bring a similar level of context to 978 00:36:08.795 --> 00:36:09.614 - an agent. 979 00:36:10.635 --> 00:36:12.015 - And to do that, 980 00:36:12.579 --> 00:36:14.679 - we are stitching up multiple, 981 00:36:15.940 --> 00:36:16.440 - you 982 00:36:16.739 --> 00:36:19.940 - know, sources of information like the intent, which 983 00:36:19.940 --> 00:36:20.839 - is the 984 00:36:21.460 --> 00:36:23.780 - the golden blueprint of what the network is 985 00:36:23.780 --> 00:36:26.280 - supposed to be, the live network states, 986 00:36:26.594 --> 00:36:28.054 - and the operational history 987 00:36:28.434 --> 00:36:29.175 - and whatnot. 988 00:36:29.554 --> 00:36:30.215 - So that, 989 00:36:31.394 --> 00:36:32.454 - when the agent 990 00:36:32.914 --> 00:36:35.014 - has to decide, it has all these, 991 00:36:36.514 --> 00:36:39.175 - information as context available to it, 992 00:36:39.659 --> 00:36:42.400 - And this reduces the scope of hallucination 993 00:36:43.019 --> 00:36:45.019 - because it's not operating on, 994 00:36:47.099 --> 00:36:49.579 - a broad horizon. It is operating on a 995 00:36:49.579 --> 00:36:50.480 - very confined 996 00:36:50.859 --> 00:36:51.920 - set of context. 997 00:36:53.355 --> 00:36:55.195 - Okay. Yeah. That's the way I would expect 998 00:36:55.195 --> 00:36:56.635 - you would solve the problem. I was just 999 00:36:56.635 --> 00:36:59.375 - curious, you know, like, if you'd encountered it. 1000 00:36:59.514 --> 00:37:00.974 - Yeah. Cool. Interesting. 1001 00:37:02.155 --> 00:37:04.474 - Is there anything else about AI ops that 1002 00:37:04.474 --> 00:37:07.275 - you want to talk about, like troubleshooting process 1003 00:37:07.275 --> 00:37:07.934 - or anything 1004 00:37:08.300 --> 00:37:10.300 - before we start thinking about wrapping up and, 1005 00:37:10.300 --> 00:37:12.780 - like, you know, ending? Is there are there 1006 00:37:12.780 --> 00:37:15.260 - other specific topics you think are of real 1007 00:37:15.260 --> 00:37:15.760 - interest, 1008 00:37:16.539 --> 00:37:19.019 - in the AI AI ops space that you're 1009 00:37:19.019 --> 00:37:20.559 - looking at or you're seeing? 1010 00:37:22.795 --> 00:37:25.914 - Maybe Sushant can take up the question of, 1011 00:37:26.875 --> 00:37:27.375 - troubleshooting 1012 00:37:27.835 --> 00:37:28.335 - and 1013 00:37:29.035 --> 00:37:30.414 - how agents are 1014 00:37:31.035 --> 00:37:33.035 - able to do troubleshooting today, what is the 1015 00:37:33.035 --> 00:37:33.855 - novel approach, 1016 00:37:34.315 --> 00:37:36.575 - which which has not been possible before. 1017 00:37:37.089 --> 00:37:38.550 - Okay. Cool. Yeah. Go ahead. 1018 00:37:39.170 --> 00:37:39.670 - Yeah. 1019 00:37:40.609 --> 00:37:43.589 - So what what difference that we do in, 1020 00:37:44.690 --> 00:37:45.409 - through through, 1021 00:37:45.730 --> 00:37:47.730 - the whole semantic thing, we call it as 1022 00:37:47.730 --> 00:37:48.469 - CR that, 1023 00:37:49.250 --> 00:37:50.550 - understands the network. 1024 00:37:51.195 --> 00:37:51.934 - First understands 1025 00:37:52.635 --> 00:37:54.015 - the intent of the network, 1026 00:37:54.555 --> 00:37:56.414 - how it's been designed to, 1027 00:37:56.795 --> 00:37:59.215 - what is the right way to troubleshoot. 1028 00:38:00.235 --> 00:38:00.735 - And, 1029 00:38:02.635 --> 00:38:05.035 - the agents look at two things. One, this 1030 00:38:05.035 --> 00:38:05.855 - is the network. 1031 00:38:06.230 --> 00:38:08.630 - This is how a particular user is supposed 1032 00:38:08.630 --> 00:38:10.789 - to authenticate. And I'm taking an example of 1033 00:38:10.789 --> 00:38:11.929 - a wireless authentication. 