Innovation from Agentic AI to Low-Code w/ Raju Vegesna – Zoho

Video Interview for Zoho by Jason Bloomberg

With the help of agentic AI, Zoho is now positioning its platform as aligning with enterprise concerns – without diminishing its value proposition to the smaller customers who have always been its bread and butter. Jason Bloomberg of @Intellyx writing for @Zoho. Read the accompanying BrainBlog here: https://intellyx.com/2025/02/18/zoho-brings-agents-to-its-cross-portfolio-ai-infused-platform/

Watch the interview video below or visit our YouTube page at: https://youtu.be/3-_eYoQStAA

Full transcript of the interview:

Jason: Hi, I’m here today with Raju Vegesna at Zoho. My name is Jason Bloomberg, Managing Director of Intellyx. So, Raju, what’s what’s new at Zoho these days?

Raju: Well, a lot of good work going on across the board on the product side. Certainly, across industries, there’s a lot of AI work going on. With the industries and the market changing on a weekly basis, that is keeping us on our toes.

And that is what makes it so exciting. The AI certainly is good, but there’s a lot of other technologies moving forward as well.

Jason: Yeah, so AI, of course, is a very hot topic these days. Zoho has incorporated AI in its whole product suite for a while now, right? It’s been several years. It’s been a decade, yeah.

So what’s sort of the 2025 story for AI with Zoho?

Raju: In the last year, things have changed. Dramatically, particularly with the evolution of agentic and that is shifting the market and more importantly, we are investing a lot on that. So today we are launching our own platform for customers to build their own agents and all of that.

So agents are a big deal.

Jason: Yeah. So, so just to make sure everybody’s on the same page, tell us what an agent is. What is agentic AI?

Raju: If you look at an LLM you have a large language model, then generative AI, you ask a question and it gives you the answer. It doesn’t always have all the context and all the information it needs.

And it doesn’t have to be, it is not necessarily the latest information. Sometimes to get a piece of information, it requires a tool to go and get something and provide you that information. So like that, when you combine a large language model with, and give it some tools where it can update information, its own information and act on it, whether you feed it data, whether you feed it some tools and say, go check on it a few more with few more information.

And few additional details and then report back that makes, expands the usage of OpenLLM. And now it’s like an agent where it performs things and that is the agentic model. So, the way I look at it is in the technology landscape or to use an analogy, if there is a new ingredient, that new ingredient makes new recipes possible.

LLM is a new ingredient that has come in to the technology world. And now, Agentic brings in a new ingredient. Now we have two new ingredients working together. What kind of solutions are possible for the customer? And that is what is making the industry exciting, excited again.

Jason: So LLMs, or generative AI in general has issues with veracity, right, hallucinations, and also the public LLMs are based upon the public web, so there’s intellectual property issues and compliance issues and now we’re talking about agents, which go out and fetch things, and so I’m even more worried now that it’s going to give me the wrong answers or it’s going to go fetch information that is, that’s confidential, or for some other reason is inappropriate for it to get. So what is Zoho doing to help reduce hallucinations and also ensure that it’s not fetching the wrong information or unsafe information?

Raju: What it fetches from, where it fetches from, what it fetches, all of that is something that user controls. So you combine LLM and give it a set of tools.

Here’s a set of tools that you can use to fetch information. Now who defines that tool? End of the day, it’s the user. That defines a tool. It’s a company that defines those tools and the parameters for that particular tool. We obviously create a few tools ourselves and we have a wide scope of tools. But now it can only do what you ask it to do because the tool itself is restricted.

Now what information you provide you really have to have all the information restrictive in terms of permission structure and all of that. So, as long as a tool goes through the authentication, authorization and access control filter in there, then it is going to do what you ask it to do. So, it is essential to have those filters in place before you randomly go and point an LLM and a tool to a random location because you want to add that trust layer in between.

So, that is, that is what we do where we not only offer a set of tools. But a series of, of pretty much permission layer and access control layer that can, that can restrict a tool to go out of control.

Jason: Yes, I know privacy has always been very important to Zoho. And privacy is sort of at odds with a lot of the LLM and generative AI on the market.

So what is, you know, what is Zoho doing to ensure? That it’s sort of privacy mantra that’s been so, so important all along applies in this new world of generative

Raju: AI. One thing we make sure is whenever an LLM is trained, it is not trained on customer data broadly. Now, you need models that work on customer data.

But the scope of that model is restricted to that particular business, that particular customer alone. That’s a significant model. It’s like a like a SaaS where you want, you don’t want your customer data to mix. The same thing applies to an LLM or any of the models we develop. So, that is a, that is a critical layer.

On top of it, add additional, the AAA authorization access control, all of those in there. Then, that adds an additional layer of protection. And we are not in the business of taking customer information, selling it to, or, or generally selling it to third parties and pushing their data without the privacy control.

