WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=1_lX0m64-aI NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/1_lX0m64-aI.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:12.560 --> 00:00:16.400 I'm going to go through a bunch of keywords and you guys get interactive. s3 00:00:16.400 --> 00:00:20.040 I'm going to ask you guys to put your hands up as I go through a bunch of keywords. s4 00:00:20.040 --> 00:00:24.240 So, hands up if you heard about agent memory. s5 00:00:24.480 --> 00:00:34.920 That makes me unbelievably happy because I did this I did a talk on agent memory about a year ago and no one in the room knew what I was talking about. s6 00:00:34.920 --> 00:00:46.440 So, it was also at AI Engineers San Francisco and once upon a time, maybe around 3 years ago, people thought I was crazy when I was talking about we need to structure memory, we need to model memory in agents. s7 00:00:46.440 --> 00:00:52.320 And now everyone here knows about agent memory and hopefully you're implementing it within your agent. s8 00:00:52.320 --> 00:00:57.160 So, I'm just going to quickly go over agent memory, just very quickly. s9 00:00:57.160 --> 00:01:04.559 So, agent memory is basically how an agent or the memory mechanism an agent uses to retain, reuse, recall, and refine information. s10 00:01:04.559 --> 00:01:07.400 I call it the four Rs, it's easy to remember. s11 00:01:07.400 --> 00:01:10.520 Um but there's also taxonomy of agent memory. s12 00:01:10.520 --> 00:01:12.320 Short-term, long-term is easy. s13 00:01:12.320 --> 00:01:19.760 Then you go into the cognitive function of agent memory, semantic, episodic, procedural, working memory. s14 00:01:19.760 --> 00:01:23.000 Then we go into an implementation that we're going to see a lot today. s15 00:01:23.000 --> 00:01:29.600 And I've done this so many times online, but let's take pictures and you guys are going to get the slides as well. s16 00:01:29.600 --> 00:01:36.680 So, this is all the taxonomy of memory, the what I'm trying to educate people on and I've been educating people on for the past 3 years. s17 00:01:36.680 --> 00:01:38.920 But in this talk, we're going to take it further. s18 00:01:38.920 --> 00:01:41.680 So, another keyword coming up. s19 00:01:41.680 --> 00:01:44.640 Hands up who's heard of the control plane? s20 00:01:44.640 --> 00:01:45.760 Nice, less so. s21 00:01:45.760 --> 00:01:48.480 We're getting We're getting less hands and I have a theory. s22 00:01:48.480 --> 00:01:53.000 As we go through the keywords, you're going to see the number of hands in the presentation reduce. s23 00:01:53.000 --> 00:01:58.680 Eventually, if things go correctly, there will be no hands or maybe one or so. s24 00:01:58.680 --> 00:02:00.972 So, the control plane, we can define it. s25 00:02:00.972 --> 00:02:02.372 Most people know what it is. s26 00:02:02.372 --> 00:02:10.892 It's essentially a centralized management system for how your agent is governed, with the tools they have access to, what they can do, and what they cannot do. s27 00:02:10.892 --> 00:02:15.412 So, there's a lot of different software companies positioning themselves as a control plane. s28 00:02:15.412 --> 00:02:18.492 And yes, we do need these sort of systems. s29 00:02:18.492 --> 00:02:22.052 But in this talk, I'm going to take you a little bit further. s30 00:02:22.052 --> 00:02:24.292 So, this is just an idea of what the control plane is. s31 00:02:24.292 --> 00:02:32.372 So, identity for your agents, the policy and guardrails, the orchestration to permission, and all the auditing. s32 00:02:32.372 --> 00:02:36.492 And these are the components of typical control planes. s33 00:02:36.492 --> 00:02:38.572 All right, another keyword is coming. s34 00:02:38.572 --> 00:02:40.852 You guys ready? s35 00:02:42.852 --> 00:02:43.772 All right, let's go. s36 00:02:43.772 --> 00:02:46.212 Software factory. s37 00:02:46.732 --> 00:02:47.732 Okay, less hands. s38 00:02:47.732 --> 00:02:48.452 We're getting there. s39 00:02:48.452 --> 00:02:52.892 It's It's We're almost in line with what I have on the slides. s40 00:02:52.892 --> 00:02:56.132 So, software factory is actually an interesting term. s41 00:02:56.132 --> 00:03:00.332 So, I did my internship at General Electric like eons ago. s42 00:03:00.332 --> 00:03:12.292 And software factory is a term that someone called Bob Bemer at GE at the time in 1968, he coined the term software factory in one of his paper where he was talking about how we can