Swapnil Bhartiya: Now, when it comes to AI, it’s very easy to get started. But moving enterprise AI from a shiny proof of concept to a secure production environment is incredibly difficult. Most projects as we have seen install, fail before they even launch. Now a lot of companies are trying to solve this problem, including Oracle, by bringing agent TKI directly into Oracle Fusion applications. Oracle is giving developers, and no code builders, a unified platform to create specialized AI agents that understand business workflows. And who else understand business better than Oracle to help them move from concept to product? And joining us today to unpack this new AI native builder experience is Kaushal Kurapati, GVP of Applications Development at Oracle, first of all, it’s great to have you on the show.

Kaushal Kurapati: Thank you Sopnil, thanks for having me.

Swapnil Bhartiya: It’s my pleasure. Let’s just get into it. Can you give us kind of a broad overview of this new AI native builder experience and how does it fit into Oracle’s overall AI strategy?

Kaushal Kurapati: Absolutely. So we are seeing positioning this, Oracle’s positioning this as a new category of enterprise application. Fusion agentic applications is what they’re called. They are different from your standard standalone agents, they’re different from your traditional enterprise application. We’ve built this from the ground up for natively based on agents so that these applications are outcome driven. They are fusion native systems that combine agent teams, use decision making agents, workflows and all the enterprise business context through connectors and so on, and provide for deterministic controls. So executing enterprise workflows in a deterministic manner and a predictable manner is very important. We know that most enterprise workflows are very predictable, very deterministic. The next step is often very clearly known. So we are providing an array of tools for deterministic execution, predictable execution and consistent execution because you don’t want the result changing for user to user or when you run it the same system the next time around. So we are providing for all that predictability, consistency and determinism along with the power of LLMs and the AI agentic reasoning and with built in sort of role based access control, security auditability, tracing and detailed sort of testing and debugging capabilities, which are enterprise class agent testing and debugging capabilities, so that you can have all the trust and the confidence to build and deploy these agentic applications within the fusion runtime. So we have not built these for any complex sort of stitching them around and essentially you don’t need to do extra work for doing all of that. We’ve built this such that you can deploy these in the runtime with a single click. And the builder experience that we are launching is designed for both no code and pro code developers. No code business users can build very simply, build a prototype very quickly and a pro code experience as well, with both paths producing governed fusion runtime artifacts. That’s the difference.

Swapnil Bhartiya: Now when it comes to AI, we are seeing a big shift that it is moving away from AI assistants or chatbots to the whole agentic AI where they are taking action on people’s behalf. Now if you look at at things like openclaw and all those things, those are some of very, very dangerous but very powerful technology. I am a woodworker, so only thing I have learned is never fear your tools, always respect them. And of course when it comes to AI you have to also worry about hallucination and mistakes it makes, which is almost always talk a bit about when it comes to enterprises. What does this mean, what does this shift mean for them? And how does Oracle look at this next phase of AI?

Kaushal Kurapati: Yeah, I mean I think what we’re doing here is a copilot or some assistant usually assists a person with an individual task. But a fusion agent application brings the intelligence, the whole experience, the orchestration with agents execution, enterprise controls all together towards solving a business objective, essentially. So in practical terms, it’s a complete business application. It’s not a chat interface added to an existing screen. It includes a command center type experience where users can monitor that objective, manage exceptions and review the priorities that the agents are bringing towards them by understanding the business context. So we don’t need to tell the agent anything. It really understands the context and the signals and prioritizes the work. Right. So what we are doing is we are advancing the state of the art in terms of an enterprise application, what agents can do. But this application can really solve business outcomes, right? So and really you can deploy with trust and confidence, which is where the biggest blocker was right? For enterprises so far they have not they’ve been shy in deploying these, they’ve been doing it for very specific point tasks. But now you can solve enterprise grade business outcomes with using these agentic applications and you can build some very easily. That’s the builder tools part of it. You can build all of these very easily and deploy them for very specific tasks. So it doesn’t mean that something that’s off the shelf, it’s not a one size fits all approach, right? You can deploy an agentic app that is built around the process in your department, in your Organization in your kind of microcosm and somebody else in an organization can deploy a slight variation of it which will meet their particular needs again. And all built on the same platform, very seamlessly built on the same platform, but contextualized and customized and personalized for your own sort of set of use cases.

