Rick Houlihan

So I have a long history in OSQL. For folks who know me, I'm a very big advocate of document database technology. So I was the worldwide technical leader for OSQL service at AWS. I moved to MongoDB. I was the field CTO for enterprise there where I built the developer relations program for strategic accounts. And I brought that back to Oracle recently because Oracle actually recently announced that they have adopted document technology and integrated that as a first-class citizen in their ecosystem. So that was very interesting to me. And in exploring the opportunity there, I realized that this converged database technology is really revolutionary versus the kind of polyglot persistence model of our competitors. So that's what brought me here.

So when you think about modern applications, AI application services, it's not just one database anymore, right? We have to operate across databases. You have vectors, you have graph, you have time series, there's spatial data, there's relational, there's document data. And RAG retrieval requires accessing multiple persistent stores. So if you start to persist data across multiple silos, what you do is you create an interconnect, which requires a lot of maintenance and a lot of overhead and a lot of compute infrastructure to kind of persist data across multiple stores, transact across multiple instances of data to give you the view that you need for modern applications. So this is where Oracle stepped in and said, you know, it's not about what database anymore. It's not about document versus relational. It's about being able to provide a converged persistence model for all of these data models where you can have a single execution framework. And now I don't have to ask questions across multiple silos within a stack. So the latency, the efficiency, the performance of the system is tremendous when you start to compare.

Efficiency and consistency. Right. And one of the things, one of the biggest problems we have with AI today is hallucinations. Right. And the way that you get AI to hallucinate with confidence is you give it inconsistent data. Right. So if you have different persistent stores and you have different views of data that might be related but inconsistent with each other, this is the recipe for hallucinations. So this is where Oracle's converged database, the 26 AI platform, gives you a significant advantage because all of these various shapes of data, they exist within one execution framework, within one transaction. So the consistency model across all of these various reflections or projections of your data is uniform. There is no eventual consistency in the model. So therefore the queries you're getting are as accurate a picture of the data as you can get at any given time.

So it's not about normalization anymore. Normalization is a relational construct. Right. So normalized data persists in tables. Right. Homogenized tables that have relationships to each other, you know, orders, items, shipments, invoices, they're all related to each other. A document view might say, let's take the rows of those tables and embed those into a view that is consistent with what the application requires. So an order might contain an array of items. That's not really a normalized data structure. It's stored as a document. So now in the Oracle execution framework, that document exists as its own persistent store, a JSON table with a document attribute. You can also break that document up into multiple relational tables. I can create a document output from those relational tables using a duality view. So there's multiple ways to access the data. It's not that the data is all normalized. It's just that it is now connected. Right. So the execution framework is a single context, single memory space. There's no we don't have to traverse TCP IP boundaries. We don't have to execute requests across multiple APIs to get the data. All right. So it makes it just much more efficient.

It really is. It's like a global filing system. You put your data in there. Is it spatial time series? Is it graph? Is it document? Yeah, I'm sorry.

But yeah, I know it's these various shapes or projections of that one canonical form, right? The canonical form is normalized data, but that only has to exist as a logical overlay. Right. We don't have to persist data as normalized data stores, right? We can persist data in any shape as long as I have a single execution framework that allows me to access that data consistently across those shapes. Now, that's the power of the converged database.

you can think of it that way, right? It's a universal data store is really what we're talking

Right now, any Oracle customer, if you have Oracle technology in your shop, which is really all 500 of the Fortune 500. And you know, I would even say all 1000 of the Fortune 1000, right? Everybody has Oracle. You have access to this technology. The 26 AI is just an extension of the Oracle family of database services and it is actually runs on Oracle exit data. If you are an Oracle customer, then talk to your Oracle reps. We can get you up to speed on 26 AI.

Well, I, you know, I could say I don't know exactly what database is being used in the back end of chat TBT or whatnot. I know these things are all tied into various systems of record, right? Then then Oracle at 26 AI has its own MCP, you know, server as well. So you can use Oracle in the context of any, you know, with any LLM that's MCP compatible, right? So I think it's really I would treat it more as a source of data for those services rather than the persistent store. All right.