create s43 00:03:12.292 --> 00:03:15.812 um programs in measurable and repeatable way. s44 00:03:15.812 --> 00:03:18.932 And essentially, he coined the term software factory. s45 00:03:18.932 --> 00:03:21.612 So, it's not a new term within the agentic era. s46 00:03:21.612 --> 00:03:26.932 It's a term that we've used in computer science a very, very long time ago. s47 00:03:26.932 --> 00:03:29.452 I'll wait for people to take pictures. s48 00:03:29.452 --> 00:03:37.372 So, software factory, you can just imagine it as you want to create a system that is able to churn out software reliably, right? s49 00:03:37.372 --> 00:03:47.172 You give it a bunch of um processes and systematic workflow it's meant to follow whenever it's building a software, and you have the agents execute this, usually multiple agents in parallel. s50 00:03:47.172 --> 00:03:51.852 Again, different companies position themselves in this space. s51 00:03:51.852 --> 00:03:54.412 All right, another keyword. s52 00:03:54.412 --> 00:03:56.972 Intelligence plane. s53 00:03:57.132 --> 00:03:58.892 Any hands up? s54 00:03:58.892 --> 00:03:59.932 No, I failed. s55 00:03:59.932 --> 00:04:01.372 I got three hands up. s56 00:04:01.372 --> 00:04:04.374 Um yeah, the intelligence plane is what we're going to be talking about today. s57 00:04:04.374 --> 00:04:12.254 And I'm going to make a promise that I did in my previous talk last year when I was talking about agent memory when people thought I was crazy. s58 00:04:12.254 --> 00:04:22.014 And this promise is the same I'm going to I'm going to make, which is the information I'm going to give you today, everything I'm going to say, is going to put you 6 months ahead s59 00:04:22.014 --> 00:04:24.814 of where most people are today. s60 00:04:24.814 --> 00:04:27.694 So, I made that promise last year talking about agent memory. s61 00:04:27.694 --> 00:04:30.294 I'm making that promise this year talking about the intelligence plane. s62 00:04:30.294 --> 00:04:36.334 So, when I come back next year and I talk about the intelligence plane, everyone's going to put their hands up. s63 00:04:36.334 --> 00:04:37.934 Don't don't quote me. s64 00:04:37.934 --> 00:04:38.654 How about this one? s65 00:04:38.654 --> 00:04:39.774 AF is Midway. s66 00:04:39.774 --> 00:04:42.214 Who's heard about this? s67 00:04:42.214 --> 00:04:42.974 No? s68 00:04:42.974 --> 00:04:45.654 Okay, let's uh do a bit of history lesson. s69 00:04:45.654 --> 00:04:56.734 So, in World War II um in World War II uh Pearl Harbor was getting um attacked by the Japanese Imperial Navy, right? s70 00:04:56.734 --> 00:05:03.774 And there was a bunch of groups that were meant to be intercepting communication from the Japanese army. s71 00:05:03.774 --> 00:05:05.294 And there were three groups. s72 00:05:05.294 --> 00:05:07.134 So, we're only going to talk about two groups here. s73 00:05:07.134 --> 00:05:12.534 One group was based in Washington and one group was based in Pearl Harbor. s74 00:05:12.534 --> 00:05:15.334 And there was a code AF. s75 00:05:15.334 --> 00:05:23.534 Now, AF stood for something and it was very critical to the outcome of the actual um Pacific War itself. s76 00:05:23.534 --> 00:05:37.094 But to cut the long story short, these two groups had a bunch of fragmented information from communication from the Japanese Navy force that they were they were that they were meant to codebreak. s77 00:05:37.094 --> 00:05:48.054 So, they had a lot of fragmented information and basically the one group, which is the group based in Pearl Harbor, a bunch of folks that were basically in a basement s78 00:05:48.054 --> 00:06:00.934 collecting loads of information and intercepted messages on paper and storing it for over months, were able to decode this and they found out that AF is actually short for Midway. s79 00:06:00.934 --> 00:06:01.334 Right? s80 00:06:01.334 --> 00:06:03.134 AF is Midway. s81 00:06:03.134 --> 00:06:03.414 Right? s82 00:06:03.414 --> 00:06:06.615 So, the Japanese were using AF as code word for Midway. s83 00:06:06.615 --> 00:06:12.575 Midway was a little island located in between Asia and North America. s84 00:06:12.575 --> 00:06:17.135 So, this was You can look this up, control of the Pacific. s85 00:06:17.135 --> 00:06:22.575 But, a key thing I wanted to tell you is the difference between control and intelligence. s86 00:06:22.575 --> 00:06:38.455 Because that small group that was in Pearl Harbor, located in a basement, and they were deciphering