Swapnil Bhartiya: You mentioned, of course, no code, low code, pro code. I will talk about no code. This is what we hear a lot is that first of all, I am not a developer, but thanks to cloud I have written a lot of applications that helped me in past. I would not even know how to actually explain to a developer what I want to do, you know what I mean? Because sometimes you get. But it does enable. But the problem that we are seeing is that a lot of time folks are creating application that had not been tested. They may have security because you got people with no experience to start doing things. So talk a bit about when you talk about no code, what kind of audience you are targeting within enterprises. And when you look at no code, low code, pro code, how does kind of a single platform, how do you envision these very different type of builders collaborating on the exact same application? Does that create more job, more work for admins or developers to actually go and wet and clean? Because then it’s taking them more time to clean up someone else’s code version is writing. So what I want to understand is it’s twofold. First of all, who is the target audience when it comes to no code and then how does it work across different teams within an organization?

Kaushal Kurapati: Yeah, very good question. I mean, I think what we are really targeting with this is different Personas, right? So you can have, for instance a business person, could be a product manager, could be a salesperson, even could be an executive who wants to have a particular kind of interaction. Could be a manager, could be anybody or marketer or a customer support person. And they can describe what they want to build in natural language. So we have something called an agent brain, which goes with the Agent Studio, which is our developer sort of interface or right IDE for building agents. And what we can do there is essentially allow a person to describe what they want to build. And the system goes off and says, these are the agents I have in the system. I can assemble this application for you very quickly and you can very quickly create a, let’s say a prototype, a working prototype of how it should be and the business case that you are interested in particularly, right? So that happens very quickly, seamlessly. Guardrails are applied behind the scenes, governance mechanisms are put in place and you can interactively using a chat interface, essentially interact with this agent brain and say this is what I want, this is what I don’t want, and so on and so forth. And it will add components to the application, delete components, whatever it is. Right. So you can very quickly assemble a prototype in matter of minutes now. But if you want to test it, we’ve given you tools like we’ve called it Metro within the agent Studio. You can test it in the interface itself and, and run a variety of tests to make sure that it is working as you intended it to be. So that’s for a, maybe even a low code or a no code kind of a Persona, citizen, developer, a business user of various kinds. Right now if you go to developers, they are more familiar with the tools like VS code and things like that. So you want to meet the developers where they are. So they are very familiar with that. They are used to controls like that. They’re increasingly using coding agents like codecs or cloud code and so on. So we’ve built a specialized AI studio skill that can be included within VS code, which can be incorporated as a plugin and essentially AI studio becomes part of VS code because it’s happening in the development environment that they’re comfortable with and you can still use natural language there. You can launch codecs from there for instance, and say I want to build this particular app. But you have far more granular control. You want to include maybe more controls for production grade sort of deployment and you want to run tests which are much more granular and maybe they’re source code controlled the regression tests that you want to run. So you can leverage the power of the development tools and the frameworks that you have built up over the years and use all of that to bring to bear into the same development environment that the developers are using anyway and build a production class system agentic app which is really talking to the same backend pretty much as what a no code developer and a business user was building on. So the platform behind the scenes is the same. You have the same governance mechanisms, you have the same testing harnesses and everything. But what you are doing is you’re now enable different Personas for different purposes and for different kind of outcomes. But ultimately the artifact is the same. You can generate a same runtime fusion agentic application. You can approach it from two different perspectives. And also a lot of these cases what is happening is that the roles are kind of getting very different, right? The, the roles within an organization are no longer the Roles that were there before because of these various tools that are available. And so you can, a developer is, can also use the agent brain for that matter, or someone like me who’s coded in the past can also feel comfortable. And I started coding in VS code because now I can use natural language even there and I can use all the power of the tools and so on. So the roles are also becoming sort of intermingled essentially as you’re talking about

Swapnil Bhartiya: support for tools like VS Code, Git and AI coding assistance such as cloud code. Why was it so important for Oracle to meet developers exactly where they’re already there and where they’re already working?

Kaushal Kurapati: Yeah, I mean it is again, I think to giving the power to developers where they need a lot more control, they need granular control, they, they need tools that they’re comfortable with. And VS code offers a lot of that. And there have been, for instance you might have done QA testing, regression tests that you’ve been stored in a source code repository, in a git repository that you might have had. So you can leverage all of that now because this is integrated. So it’s an integrated work suite which they feel comfortable with. And so we have enabled a big ecosystem in partners as well. We have invested a lot in a partner ecosystem. We have about 80,000 plus people trained on Fusion Agent Studio and certified on the Agent platform. And we’ve heard repeatedly that look, we are very comfortable with these coding tools and coding agents and so on that have become popular and can you allow us to kind of build agentic apps right from there? So we heard feedback in our partner, from our partners as well. So to enable our SI partners, to enable internal teams as well, a lot of Fusion developers obviously within Oracle and of course among customers as well to give them more granular control. And there’s this huge pent up demand for developers in various places within Oracle, within customers and within partners. We said it was appropriate to build up these set of tools and give that power to these developers. And of course a lot of people, like I said, the roles are changing and somebody who wanted to dust off their old VS code skills and wanted to get into that and feel that ability to sort of test the agent gaps, build them and shape the entire development process by themselves. We wanted to make sure that we accelerate development so it accelerates the time to market, it accelerates, lowers the barriers to get them into production for various companies in various ways. So this is again an effort to sort of do that and we’re seeing strong resonance from various Folks who’ve been starting to using this now when it comes to as in the beginning, I was saying that most of those pilots feel moving an AI product from a proof of concept to a real production workload is a massive hurdle. Of course your teams work with customers. What kind of lessons your teams have learned working with customers when they do try to make that leap from proof of concept to a product?

