Adi Kuruganti on Automation Anywhere’s AI Vision for the Autonomous Enterprise

Hey everyone, welcome back to our coverage here at Techstrong TV. We're in Dallas for the Automation Anywhere Imagine Conference. We've been talking to some of the attendees here, we've been talking to some of the Automation Anywhere folks, and we're going to continue our conversation today.

I want to introduce you to Adi Koraganti. I hope I said that right. Yeah, you did.

Adi, welcome. I don't know how many, probably very few of you would remember of our coverage last year, but it is available on Techstrong TV. We started this conversation with Adi about a year ago.

Adi, first of all, welcome back, it's good to see you. Thank you for having me. My pleasure.

So I'm not going to steal your thunder. Why don't you share with the audience a little bit about you and your position and kind of how you got here? Great.

So Adi Koraganti, as you mentioned, chief AI and development officer. I run products, technology, our AI teams, so basically responsible for the entire product and technology roadmap, innovation, and also making sure that innovation sticks in terms of customer outcomes. I've been here about four and a half years.

com, building a lot of their core SFA and then CRM offerings. Ran a business there towards the last four years, spent a long time there. Mm-hmm.

And so now, truly building and deploying offerings like what we'd call HD process automation and ensuring that during Imagine that we highlight all the incredible innovation to our customers. So Adi, right, the Enterprise Cloud is an interesting thing, but you said there were a couple of other- Yeah ... product announcements that you really like.

I'm really excited about our new solutions that we've created for autonomous service desk, autonomous IT, things like AIOps and FinOps, as well as autonomous finance. Because these are, think of it as driving outcomes for our customers. That way, there's obviously a technology part of it, and our developers, our more technical-oriented customers are really excited about that.

But then there are outcomes that line of business buyers really care about. And this essentially takes all what we're learning from existing customer deployments and creating those best practices and blueprints for customers to accelerate their time to market. So I'm really excited about that as well.

So Enterprise Cloud has their solutions, and the third piece, because I like to think in terms of three, is our context intelligence graph. One of the ideas we've been thinking about is, and working with customers deploying more and more of these agents, is how do we improve accuracy and reliability? And the way we do, we have our internal benchmarking tools, we do industry-wide benchmarks.

And what we realized is these agents, beyond data, they also need transactional history of what they did in the past, which is what we call the context, not only of the agent, but also of the process. So we're introducing a new product called Context Intelligence Graph. The entire goal is to drive higher accuracy and reliability of these agent .

It's interesting what you're mentioning, though. They all build on sort of Automation Anywhere's base. Mm-hmm.

These are the kinds of solutions you were offering before there was necessarily AI and agents and all that. But now, excuse me, now with this increased functionality and capability, it takes it to another level. You mentioned something, Adi, that I wanted to dig in on, and that is, to put it bluntly, gen AI and chatbots are so 2025.

Yeah. Right? Here we are in 2026, everybody's talking about agents, and it's this agentic era.

When we were talking last year about chatbots and that generative AI, when we look at what Automation Anywhere has been building around BPA and RPA for 20 years now. Yeah. Last year, I think it was APA was- That's still the category name.

Agentic process automation is the broader category. But what's changed now? I think it's the level of agency.

If you think about Copilots and others, they were essentially assistants. They're not agents. They're assistants where there's human interaction, essentially looking at a RAG or creating a RAG database, retrieval-augmented generation for those who don't know what that means.

But essentially, providing some context so you can get information back. And that was the kind of phase one of gen one of... I guess gen one is models and gen two was the assistants.

Understood. But with agents, it's not only about creating information, it's about taking action. And we launched AI agents about, I want to say 14 months ago.

And we are already seeing now over six million AI agent executions on our platform. Has grown specifically in the last six months. Wow.

And because when you combine query or search, or knowledge retrieval with action, then you're getting work done. Yeah. And in our world, as a process automation company, customers are using our platform to get work done, whether it is in finance or service or whatever the industry might, the function might be.

So that's definitely driven a whole new level of usage. Obviously, that also uses models and other pieces of it. And that's kind of a using the term agent a lot- Mm-hmm ...

but essentially the levels of agency, right? " So essentially mimicking a human, but letting the agent do it. And then you also have human in the loop for the human to, or the employee to confirm before you transact.