all the um Japanese communication, and using a lot of papers to record everything for months, which is essentially memory, s87 00:06:38.455 --> 00:06:45.095 what was able to actually decode and position um the American Navy at an advantage. s88 00:06:45.095 --> 00:06:51.575 So, whereas Washington, which was more centralized, had the same information, right? s89 00:06:51.575 --> 00:06:52.815 There was no difference in information. s90 00:06:52.815 --> 00:07:06.535 They had the same information, probably more resources, and they had control, and they had organization, came up with an entirely different um uh interpretation of the data. s91 00:07:06.535 --> 00:07:12.415 And the one that was chosen was the folks of our Station Hypo, um the code breakers there. s92 00:07:12.415 --> 00:07:18.895 And the reason they were able to also do our hypothesis on the fact that AF does mean Mid- Midway. s93 00:07:18.895 --> 00:07:31.855 So, what I'm trying to say is everyone in this room has the same models, probably the same access to the same information, but that doesn't equate to intelligence, and it doesn't equate to you having that unique differentiator. s94 00:07:31.855 --> 00:07:37.815 So, today, I'm going to talk about why we really need to start talking about the intelligence plane. s95 00:07:37.815 --> 00:07:40.095 So, I'm going to go through the agent stack. s96 00:07:40.095 --> 00:07:41.335 I've spoken about the agent stack. s97 00:07:41.335 --> 00:07:47.375 We know what the agent stack is, and then the control plane cuts through a certain level of the agent stack. s98 00:07:47.375 --> 00:07:49.855 Now, what I'm proposing to you is nothing new. s99 00:07:49.855 --> 00:07:53.655 It's just a way of thinking about the software you're building today. s100 00:07:53.655 --> 00:07:55.975 So, this is the intelligence plane. s101 00:07:55.975 --> 00:08:09.497 It cuts through the entire stack, and creates a dimension where you as engineers can start to build agents that are not only orchestrated, but they can actually learn from data and memory as well. s102 00:08:09.497 --> 00:08:16.137 So, we're going to we're going to move from theory into more practical concepts. s103 00:08:16.137 --> 00:08:19.297 So, what is the intelligence plane? s104 00:08:19.297 --> 00:08:28.737 It's essentially a part of your agentic system or your AI infrastructure that allows you to understand or allows your agent to get better over time. s105 00:08:28.737 --> 00:08:40.297 It encapsulate what the agent should know, which is basically knowledge, what it should remember, which is basically memory, what it's allowed to see, which is just governance, data access, s106 00:08:40.297 --> 00:08:43.537 and how it gets better, which is continual learning. s107 00:08:43.537 --> 00:08:44.857 So, there's nothing new here. s108 00:08:44.857 --> 00:08:51.177 Some of you are already doing some of these things already, and some of you are going to start doing it after this talk. s109 00:08:51.177 --> 00:08:54.497 So, that is the difference between the control plane and the intelligence plane. s110 00:08:54.497 --> 00:08:55.977 Are we following? s111 00:08:55.977 --> 00:08:56.257 Right? s112 00:08:56.257 --> 00:08:57.137 Nodding? s113 00:08:57.137 --> 00:08:58.217 Okay, this is good. s114 00:08:58.217 --> 00:08:59.297 Let's go. s115 00:08:59.297 --> 00:09:05.337 Um I'm going to take you through this timeline of where most of you have been and where most of you are going to get to. s116 00:09:05.337 --> 00:09:11.977 First, we were all prompting the models, and we all had like prompt layers or tools where we can see all our prompts. s117 00:09:11.977 --> 00:09:15.777 Then we moved into tools that can allow us to see the context, context layer. s118 00:09:15.777 --> 00:09:23.857 And most of you are now in the control plane error, when you guys are looking for softwares or building softwares that allows you to orchestrate this agent. s119 00:09:23.857 --> 00:09:30.617 So, where we're going to go to is intelligence plane or where you should be heading to is intelligence plane. s120 00:09:30.617 --> 00:09:39.137 And just to get this out there, throughout all the slides, one thing we can understand and agree on, information does not equal intelligence. s121 00:09:39.137 --> 00:09:42.377 So, take that away, and now we can start getting a bit technical. s122 00:09:42.377 --> 00:09:43.817 So, this is the agent stack. s123 00:09:43.817 --> 00:09:46.017 We spoke about the agent