1034 00:38:12.549 --> 00:38:13.449 - So let's say 1035 00:38:13.829 --> 00:38:15.750 - it goes through a multiple things. You know? 1036 00:38:15.750 --> 00:38:16.809 - It goes through your 1037 00:38:18.324 --> 00:38:21.125 - Wi Fi access points, and then it goes 1038 00:38:21.125 --> 00:38:21.864 - on to, 1039 00:38:22.885 --> 00:38:25.065 - let's say, an authentication server. 1040 00:38:25.684 --> 00:38:26.184 - So, 1041 00:38:27.364 --> 00:38:29.864 - here, what we do first is we understand 1042 00:38:29.925 --> 00:38:32.244 - the intent. We understand what it is configured 1043 00:38:32.244 --> 00:38:33.819 - for, how it is supposed to work. We 1044 00:38:33.819 --> 00:38:35.119 - take context from 1045 00:38:35.579 --> 00:38:36.079 - everywhere. 1046 00:38:36.539 --> 00:38:38.480 - And this context becomes the blueprint 1047 00:38:38.780 --> 00:38:41.019 - for us to understand the SOP. You know, 1048 00:38:41.019 --> 00:38:43.679 - it's a dynamic SOP that the agent writes 1049 00:38:43.980 --> 00:38:45.119 - based on the context. 1050 00:38:45.675 --> 00:38:46.175 - And 1051 00:38:46.635 --> 00:38:47.755 - that SOPs are, 1052 00:38:48.155 --> 00:38:49.855 - you know, taken forward onto, 1053 00:38:50.394 --> 00:38:51.795 - let's say, troubleshoot, 1054 00:38:52.235 --> 00:38:53.295 - particular issues. 1055 00:38:53.835 --> 00:38:54.335 - Similarly, 1056 00:38:54.954 --> 00:38:57.035 - since we look the whole network as a 1057 00:38:57.035 --> 00:38:57.855 - single entity, 1058 00:38:58.800 --> 00:39:00.880 - any use cases that come up, any any 1059 00:39:00.960 --> 00:39:03.280 - anything dynamic city that comes up in the 1060 00:39:03.280 --> 00:39:04.420 - network that becomes, 1061 00:39:05.199 --> 00:39:07.539 - far more easier for the agents 1062 00:39:07.920 --> 00:39:11.219 - to work on, to, you know, to troubleshoot 1063 00:39:11.440 --> 00:39:12.180 - on, etcetera. 1064 00:39:13.734 --> 00:39:14.234 - Okay. 1065 00:39:14.855 --> 00:39:16.554 - Okay. Interesting. Alright. 1066 00:39:16.934 --> 00:39:17.434 - Cool. 1067 00:39:18.614 --> 00:39:21.655 - So, basically, the troubleshooting assistant, is is it 1068 00:39:21.655 --> 00:39:23.815 - more of an assistant at this point, or 1069 00:39:23.815 --> 00:39:25.894 - is it doing the job itself? Is it 1070 00:39:25.894 --> 00:39:27.675 - noticing things that are broken 1071 00:39:28.089 --> 00:39:29.630 - and doing the troubleshooting 1072 00:39:30.329 --> 00:39:32.089 - and saying, I think this is what the 1073 00:39:32.089 --> 00:39:34.670 - root cause is, or is it more interactive? 1074 00:39:34.969 --> 00:39:36.730 - Like, I see a problem. Let me use 1075 00:39:36.730 --> 00:39:38.269 - the AI agent 1076 00:39:38.730 --> 00:39:41.844 - to help me understand the problem. Which of 1077 00:39:41.844 --> 00:39:43.684 - those would you say you're closer to right 1078 00:39:43.684 --> 00:39:44.184 - now? 1079 00:39:45.204 --> 00:39:48.164 - It understands the problem. It understands the drift 1080 00:39:48.164 --> 00:39:49.625 - and why it has happened. 1081 00:39:50.005 --> 00:39:52.324 - It also understands what has happened in the 1082 00:39:52.324 --> 00:39:54.909 - past. Let's say there was some ideas and 1083 00:39:54.909 --> 00:39:56.989 - ticket about it or something and how it 1084 00:39:56.989 --> 00:40:00.269 - was resolved. It gives you suggestion to solve 1085 00:40:00.269 --> 00:40:02.289 - it. Obviously, it gives you the recommendation. 1086 00:40:03.070 --> 00:40:05.809 - And there's an option to have it remediated, 1087 00:40:07.255 --> 00:40:09.414 - with human in the loop. There's a lot 1088 00:40:09.414 --> 00:40:12.315 - of trust issues that most of the organizations 1089 00:40:12.614 --> 00:40:14.554 - do have on agent taking actions. 