Oh, I'd have to say the evolution of the coding agents, right? I think we talked earlier that six months ago. I don't know if I really trusted these tools for doing anything other than maybe small scripts. I did use them for production code, but I checked every output. You know, recently I put a post up on LinkedIn where I talked about this, the trust deficit, right? This is the every abstraction over time survives because it provides more value than the trust deficit that it erodes. And this is goes back to every technology evolution, punch cards, compilers, cloud technologies, HTTP parsers, right? No one looks at compiler output anymore. We used to, you know, back in the late 60s, early 70s, that people would double check the compiler output to make sure the machine code was what they expected. And people would sometimes when you use compilers, you would be ridiculed for using them. You're going to let a program write your code. Well, today no one looks at compiler output, right? And this is the same thing that's happening with the Gentic AI. What I'm seeing is the reliability, the trust, the trustability of the output is significantly improved. And it's just like everything. If you don't, you have to know how to verify if it's good. And if you give the tool, the mechanism, the tests to verify that it what is correct, it can produce very, very, very quality output. So again, it's all about what you the input, spectra of an input.

And I think right now what it requires is, you know, that architect level skill set. You have to be able to and a lot of it is becoming for me, it's becoming less of reviewing the output and more about reviewing the input. Right. So I talk about reviewing the input. What I mean is the spec, the spec that we generate that is the input to the prompt. So oftentimes, that's I spend more time going back and forth with the AI tools, generating the product specification and the implementation plan. Once those things are generated, they're very detailed, usually multi-page documents, then I'll just tell the AI, okay, let's go according to this plan using tests, you know, test driven development standards. And I'll watch it. I'll watch the output as it goes and I'll check, test check it every now and then. But the reality is I'm getting more closer and closer to the point where once I have a well-defined specification and a well-defined implementation plan that these tools are becoming like compilers.

The improper use of the tools, right? I mean, this is what I think that people tend to, right now, what we have is a lot of people running around creating what they believe are products. And, you know, they're really kind of like prototypes, maybe demos that I might show, but they're not products because they're being built by people who don't understand what products are. So they don't even know what kind of specification to define as input. They're just giving it functional requirements and they're saying go. And I just was reading the other day, I think one of the people I follow on Twitter, Teri Radichelle, she warned some time ago about the security credentials that could be leaked from some of these tools. And I read the other day that like Claude Bot is exposed, you know, hundreds of thousands of instances that people haven't updated. Right. So even though the tool has been updated to prevent the credentials from being exposed, there's a hundred thousand instances out there. People installed it two months ago who haven't updated it since and all their stuff is flying in the breeze. And this is the thing that scares me is I don't think people understand how exposed they are when they plug these tools in. And I think that's caused that's going to create an enormous risk that we have yet to even realize.

So sure.

I think it's slowed things down from what I've noticed. I mean, I definitely don't think that the across the big tech across the technology sector across all sectors, I think we're hiring at a slower pace. I think as people start to kind of become familiar with these tools, they're force multipliers. So what you're seeing is people become much more productive, but they're not yet at the point where they can replace us. I don't believe that we're there yet. I think that, you know, and I don't know how far away are we. That's that's that's the scary part. I don't know. Are we a year away, five years away, 10 years away? But we're headed there. There's no doubt that these tools are on track to replace a significant portion of the current skill set in the workforce. What does that mean? I don't know. We face these things many times, right? I heard back in the day that so many times have I heard technology will replace us. The job force is going to be destroyed only to find out that what the technology actually did was open up new opportunities, open up new capabilities that people could take advantage of. And I think that's where we are. We're at an inflection point. Again, you need to embrace the tools. If you don't embrace the tools, you will age out in the 70s. We saw it fax machines and email and the 80 and through the 80s and the 90s. And I remember people saying, you know, email is an unreliable fax machine. Got to use that fax machine. You know, by 1992 93, those people are aging out of the workforce. Right. So this is the key to any technical technical Renaissance. Right. You have to be part of it.

I would say somewhere in between. Obviously, I'm excited in so many ways because I think this technology is going to unlock innovation and unlocking invention invention in ways that are going to help make society and human and the human experience better. I'm fearful of how these these technologies will be used in ways that are not beneficial. And I think that, you know, your experience in media is relevant here because I think media in many ways is being destroyed by AI and deep fake technology. So it's a double-edged sword. Right. This could be the savior or the destroyer of modern human life.