So that is what goal-based agents what we launched about 14 months ago. Now with Enterprise Crawl and others, you can get further agency, higher agency, which basically says you just give it a goal, some security parameters in terms of credential access and governance, observability. You can give it access to some tools, but frankly, these agents and this new class of agents can create their own tools.

Yeah. So if they need to access a website to get information, they can use computer use to access that website. You don't need to tell it to access that website.

No, it just spawns. Just spawns, and that's where these higher agency agents, they're the future and they're exciting, but there's also a governance aspect of it, which is why we've kind of built Enterprise Crawl to have that right governance angle to it. I agree with you.

It's exactly what you described, and I think that scares a lot of people. Yes. Because fundamentally, it's not even a question of the functionality.

Yeah. These agents are very functioning. That's right.

They have great capability. It's a question of trust. And when we talk about- And control ...

right. Trust and control. That's right.

So we talk about, we put nice words on it, like governance. That's right. But really, we're talking about trust and control.

Yeah. And that's funny because in my mind, trust and control is a sliding scale. That's right.

As we get more confidence, as we've got more success under our belt, I'll trust it a little more- Yeah ... and maybe give up a little bit more control. And your risk appetite.

Yeah. So there's certain industries you might trust and you might control, want to have the controls, but you might not have the risk appetite. So I definitely agree there is, you need to roll out those initial use cases with agents and deterministic process.

That's why our core belief is why we call it agentic process automation, is you're not just throwing away process automation. It's always going to be a combination of deterministic, cognitive, probabilistic, and human decisioning, all in a unified system, which is what we call agentic process automation. It's the 80/20 rule, 80% is still deterministic plus human, 20% is probabilistic.

Now, how, whether that 20% goes 40% or 60% is a risk appetite. And I think for highly regulated industries, I think it'll take a little bit more time to move beyond 20%. For some of the others, like tech, high tech and some of the others, and specific use cases that, I don't know, let's say in finance or even service, it could be even higher.

Mm-hmm. Right? It could be 40, 50% probabilistic.

But I think the risk appetite as well as improving the model and showing certain outcomes, both are really important. I don't disagree with you at all. I think that's dead on.

You're talking about risk, we're talking about trust, control, all of that gets tied into this observability- Sure ... issue, right? We've got to be able to see, we got to have clear path into what's happening, how it's happening, what's going on.

When we talk about observa... Now observability's a loaded word. It's been, yeah.

Right? It means a lot of different things to different people. But when we talk about observability for systems, where reasoning itself becomes part of that- Yes ...

runtime behavior- Right ... well, what does that mean for our audience out here, right? So one of the ways you build trust, and frankly control, is through observing what's happening.

Yeah. Right? Again, the lens here is customers automating these processes.

They're mission-critical, they're touching their financial systems, their healthcare systems. It's not just another chatbot trying to get information. So for us, observability is a couple of different levels, right?

One is what is the process doing? What systems is it touching? And then when there are times that the process is stuck, how can you unstuck it?

Right. Namely, the API call might be slow, or it's calling a system that is down. Can it retry on its own without actually any human involvement?

So there's one piece of that, the troubleshooting aspect. Sure. Then when you're looking about agents, okay, well, what's the plan?

What's the goal? What's the planning it's going through to get to achieve that goal? And what tools is it actually using?

Because it might have a whole set of tools it can use, but it's using a specific set of tools because of what it's decided. Mm-hmm. And then what is it actually transacting?

Right? And in many cases, not only do we show that information, but in some cases, you might say that against a benchmark of what, because we have this ability for customers to do agent benchmarking at design time, to just test it, because you have to test it before you deploy it. So for that, what we also do at runtime is we check for any drift.

Mm-hmm. So let's say you're observing that at design time, it did a certain set of steps, but in production, it was actually doing a very different set of steps. We need the admin to be aware of it because maybe these are probabilistic systems.

Sure. So maybe it's doing something different. It may not be wrong.

No, no. At least they need full... kind of visibility into the fact that it's doing something different.

All that I'd say is observability, but observability is not the only thing. It's observability plus security, plus governance. All three go hand in hand.