1090 00:40:15.014 --> 00:40:15.835 - So Yeah. 1091 00:40:16.215 --> 00:40:18.215 - There's human in the loop to make sure 1092 00:40:18.215 --> 00:40:20.855 - that right action is taken on, and it's 1093 00:40:20.855 --> 00:40:21.755 - on and over. 1094 00:40:22.590 --> 00:40:23.090 - Alright. 1095 00:40:23.710 --> 00:40:24.610 - Cool. Alright. 1096 00:40:24.910 --> 00:40:26.829 - That all sounds really cool. I mean, there's 1097 00:40:26.829 --> 00:40:29.230 - a lot of things I would think of 1098 00:40:29.230 --> 00:40:31.470 - in ways to do this, but, yeah, I'm 1099 00:40:31.470 --> 00:40:33.470 - just trying to get my my arms around 1100 00:40:33.470 --> 00:40:35.809 - for listeners, like, what this space entails 1101 00:40:36.430 --> 00:40:39.464 - right now Right. And and where things are. 1102 00:40:39.684 --> 00:40:42.744 - So alright. Anything else before we wrap up? 1103 00:40:43.684 --> 00:40:44.184 - Any 1104 00:40:44.565 --> 00:40:46.565 - comments or thoughts? I would normally ask Tom. 1105 00:40:46.565 --> 00:40:49.284 - But Tom left us. Tom and eight and 1106 00:40:49.284 --> 00:40:51.545 - and Audrey left us, the bums. 1107 00:40:56.590 --> 00:40:58.910 - Thanks for having us on your podcast, Les. 1108 00:40:58.910 --> 00:41:01.869 - It's a privilege and honor speaking to you. 1109 00:41:01.869 --> 00:41:03.570 - Any any old time. So, 1110 00:41:04.030 --> 00:41:04.530 - alright. 1111 00:41:05.070 --> 00:41:05.570 - So, 1112 00:41:06.385 --> 00:41:08.545 - actually, do you, actually, do you have a 1113 00:41:08.545 --> 00:41:10.005 - pa a buy a 1114 00:41:10.385 --> 00:41:12.785 - blog, or do you write regularly at LinkI 1115 00:41:12.785 --> 00:41:14.005 - or any other places 1116 00:41:14.465 --> 00:41:16.864 - where people can follow you or, like, x 1117 00:41:16.864 --> 00:41:17.985 - Twitter or x, 1118 00:41:18.670 --> 00:41:20.670 - LinkedIn or anything like that? Is there any 1119 00:41:20.670 --> 00:41:22.670 - other like, if people wanna find out more 1120 00:41:22.670 --> 00:41:24.929 - or want to follow you or whatever? 1121 00:41:26.750 --> 00:41:29.510 - I'm quite active on LinkedIn, and I Okay. 1122 00:41:29.630 --> 00:41:31.329 - Blog on LinkedIn. Yeah. 1123 00:41:31.630 --> 00:41:33.994 - Alright. Cool. And you, Shushant, anything? 1124 00:41:35.574 --> 00:41:37.494 - I'm I believe we're on LinkedIn, but I 1125 00:41:37.494 --> 00:41:40.554 - don't, you know, blog a lot. Yeah. Okay. 1126 00:41:40.614 --> 00:41:43.194 - So Rakshit has most of the 10 guests. 1127 00:41:43.335 --> 00:41:45.510 - Alright. That's cool. Just asking cause if people 1128 00:41:45.510 --> 00:41:47.190 - wanna find you, you know, it's good to 1129 00:41:47.190 --> 00:41:49.030 - to let them know, like, if there's a 1130 00:41:49.030 --> 00:41:50.869 - blog or something, what that URL is. But 1131 00:41:50.869 --> 00:41:52.869 - if you're just on LinkedIn, that's cool. People 1132 00:41:52.869 --> 00:41:53.929 - can look you up. 1133 00:41:54.550 --> 00:41:55.050 - Alright. 1134 00:41:55.510 --> 00:41:57.110 - I am Russ White. You can always find 1135 00:41:57.110 --> 00:41:58.309 - me here at the head. You can find 1136 00:41:58.309 --> 00:42:00.635 - me at rule eleven dot tech. 1137 00:42:01.255 --> 00:42:03.094 - You can find me on LinkedIn. You can 1138 00:42:03.094 --> 00:42:04.635 - sometimes find me on x, 1139 00:42:05.094 --> 00:42:07.355 - here and there, depends on what's going on. 1140 00:42:07.494 --> 00:42:09.655 - And, we know we live in an attention 1141 00:42:09.655 --> 00:42:11.735 - driven economy. We thank you for listening all 1142 00:42:11.735 --> 00:42:13.414 - the way to the bitter end of this 1143 00:42:13.414 --> 00:42:14.875 - episode of the hedge, 1144 00:42:15.230 --> 00:42:17.010 - and we will catch you next time.