plane. s124 00:09:46.017 --> 00:09:48.977 And when we start looking about let's talk about the control plane. s125 00:09:48.977 --> 00:09:58.617 When we start thinking about intelligence in the control plane or intelligence added onto the control plane, we start to think about when you're looking at gateway and our connectivity, s126 00:09:58.617 --> 00:10:03.417 you start to think about end-to-end user identity connected to the data. s127 00:10:03.417 --> 00:10:04.217 Right? s128 00:10:04.217 --> 00:10:13.818 When you start to think about orchestration, you start to think about the context assembly, you start to think about things such as memory and contest construction. s129 00:10:13.818 --> 00:10:15.538 Then all the way to the memory core. s130 00:10:15.538 --> 00:10:19.018 I'm skipping to the memory core because that's my sweet spot, database. s131 00:10:19.018 --> 00:10:21.938 Um we start to talk about database, but we take things further. s132 00:10:21.938 --> 00:10:26.578 We start to talk about ontology and semantic layers and memory as well. s133 00:10:26.578 --> 00:10:29.338 Then infrastructure and platform is the important bit. s134 00:10:29.338 --> 00:10:35.298 Which is where we start to talk about inference, fine-tuning, and continuous learning, right? s135 00:10:35.298 --> 00:10:39.778 When we go beyond the deployment that the control plane gives you. s136 00:10:39.778 --> 00:10:44.898 So, these are all the components of the control plane and the intelligence plane. s137 00:10:44.898 --> 00:10:48.658 There's going to be a reference architecture that you would be able to take a picture of. s138 00:10:48.658 --> 00:10:54.098 And this idea, I wish I came up with new ideas, but this idea I'm actually proposing to you is nothing new. s139 00:10:54.098 --> 00:10:55.578 Um it's been proposed several times. s140 00:10:55.578 --> 00:11:00.498 In fact, in if you If anyone's done networking, this might be familiar to you. s141 00:11:00.498 --> 00:11:14.258 Um some folks have proposed that there needs to be a knowledge plane on top of management and configuration platforms that are meant to actually see um the the routes of different circuits, s142 00:11:14.258 --> 00:11:16.778 which is essentially a a control plane, right? s143 00:11:16.778 --> 00:11:23.618 So, a control plane, a management plane, then folks have um uh proposed a knowledge plane. s144 00:11:23.618 --> 00:11:28.698 And this is what I'm proposing for the agentic systems we're building, an intelligence plane. s145 00:11:28.698 --> 00:11:29.738 We're going to have a demo. s146 00:11:29.738 --> 00:11:32.298 I'm looking at the time to make sure I get to the demo. s147 00:11:32.298 --> 00:11:33.818 And we like equations. s148 00:11:33.818 --> 00:11:35.978 Agent equals model plus harness. s149 00:11:35.978 --> 00:11:40.298 Then you take that into control, that equals governance plus security. s150 00:11:40.298 --> 00:11:50.658 But when you bring intelligence, you add knowledge and memory and learning, it gives you that multiplier effect on your entire on the entire way you're building agents today. s151 00:11:50.658 --> 00:11:52.538 And I'm going to show you what that looks like. s152 00:11:52.538 --> 00:11:55.258 I'm trying to think, should I go to the demo now or later? s153 00:11:55.258 --> 00:11:57.058 Um we'll do a demo version. s154 00:11:57.058 --> 00:11:57.898 Oh, I'll go back. s155 00:11:57.898 --> 00:11:59.418 People were taking pictures. s156 00:11:59.418 --> 00:12:03.658 You are going to get the slides um in a second. s157 00:12:03.658 --> 00:12:03.938 All right. s158 00:12:03.938 --> 00:12:13.458 So, everything I talk about within the intelligence plane, I've tried to capture as just high-level system components. s159 00:12:13.458 --> 00:12:16.299 So, all of these would be different providers today. s160 00:12:16.299 --> 00:12:16.819 Right? s161 00:12:16.819 --> 00:12:22.379 So, we have the agent runtime that your agent is going to be executing on. s162 00:12:22.379 --> 00:12:34.379 Then your agent runtime needs memory and knowledge, probably has a memory manager on top that sits on top of data platform, and then your agent runtime is able to put in traces of memory s163 00:12:34.379 --> 00:12:43.099 into your memory manager, and that could then go into a continual learning loop that allows you to then change models back into your agent runtime. s164 00:12:43.099 --> 00:12:43.459 Right? s165 00:12:43.459 --> 00:12:47.339 And you can put this all on the