No Memory, No Harness: Why the Database Is the Last Line of Defense — Kay Malcolm, Oracle

AI Engineer · 21 min · 301 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:12Everyone, are we having fun?
00:15Oh, you've got to give me way more than that.
00:18So, let me tell you, um, my name is Kay Malcolm.
00:21I am a retired hiphop instructor.
00:24So, if I don't get more energy than that, we will start.
00:28We'll start.
00:29Are you having fun?
00:31[cheering]
00:32Okay.
00:32All right.
00:32So, here's what we're going to talk about today.
00:36Now, you guys have heard a lot about two letters.
00:39Does anyone want to guess what those two letters are that I'm going to talk about today?
00:46Data.
00:46That was pretty good.
00:47DB.
00:48I'm going to talk about AI, but I'm specifically going to talk about agent harnesses.
00:54But before I do that, I want to introduce you all to a few people.
00:57Is that okay?
00:59Yes or yes.
01:00Is that okay?
01:02I gave choices.
01:04Yes.
01:04Anyway.
01:05All right.
01:05Okay.
01:05All right.
01:06This is my team.
01:09I run an outbound database product management team at Oracle.
01:14I've been at Oracle a really long time, 20 years.
01:18Funny story, I started when I was 12.
01:20So, don't do the math and don't start start adding in your head.
01:23Um, and we've got a problem.
01:27That problem is I've got one group that does platform development and then I have another group that does content development for live labs, a platform that I wrote myself.
01:40So yeah, I'm an engineer but I'm kind of a developer poser too.
01:44And then I've got another group who does QA and then I have another group who does my front-end development with AI.
01:53Here's what I found out as a leader.
01:57Because in the token maxing era of 2025, because you know, we're not token maxing anymore, right?
02:05We are responsible AI now.
02:08But in the token maxing era, the thing that I found out was while AI was making the individuals on my team faster, there was another problem it was creating.
02:23It wasn't making my team more productive.
02:28And the reason was when one team from the Netherlands checked in code at my 4:00 a.m. in the morning because I've got half of my team that's in AMIA and I have half of my team who that are here in the United States.
02:43They checked in the code but they didn't check in their context from Codeex.
02:49We use Codeex at Oracle.
02:51So then when the US team woke up, they got the code but no information about the context.
03:01So we used AI to solve a problem that AI created.
03:07And here's what we did.
03:08Oh well, let me talk about this first.
03:10So some of the issues, the context, like I said, wasn't shared.
03:15GitHub wasn't tracking that.
03:18I had repositories that were diverging and I was asking the managers who work for me, what's happening to your teams?
03:26Why why are we not going faster?
03:29We're spending all of this money on tokens.
03:31We're spending all this money on AI, yet something is missing because we're still spending time doing testing and validation.
03:40So our netn net wasn't really wasn't really working for us because git records the code and not human intent.
03:49So it's a problem and even though code creation was no longer our problem still had a bottleneck.
04:03We needed a collaboration layer.
04:06Now, I do have members of my team in the audience, so don't judge me, and you know who you are.
04:14I'm not saying that you all didn't collaborate.
04:18But now, we've got a new team member, and that new team member is AI.
04:23So, we needed to figure out how to track our progress in our next steps.
04:29how to rationalize decisions that the agent was making.
04:39We needed to figure out how to resolve questions in conflicts.
04:45Okay, hold my problem.
04:49Will you all hold my problem for me right here?
04:51We're going to just tuck that in a little box.
04:54Let me define what a enterprise agent actually is.
04:59Now, most people think that an enterprise agent is the model and workflow.
05:05How many people agree with me?
05:08Man, this tough crowd you.
05:11Okay, one person.
05:12Okay, the rest of you think it's a little bit more.
05:15Okay, let's see what could it be that a real enterprise agent has tools.
05:26Tools are how it does things.
05:30Context.
05:33The context.
05:34That's a context window.
05:35That's what's in the actual prompt.