Kaushal Kurapati: Absolutely. What we have seen and learned from customers and partners in this journey is a make our development easy and hurdle free. Right. So that’s part of what we are announcing. That’s why as these builder tools and second is they want to be able to have the governance and the tools for governance and governance is a broad term here. I’ll expand on that one is deterministic execution. And we have launched specific innovations called policy models within AI Studio where it can take a natural language business policy like your entertainment policy or refund policy, your vendor procurement policy, or any contractual policies that you have, and turn that natural language policy into deterministic executable code, actual code that you can execute and it runs consistently and accurately every single time. So we heard from our customers and partners and we gave them, we’re giving them the tools for deterministic execution. That’s one big aspect of how we are differentiating ourselves and how customers and partners are able to get the confidence to put these into production. The third thing is manual oversight. Many companies have built up over the years a huge process of processes for approvals. So if somebody is a vice president, somebody is a senior vice president in a company, they have certain capabilities and authorities to approve certain things. I can approve a code of $50,000 or $100,000 or $10,000, depending on my level. So how do you provide for structured manual oversight but including agents? So we built a manual oversight, manual approval node and within our Agent Studio, which you can incorporate in your agent workflows so you can build all the enterprise processes that you’re used to into the agent agent gap, so that it pauses for when an agent needs an approval from a human being and it really waits for that person. A long running workflow is important. Maintaining context, maintaining that memory. So all of these governance and deterministic execution and manual oversight tools give you the confidence to deploy these into production with accuracy, with trust and so on. And I already spoke about testing and debugging tools which allow you to sort of investigate and interrogate these systems step by step, like in a debugger, classic debugger. You could stop workflow like that. You can stop workflow execution every single node. What is the input, what is the output, what are the LLM calls, what is the contextual variables so you can investigate all of that. So you have a lot of tools being given to developers and non developers so that you can build agentic applications and deploy them with confidence. So that’s what we are doing. We are building the harness and the tools and the capabilities for deterministic execution, but at the same time we’re giving them the power of agentic reasoning so that they get the best of the both worlds.

Swapnil Bhartiya: Let’s talk about how do things look like or what does success look like for a developer using this platform six months from now or before the way they were doing things? Are we talking about faster build, reduced operational complexity or entirely new business experiences that were not possible earlier?

Kaushal Kurapati: Yeah, absolutely. And I think the so multiple things there acceleration of development is one critical aspect of it and we are already seeing that how much our own teams have built up in the short period of time. The second is they are able to execute workflows in a deterministic and predictable and consistent manner now, which they were not able to do that before. And we’ve taken out the non determinism so to speak. But preserve the power of agentic reasoning in here and we’re able to preserve what enterprises have built up over a period of time and codified their business processes and all of that into natural language in documents. We’re taking that and producing deterministic code for agents to execute so it’ll run consistently, accurately and predictably every single time. We’ve incorporated manual processes so that wherever you need manual oversight you can incorporate that. We’ve built debugging, testing, harness systems so that you can choose the right model, optimization, choosing the right model, balancing cost, accuracy, latency and token efficiency, all of those are important. So we built test harnesses which will automatically optimize the model for the task at hand. So given all of these, you can do new things which were not possible before and you can do them with trust and confidence with the governance and auditability and tracing and everything built in. And of course you can do it with great amount of speed, so time to market as well.

Swapnil Bhartiya: Kaushal, thank you so much for joining me today and sharing these insights into Oracle’s agent TKI strategy. Thank you so much for your time today and I look forward to chat with you again. Thank you.

Kaushal Kurapati: Thank you. Swapnil, thanks for the opportunity for sharing this and for those who are watching, make sure to check out Oracle to learn more about this new AI native builder experience in fusion applications, and I look forward to having another great conversation with you soon. Thank you.