plane that we've been talking about. s166 00:12:47.339 --> 00:12:51.619 So, we have the data plane, the control plane, the intelligence plane. s167 00:12:51.619 --> 00:12:54.739 I'm not going to go over this cuz we already know what the control plane is. s168 00:12:54.739 --> 00:12:58.819 We know what the intelligence plane is, and that's the key thing that is missing. s169 00:12:58.819 --> 00:13:03.459 So, in the intelligence plane, we have a couple things that are actually quite important. s170 00:13:03.459 --> 00:13:05.139 I'm going a bit too fast. s171 00:13:05.139 --> 00:13:08.299 We have a couple things that are quite important I'm going to show in a demo. s172 00:13:08.299 --> 00:13:10.899 And that is what I'm going to be showing in a second. s173 00:13:10.899 --> 00:13:13.659 So, why is the industry ready for the intelligence plane? s174 00:13:13.659 --> 00:13:15.939 Why am I talking about the intelligence plane now? s175 00:13:15.939 --> 00:13:18.899 So, very quickly, we have frontier AI models. s176 00:13:18.899 --> 00:13:28.579 Um the scaling law allowed us to actually predict the benefits we could get from these models, and then we have the open ways are very like 6 months or 3 months behind. s177 00:13:28.579 --> 00:13:32.219 But we have we had something that happened recently, which was decision models. s178 00:13:32.219 --> 00:13:32.459 Right? s179 00:13:32.459 --> 00:13:35.939 And they changed the trajectory or the way that we started to think of these systems. s180 00:13:35.939 --> 00:13:37.859 The way I thought about the systems, right? s181 00:13:37.859 --> 00:13:42.219 It made me start to think systems that are very fundamental architectural level. s182 00:13:42.219 --> 00:13:47.339 Looking at my system and thinking, okay, what here could I replace as a decision model? s183 00:13:47.339 --> 00:13:56.939 So, this was the traditional way we were seeing things a few years ago, and today you can have models all through your retrieval pipeline. s184 00:13:56.939 --> 00:14:01.459 So, one thing I'm going to do is I'm going to go into a demo right here. s185 00:14:01.459 --> 00:14:03.139 This is the QR for the demo. s186 00:14:03.139 --> 00:14:05.979 A bit going ahead of myself. s187 00:14:07.499 --> 00:14:08.579 Here we go. s188 00:14:08.579 --> 00:14:16.621 So, in this demo, which you're going to get the code to, I'm not going to run it live cuz the internet here isn't great as you guys probably know. s189 00:14:16.621 --> 00:14:18.821 Um this is a simple chat interface. s190 00:14:18.821 --> 00:14:21.701 I have a governed analyst, I have a naive analyst. s191 00:14:21.701 --> 00:14:24.861 My governed analyst is built on what is an intelligence plane. s192 00:14:24.861 --> 00:14:30.621 It gives it the ontology layer, semantic layer, gives it everything we've seen in that reference architecture. s193 00:14:30.621 --> 00:14:36.501 And very quickly, just to show you the effect of having an intelligence plane. s194 00:14:36.861 --> 00:14:41.141 The key thing is having one, saving on tokenomics, right? s195 00:14:41.141 --> 00:14:44.821 Tokenomics is cost, accuracy, and latency. s196 00:14:44.821 --> 00:14:54.301 And we have two agents that are able to execute the same questions, but the one that is powered by an intelligence plane is a bit smarter. s197 00:14:54.301 --> 00:14:57.101 It gets things right 99% of the time. s198 00:14:57.101 --> 00:14:59.661 And it does it with less token consumption. s199 00:14:59.661 --> 00:15:05.381 In 32,000 token compared to the 200 thousand token with the naive agent. s200 00:15:05.381 --> 00:15:09.261 And then we have all of the guard controls in place and the cache as well. s201 00:15:09.261 --> 00:15:12.301 And we have some prompt caching mechanism as well. s202 00:15:12.301 --> 00:15:15.181 So, when you actually start I'm working backwards here. s203 00:15:15.181 --> 00:15:25.541 When you start to implement some of the things I'm going to show you to you in a second, you start to implement systems that are able to actually one, be optimized for production, s204 00:15:25.541 --> 00:15:29.981 and two, learn and grow with the data that you're capturing. s205 00:15:29.981 --> 00:15:33.181 So, over here I just want to talk about the ontology and the semantic layer. s206 00:15:33.181 --> 00:15:36.941 I'm kind of skipping between demos here cuz we're short on time. s207 00:15:36.941 --> 00:15:46.701 So, with the ontology and semantic layer, I asked the question uh to both agents