05:39Memory.
05:40Huh?
05:43And if you're thinking, "But wait, K, memory, you just said that the model is kind of like the brain of the operation."
05:51Hold tight.
05:52We're going to talk a little bit more about memory retrieval because you don't want to get everything back.
05:58So that's being able to to retrieve the right information back.
06:04And then I know that there are a lot of developers here and you all don't care about security.
06:11I care about security because I work for the most secure database company and I used to work for um agency that has no name.
06:22But guard rails is also important.
06:25This is the harness.
06:28I speak in analogies and I and I speak in stories because if I tell you this and Marvel, you know exactly what I'm talking about.
06:38So the agent, think of it as the model, little brain floating a in a glass jar plus this harness.
06:49This harness is the body.
06:52So it's how the agent can actually do things and get things done.
06:58That memory, that's the part of the central nervous system.
07:03And you remember the central nervous system connects the brain to the rest of the body, legs, arms.
07:10That's the part of the central nervous system that carries context.
07:16So you remember my problem with Git.
07:20What I needed was memory.
07:22Okay, so there are a number of memory types.
07:26I chose five, the five most common ones that people talk about and these are the ones that I want you to remember.
07:32The first one is short-term memory.
07:34That's the session, right?
07:36And so if you're storing memory of an AI uh process, that is the short-term memory is if you're with chat, cloud code, right?
07:47Codeex, pick your poison.
07:49The long-term memory is what persists across sessions.
07:55Episodic memory.
07:56Hm.
07:58What happened the last time I interacted with fill in the blank?
08:06That's your episodic memory.
08:09Procedural memory tools, steps that were taken.
08:15And then finally, semantic memory.
08:17And semantic memory because we're talking enterprise agents.
08:21We're not talking the agent that I built, Sasha Fierce, because remember I told you guys that I was a I'm a a dancer.
08:28So, of course, my my chief of staff is going to be called Sasha Fierce because that was Beyonce.
08:34Any Beyonce fans?
08:37Okay, I'm sorry.
08:38All right, we got to focus.
08:40Okay, so these are the memory types.
08:43Now, when you're defining this real enterprise agent in this memory, there's something you need to consider where to store it.
08:53And so, I'm going to tell you guys a story.
08:55But when I tell you the story, you have to promise me that you're not going to judge me.
09:00Do you promise?
09:06Do you promise
09:08you're not recording me, right?
09:10Because this doesn't paint me in a good light.
09:12Okay.
09:12All right.
09:13The world of data was one simple.
09:14I've been at Oracle a long time, but I came from a customer.
09:18That customer's name was Southern Company.
09:19It was a power company.
09:20I'm based out of Atlanta.
09:22And I was hired at Southern Company because I was a rockar performance tuner.
09:28You had a SQL query.
09:29I mean, I'm dating myself, but whatever.
09:31You had a SQL query.
09:32I knew all of the innit.org parameters.
09:34Even the ones when you called support and they said, "Don't remember these.
09:37Don't write them down."
09:38I wrote them down in my little notebook.
09:40I could tune a query within one inch of its life.
09:44Then one of you came to my desk because I mean the world the world was rows and columns.
09:51It was a great time back in my Albundy days um and said hey I need to store data unstructured.
10:01Why I need to do that?
10:04And so me being K the the diligent DBA, I was like, "Let me figure it out and get back to you."
10:14Did I get back to him?
10:16I didn't get back to him.
10:18Now, the thing you have to know about Southern Company was for every database system that a DBA managed, I had to attend two meetings.
10:26Today, when I hear Sarbain Oxley, I still throw up a little bit in the back of my throat.
10:31So I had to attend a security meeting and a patching meeting every week.
10:36Never failed.
10:37Now because this developer installed a database that was specialized for unstructured.
10:45Okay, there are really smart people in the room.
10:47How many meetings am I going to now?
10:54Four.
10:55Okay, I'm a little annoyed, but I'm like, okay, we can we can we can do this.