We let the user define what they wanted to. The default option within Zoho is to keep it private. The user has to explicitly say, and go through a few steps to say, I want to explicitly push my data to a third party LLM, including my private data, but the default is always no. So, it is an explicit permission that they have to give, including training the models and what not.

But, of course, we do not use Customer data to train our generic models.

Jason: Okay, that makes sense. So one of the important use cases for Gen AI these days is creating applications or code generation more generally. So what, what is Zoho doing in that area?

Raju: We have a very good low code, no code platform called Creator.

We have over 5 million applications created on the platform. Now we are using the LLM to make it easy for anyone to create applications. So you input details on what kind of application you want and say generate an application. It can generate the app for you. That’s one aspect.

But then generating an application is not just one, one and done, there are going to be incremental updates. Say you generated an application, it has a form, you want to develop that form. Add new fields to it and there are generative capabilities to that. You want to create a workflow to that, then there is generative AI code creation, workflow creation, all of that.

If you want to generate additional code, we have done that. So these are some of the trends we do for the low code, no code things. There are also other experiments that we are doing, still a little early, where you take a screenshot or an image and say, here is the mockup of the image that the app I want.

Then input that to the LLM, it automatically understands the concept of that application and tries to create an application for you, including the logic that you want. That is another parallel thread that, that’s going on. Then there are other contextual embeds across the board. Let’s say I want to grab some data from a particular database and pump it to my, my analytic system.

You can even use LLMs to generate the SQL queries for you. And embed them and execute them as well. So at various levels, we have done a lot of work in there, creating some custom functions using our own language called Deluge, to hosting them on our serverless system called Catalyst. So there is a series of work we are doing for developer productivity, and we think it can bump up developer productivity significantly.

Jason: So this is an interesting area of innovation in the industry, where you had the low code tools, which were, you know, visual development environments, drag and drop, or some other visual metaphor. But along came generative AI, and now you had prompt based application creation, where you create an English language or natural language prompt, and theoretically, it can build the application.

But the question is then, you know, are these two different ways of creating applications merging? This is low code, and Prompt based Gen AI, is it, are these two sides of the same coin, or are you going to have different people build different applications with the different technologies, or what do you see happening there?

Raju: It depends on the skill level. There are some people that are, you know, That don’t think themselves as technical people, but they have clarity on what exactly that they want to that’s where the prompt based thing will be a good jumpstart. After it reaches a certain stage, then they can pass it on to someone else, but at least that gives them that jumpstart.

We call them maybe citizen developers where anyone can now create an application without knowing what a form is, what a field is. You explain what you want and then it develops, it creates that application for you. And then there are incremental updates that can be done. Then there are semi advanced users, like call them like Excel macro creators.

They, they are a little more adventurous than. They’re a novice user. They can use LLM and the power of that to create advanced functionality. Historically, the way it’s been is what was otherwise not possible becomes possible with the new technology. That’s where the problem based engineering is. And what was otherwise difficult would become easy.

And that is what some of the technologies will do. I think we are, we are seeing the same transition again here.

Jason: So I guess one of the important points to take away from this is Zoho doesn’t provide a one size fits all solution, right, for different developers with different skill levels or different understanding of the application they want to build.

You have different tools, right? So you might use Gen AI, you might use a creator low code tool, you might use a pro code tool if, if that’s what the Particular person, is there an appropriate tool for that person? So and that is different from a lot of the other products in the market where they pick one, you know, one or the other, but don’t have the diversity of capabilities.

Raju: We have an entire matrix where there are tools for no code, tools for low code, and tools for pro code as well across the spectrum. And then there are tools that extend The, the ecosystem and extend the applications, integrate the applications and, and really go, go full on deep as well.

So if you draw a 2 by 3 by 3 matrix you can put all of our applications, no low and pro code applications in that matrix. And that is where we are, we are going. So we want to cover the entire spectrum. Because that is the reality of the customer base. When you have close to a million customers, you know, they’re, they’re all over the place.

And we have to have the right application suite and the developer tool for that, for that group.

Jason: Very good. So what else would you want to talk about today?

Raju: So it’s, it’s exciting time in terms of broader market, particularly in the context of AI, but the beauty of AI, what we notice is it’s being embedded across the board and business models are going to change, certainly, you know, the deployment models are going to change.

And so all of that brings the excitement back in the industry. So I’m certainly excited about what’s, what’s going to come ahead in the future. So yeah, it’s agentic is where the action is at this point on the business side. But there are also parallel tracks going on on the application development, developer productivity as well.

Jason: Very good. Well, that sounds like a good place to wrap up. So thank you very much, Raju. Really appreciate it.