around um what was a counterparty, right? s208 00:15:46.701 --> 00:15:53.341 And the ontology layer adds a level of accuracy that you can't get by just doing simple keyword search. s209 00:15:53.341 --> 00:15:59.781 So, in the demo, this is a keyword search where we're able to actually find some um entities, right? s210 00:15:59.781 --> 00:16:09.701 And the agent was able to answer without the ontology, but with the ontology, the answer actually improves and it gets the right answer using the ontology. s211 00:16:09.701 --> 00:16:11.941 Same with the semantic layer. s212 00:16:11.941 --> 00:16:20.542 The agent the naive agent without any semantic layer, we asked it a question and it got the we asked it what does concentration mean? s213 00:16:20.542 --> 00:16:21.982 And it got it wrong. s214 00:16:21.982 --> 00:16:30.542 The naive agent got it wrong, whereas the semantic the intelligent agent built with the semantic play with the intelligence plane got it right. s215 00:16:30.542 --> 00:16:36.182 And now that margin of error is just five 5%. s216 00:16:36.182 --> 00:16:42.702 But when you're working with the finance sector in any mission critical workload, that becomes very critical. s217 00:16:42.702 --> 00:16:48.542 So getting your semantic layer built out, get your ontology layer built out, it becomes very important. s218 00:16:48.542 --> 00:16:57.142 I'm going to you're going to get access to this demo and I'm going to go back to the slides very quickly. s219 00:16:57.542 --> 00:17:07.342 So there's a bunch of models that we're adding into the intelligence plane and the demo I showed you actually has all of these models and all of this architecture in place. s220 00:17:07.342 --> 00:17:09.182 So what should you your team be doing now? s221 00:17:09.182 --> 00:17:11.622 One, you treat memory as data. s222 00:17:11.622 --> 00:17:11.902 Right? s223 00:17:11.902 --> 00:17:14.102 Don't look at it as different things. s224 00:17:14.102 --> 00:17:16.742 Write down your entities and definition. s225 00:17:16.742 --> 00:17:19.382 You push entitlements into the data layer. s226 00:17:19.382 --> 00:17:25.382 Over Oracle, we have features that actually enable you to do this very easily and capture traces. s227 00:17:25.382 --> 00:17:28.102 So memory is is basically your moat. s228 00:17:28.102 --> 00:17:30.102 You need to own your own memory. s229 00:17:30.102 --> 00:17:35.902 So over Oracle, we have this reference architecture we saw before. s230 00:17:35.902 --> 00:17:39.022 But at Oracle, we have product features that map to it. s231 00:17:39.022 --> 00:17:44.582 So we give you the system components to build your own intelligence plane. s232 00:17:44.582 --> 00:17:55.742 So that means that you have all the tools you need to build something that actually learns and grow as you continue to execute build as you continue to build and execute your agent. s233 00:17:55.742 --> 00:17:57.382 I'm going to touch on a few ones. s234 00:17:57.382 --> 00:18:02.902 Memory, security, deep data security, private inference and ontology and semantics. s235 00:18:02.902 --> 00:18:07.862 So these are the key five areas I want to touch on very quickly. s236 00:18:08.022 --> 00:18:09.102 So agent memory. s237 00:18:09.102 --> 00:18:14.902 We've spoken about it, but now what we did at Oracle is we launched agent memory as a Python package. s238 00:18:14.902 --> 00:18:24.503 We put memory engineering into a box, which allows you to just not worry about all of the complexities of making agents remember, we just convert it into a Python package s239 00:18:24.503 --> 00:18:29.503 and it generates context card for you where you don't have to actually do it yourself. s240 00:18:29.503 --> 00:18:38.623 Then, one thing you have to actually have to start capturing is your memory traces because this become data for fine-tuning. s241 00:18:39.023 --> 00:18:47.543 So, right here we have a slide that just shows where memory engineering and context engineering meets and this is for you to take and go through when I give you the slides. s242 00:18:47.543 --> 00:18:50.303 I'm just going to skip through this very quickly. s243 00:18:50.303 --> 00:18:55.023 So, agent memory and the Oracle AI database is what we just covered in the demo. s244 00:18:55.023 --> 00:18:55.903 I skipped it. s245 00:18:55.903 --> 00:18:57.743 But, the next thing is deep data security. s246 00:18:57.743 --> 00:19:06.223 I spoke to a few folks yesterday and today and they were talking about