10:59Then they said, "Okay, since you're such a good tuner, I need you to figure out this relationship."
11:06Now, the way that Southern Company worked, there was this people could die application, um, and it was a it was like a Nokia phone that people who were climbing the towers, right?
11:18So, you guys have have been in a storm and the power goes out, right?
11:23And then you're pretty sure that within maybe an hour or two the power will go on.
11:27Well, that system that would tell the people who were climbing those trees and risking their lives to turn the power back on sometimes would have false positives or false negatives.
11:39So, they wanted to look at all of the other um uh polls in the area to try to get away from the false positive or the false negative.
11:50And so I did that in a SQL query and it was it was amazing.
11:54It was a five nested union all statement.
11:58It was some of my best work.
11:59Now it might have taken like 20 minutes to work but it was like a predecessor to graph.
12:06Yeah, they they install Neo4j.
12:10So now how many meetings am I going to?
12:13Six.
12:14That's a problem.
12:17So I um Oh, let me I got ahead of myself.
12:19So you know what I did?
12:20I quit.
12:23I left and I came to Oracle because I was like this is a problem and maybe I can go to Oracle to help solve it.
12:27So then Joe Mundy called me and he said hey um we are installing Reddus.
12:33Oracle is late to the game.
12:35We've got a vector database.
12:38Okay.
12:40But here's your problem, Joe.
12:45Agents now need access to all of this data.
12:49So if data is in an Oracle database, if then it's also in an unstructured JSON database, if it's in a graph database and it's in a vector database, where is your single source of the truth?
13:03The agent has to figure that out.
13:05Sometimes it'll get it right.
13:08Most times it'll get it wrong and it's going to burn up a whole bunch of tokens.
13:12And so now if you want to store your memory somewhere, you can store it in a file system.
13:20You can store it in clawed or chatgpt because we all know about the memory.md file.
13:29But that's going to be a problem.
13:31Now I want to illustrate this.
13:32I need four volunteers.
13:33I can see you raise your hand.
13:35One, two.
13:37Okay, I can't.
13:38Maybe I can.
13:39Three.
13:40I need a fourth.
13:41Ah, fourth in the back.
13:43Okay.
13:43Fourth in the back.
13:44You are going to be our old reliable.
13:47You're going to be a relational database.
13:49Yes or yes.
13:50You got your So, you have your assignment.
13:52Okay.
13:53And then there was someone here.
13:54You're going to be my unstructured database.
13:57Okay.
13:58And then where was my other?
13:59Ah, very good.
14:00You're going to be my graph database.
14:02You good?
14:03Relationship guy.
14:03You look like a relationship guy.
14:05All right.
14:06Very good.
14:06Fourth.
14:06Where was my fourth?
14:09Was it you?
14:10Yes.
14:12Yeah.
14:12You are my vector database.
14:15Okay.
14:16Now, everybody be really, really quiet.
14:20For my four volunteers, I need you all.
14:24I'm going to say something to you.
14:27And I need you all to decide how you're going to store it and who's going to have the single source of the truth.
14:34You can't get up from your seats and you have to whisper because if you talk loud that's 5x the tokens for you.
14:40Yes.
14:41Okay.
14:42Are we ready?
14:43All right.
14:44The cow jumped over the moon.
14:49Go.
14:53Doesn't really work, does it?
14:55That's a problem.
14:56Okay, Oracle.
14:59And if you don't forget one, if you forget everything I say and you remember one thing, Oracle is not the Oracle that you think that is why I am here today.
15:12How many of you knew that Oracle could natively in the same table down to the same partition store JSON graph vector my vector friend over there my JSON friend
15:26spatial you want your memory to be immutable blockchain in the same database raise your hand yeah we have a marketing problem so any data type can be stored in a 26AI
15:44database any workload anywhere AWS GCP Azure OCI on prem choice and flexibility so now when we take this and we talk about the agent I want to be able to store my long-term and procedural memory in relational
16:01on in JSON I want to store my short-term and my long-term memory graph I want to store procedural because procedural that's how I figure out the relationships right the steps
16:11my episodic and semantic memory.