agent identity and how they were using different solution providers for it. s247 00:19:06.223 --> 00:19:12.663 In agent in Oracle, one thing we did is we made security a first-class citizen of the database. s248 00:19:12.663 --> 00:19:25.303 Which means with a product called deep data security, you can take all of the context of your application and the data access and row access of your user and we propagate that into the data layer. s249 00:19:25.303 --> 00:19:35.903 Which means your agents can take on the identity of the user it's acting on behalf and access the same data that it's only allowed for is only that the user is only allowed to access. s250 00:19:35.903 --> 00:19:37.583 Fumbling my words there. s251 00:19:37.583 --> 00:19:38.223 Okay. s252 00:19:38.223 --> 00:19:46.103 So, all of this is all great including the demo I showed you it uses all of this infrastructure I spoke about and I showed you. s253 00:19:46.103 --> 00:19:48.383 And this is how deep data security works. s254 00:19:48.383 --> 00:19:49.023 That's right. s255 00:19:49.023 --> 00:19:51.783 We have the analyst, we have our identity provider. s256 00:19:51.783 --> 00:20:01.223 The identity provider provides all of the data access that the analyst is meant to have and we can propagate that into the agent and the agent inherits that as well. s257 00:20:01.223 --> 00:20:06.063 And that limits all of the data access for that particular agent. s258 00:20:06.063 --> 00:20:06.943 Okay. s259 00:20:06.943 --> 00:20:11.223 The next couple of slides I'm going to talk about private inference. s260 00:20:11.223 --> 00:20:11.503 Right? s261 00:20:11.503 --> 00:20:17.103 A lot of you work in finance companies where you want to actually have your inference on prem. s262 00:20:17.103 --> 00:20:26.785 So, we have the private AI service container which contains all the infrastructure you need to host your own your own embedding, your own LLM in your own LLM inference, s263 00:20:26.785 --> 00:20:28.345 and re-rankers. s264 00:20:28.345 --> 00:20:31.505 And this is all around owning your own intelligence plane. s265 00:20:31.505 --> 00:20:33.185 So, giving you the components. s266 00:20:33.185 --> 00:20:39.385 But, if you want to go to the cloud, we have Oracle Cloud where we host some of these models that you're hearing about. s267 00:20:39.385 --> 00:20:43.105 So, Grok models and some Open AI models as well. s268 00:20:43.105 --> 00:20:52.145 Either way, air-gapped um infrastructure or cloud infrastructure, you have the components within Oracle itself. s269 00:20:52.145 --> 00:20:52.465 All right. s270 00:20:52.465 --> 00:20:57.785 So, I have a second demo before I go off. s271 00:20:58.185 --> 00:21:03.465 So, this demo is an open-source tool that I built called Memories. s272 00:21:03.465 --> 00:21:10.065 And it's just to show you what how you can actually start to get into the intelligence plane domain. s273 00:21:10.065 --> 00:21:14.985 So, right now there are people working on agent memory, like you guys have mentioned. s274 00:21:14.985 --> 00:21:21.865 And within an agent memory system, there are other ways you can start to bring in models to improve the system itself. s275 00:21:21.865 --> 00:21:28.145 So, decision model is one of it, and this one makes ex- uses example of three different models. s276 00:21:28.145 --> 00:21:31.305 A decision model, an embedding model, and a re-ranker. s277 00:21:31.305 --> 00:21:35.185 And within my agent memory system, I typically have these models. s278 00:21:35.185 --> 00:21:44.745 But, with the intelligence plane or with an understanding that I need to own my own memory, and I own my own memory, I can actually start to fine-tune some of these open-source models. s279 00:21:44.745 --> 00:21:46.585 And that's what we're doing here. s280 00:21:46.585 --> 00:21:50.785 So, over here in memory routing, I'm just going to focus on memory routing for a second. s281 00:21:50.785 --> 00:22:00.985 And what memory routing means is we get some output from the context layer, or we get some output from our LLM, and what we're able to do is route it to the right store. s282 00:22:00.985 --> 00:22:06.305 So, whatever comes out, I can route it into conversation memory or workflow memory. s283 00:22:06.305 --> 00:22:13.745 Normally, you would do this using some code, but with decision models, this becomes very easy and very cheap. s284 00:22:13.745 --> 00:22:17.985 So, right here we have some fine-tuned Oh, skipped it. s285 00:22:17.985 --> 00:22:19.665 Let's go back. s286 00:22:19.665 --> 00:22:20.265 All right. s287 00:22:20.265 --> 00:22:26.786 Right here we have some fine-tuned models for memory routing, and we can see how it was trained. s288 00:22:26.786 --> 00:22:32.026 So we fine tune the equine model and we use it and we fine tune it into a decision model. s289 00:22:32.026 --> 00:22:44.026 The data that we use to fine tune is all here and obtained from the conversation I was have been with the agent and I was able to classify what type of memory it was. s290 00:22:44.026 --> 00:22:46.946 And this allows you to do memory routing very easily. s291 00:22:46.946 --> 00:22:55.906 So once you get any information coming in, the different types of memory you want to route it effectively and cheaply, you can do that with decision models. s292 00:22:55.906 --> 00:23:04.666 So memory routing, embeddings, and re-ranking are some of the lowest hanging fruits you guys can actually start to work with today. s293 00:23:04.666 --> 00:23:11.706 Once you understand your intelligence plane, you're capturing your memory traces, you're formatting them effectively, and you're retraining the models. s294 00:23:11.706 --> 00:23:18.426 And you get the benefit with improvement in quality and in some cases reduction in latency as well. s295 00:23:18.426 --> 00:23:20.626 Okay, so we're going to go back to the slide again. s296 00:23:20.626 --> 00:23:23.866 We're going back and forth a few times, which is good. s297 00:23:23.866 --> 00:23:27.226 Um and we're just going to wrap up. s298 00:23:27.226 --> 00:23:31.226 So there are different ways we're helping developers build agents in in Oracle. s299 00:23:31.226 --> 00:23:33.666 And one of it is private agent factory. s300 00:23:33.666 --> 00:23:43.266 And this is a no code platform we can just drag and drop and build your agents and have the whole control plane at least within the private agent factory. s301 00:23:43.266 --> 00:23:51.066 And the private agent factory also has the same agent memory system that we've we've been building across all our product surfaces. s302 00:23:51.066 --> 00:24:00.146 So again, the private agent factory is built on top of the Oracle AI database, so you get all of the memory core functionalities I've been talking about. s303 00:24:00.146 --> 00:24:10.226 I already showed the ontology and semantic layer the effect it has on the accuracy that you get within agents that have and don't have ontology and semantic layer. s304 00:24:10.226 --> 00:24:16.866 But again, Oracle gives you all the infrastructure and components to build all of this. s305 00:24:16.866 --> 00:24:21.386 So you can ask the same question with and without meaning, right? s306 00:24:21.386 --> 00:24:22.546 So we've already done the demo. s307 00:24:22.546 --> 00:24:32.908 Again, I skipped I'm kind of like going around the slides and the demos, but you saw what we had with ontology layers and without ontology layers and the accuracy improvements s308 00:24:32.908 --> 00:24:41.347 can be very minimal, but in mission-critical workload, they can mean getting a 100 million fine for handling fraudulent transactions or not. s309 00:24:41.347 --> 00:24:44.987 Okay, so we touched on continual learning a bit. s310 00:24:44.987 --> 00:24:47.427 What we saw in the demo was continual learning. s311 00:24:47.427 --> 00:24:56.987 And which is there different ways you can look at continual learning in token space and in weight space and also in in latent space as well if you're going into the prompts caching. s312 00:24:56.987 --> 00:24:58.028 So we've just gone over this. s313 00:24:58.028 --> 00:25:00.827 I'm not going to rehash this again. s314 00:25:00.827 --> 00:25:11.788 But one of the key thing and the key takeaway here is is developers and engineers of these systems, we need to actually own our own memory, engineer the capturing of traces, s315 00:25:11.788 --> 00:25:17.148 and actually build it into a system that allows for continual learning to be possible. s316 00:25:17.148 --> 00:25:19.067 So I'm going to end right here. s317 00:25:19.067 --> 00:25:28.388 Um I think I'm going over time and I'm going to give you guys the QR code to the demo I just showed. s318 00:25:28.388 --> 00:25:38.187 So if you just scan this, but the best thing is you can come meet me at the Oracle booth and you will get this QR code again along with other technical assets and we can have a conversation. s319 00:25:38.187 --> 00:25:40.587 Thank you for your time.