16:14I need to do some vector and then store it also as text.
16:18Now, if I have four different databases, you all saw they can't talk to each other.
16:25It's going to be a problem.
16:27And so, what I'm saying to you today is the Oracle AI database is the best place to store this agent memory that's going to power your harness.
16:38Remember your harness is your body
16:40[music]
16:40and that memory is your central nervous system.
16:44Okay, back to PY.
16:46So the problem that I had, we solved it with a memory broker named Py.
16:51We used agent memory.
16:54We got out of that automatic continuity.
16:58So with my team, they were able to share not just their code, but PY also kept track of the context.
17:07So if one context window was had procedural memory, episodic memory, information about the long-term memory, that was then shared with the other folks on the team.
17:20You could call them agents, if you will.
17:21They're just human agents shared across forks.
17:26The developers on the team remained in control while Polly was able to create the context, figure out which fork and branch it belonged to and which commit it belonged to.
17:41Now this is a very simplistic example but when you take this to the enterprise here's what happens.
17:49Memory is the thing that becomes non-negotiable in an agent's harness.
17:54Now these are three papers that that I read um on the airplane.
17:59This first one is from open AI and it's about its in-house data agent and the thing that it says is it is saying that its in-house data agent actually needs memory.
18:11Memory was crucially important to ensure that its agent was able to filter correctly instead of trying to string match.
18:20Harrison Chase said, "Your harness, your memory.
18:23And if you don't own your harness, you don't own your memory."
18:26Which is key.
18:28And then I'm sure you all are wondering, "Well, Claude has memory.
18:31Why can't I use that?"
18:32Well, it's kind of like file system memory and it works with one, but just like in my example, when you scale past one, and you're going to scale past one in the enterprise, it creates a problem.
18:45So Oracle has a Oracle agent memory pap package.
18:51PIP install Oracle agent memory.
18:54You get access to it.
18:56And this memory is this SDK that we have is the thing that will hold your live conversations, your memories, your facts and figure out what is worth keeping.
19:08So if we look at Py now, Kevin can share his context with Py, our memory broker.
19:16We use the Oracle agent memory SDK.
19:19It's stored in an Oracle autonomous database.
19:23We can use the LLM of our choice or we can use a local model through the Oracle private AI services container.
19:31And then Linda, who's actually sitting right here, can interact and work with Kevin, no issues.
19:38So yes, AI makes individuals faster.
19:42Shared memory on an Oracle AI database makes teams faster.
19:47So I don't want you all to compromise.
19:49In the age of AI, what 26AI does is you can choose and pick what's best for agent memory file system stored in a database file system or in the database.
20:03If you need to do data modeling, you've got JSON, you've got relational.
20:07We've got choice.
20:10Okay, I've got some goodies for you.
20:12The Oracle AI developer hub.
20:14That's where you can guys you guys can get coding materials, the applications, what I talked about today.
20:19Livelabs.oracle.com.
20:21If you've done any of our workshops today, that happens to be something that I wrote myself about six years ago and 40 million users um ago.
20:30Spend my OCI tenency money.
20:33kick the tires on any Oracle technology um for six hours, 12 hours, however long you need.
20:40Um and then I'm giving you all all a Mac Mini.
20:44No, I'm just kidding.
20:45I'm giving you an OCI Mini.
20:47So, I don't know if you knew, but there is an always free OCI.
20:50It is the most generous of any of the hyperscalers where you can get a free Oracle database, free compute, you can send 3,000 emails a month, 200 gig in storage,
21:02and if you click on that, you can get access to it.
21:05Or just search Google for Oracle Cloud, always free.
21:10Connect with me.
21:11If you build something, will you all message me and let me know?
21:15Yes or yes?
21:18Thank you.
21:34[music]