Right. So security's my background, right? Mm-hmm.

I spent 30 years. What you just described is exactly it. Some people say you can't apply the security without observability- Mm-hmm ...

right? Because otherwise you're just flying blind. Yeah.

And then governance, these all go- Yeah ... hand in hand here. Adi, I want to move along a little bit and hit another area, and that is around architecture- Mm-hmm ...

agent architecture. We were having a conversation in the room earlier, ServiceNow, companies like ServiceNow, they want to be the manager of your agents- Yeah ... or Salesforce wants to be the manager of your agents, or AWS.

Everybody wants you to use their architecture. Yeah. Right?

By the same token, everyone seems to feel that you're going to build your own agents. You're not necessarily going to use my agents. I was talking to a company last week where the CEO said he expects each employee to have 100 agents, which is, when you think- That's a lot ...

about it, that's a lot of agents. That's right. How do you reconcile that?

Is that a future that you think is possible, or? I think the fact that, I'd say a second question, the fact that you have lots of different agents for driving lots of different, trying to achieve different goals, definitely is a future. And frankly, if you think about coding agents- Mm-hmm ...

there's no one coding agent. You can have multiple agents building out a product or feature, right? My developers, when we build products, can have multiple agents doing that- Mm-hmm ...

versus one, right? So that's already happening in certain areas. To your question on, I want to kind of the bigger question around, every vendor talking about we are the platform of choice.

I'll tell you, I don't want to talk about other vendors, I want to talk about what Automation Anywhere does. So our entire lens here is because we are automating mission critical, that typically span different applications and systems. Some of those application systems could be in the cloud- Mm-hmm ...

some are going to be within the firewall, some are going to be air gapped, namely no internet access. We have those kinds of customers. But processes have to orchestrate across all of that.

And some customers will be okay, we have 70% of our customers on our public cloud, but 30% of some of the largest customers are in private cloud or in completely air gap on-prem environment. So it's kind of that hybrid distributed architecture that is essential to automate mission-critical long-running processes. If we look at some of the other vendors you're talking about, they either are very much cloud centric, typically their cloud.

If you think about Salesforce, it's all Salesforce's cloud. They don't have a concept of on-prem. ServiceNow has cloud and single-tenant cloud, their environment.

They don't deploy within the customer's environment. And AWS, obviously great partner, we partner with them- Sure ... but again, it's very much a cloud standpoint.

And so while we partner with all of them, our perspective is to truly orchestrate these processes and agents as part of these processes, you need to basically have agents running close to where the data actually is, not try to pull the data out of these air-gapped environments into the cloud so that agents can use it. Nobody's going to do that. That's not secure- Nope ...

and you don't have any controls and observability, all that stuff- Observability, control ... all that goes out the door, right? So our goal is you have a centralized control plane, which is how we've built it, centralized control plane where the agents and the tasks run distributed, closest to where the data is.

And it's all built on our cloud native architecture. So microservices based cloud native. Mm-hmm.

So I think that'll be the basis for everything, but I think we are truly positioned to run these more mission critical complex scenarios. And that's why we think the governance and observability for those mission critical processes, we are the best positioned to that. Others have a snapshot of it.

They cannot have the full snapshot because they don't actually run in those environments. Absolutely. You know, it's funny, I'm listening to you talk about the stack here.

It sounds an awful lot like the cloud native stack, right? That's correct. It is a cloud native, Mike, yes.

And this is something I think we're seeing, that the cloud native stack becomes the AI stack. Adi, we're almost out of time. I wanted to skip ahead here to the big picture.

In this headlong rush, and make no mistake, I've never seen the technology markets this foamy. Irrational exuberance- That's right ... as someone once said, right?

What do you think a lot of execs are underestimating or misunderstanding, and it could come back and hurt them? So at first stage, from a personal standpoint, it's the most exciting time to be in tech- No doubt about it ... I mean, I'm sure the internet era was like this, but this is even better, I think.

I was in the internet era. This is much more- Much, much better. Yeah.

All right. So I think couple of things. One is there are certain execs, like they're top-down mandates, go use agents.

Mm-hmm. And there's a run to use, whether it's, we see a lot of rebranding of copilots as now agents, and assistants as our agents. There's a lot of noise and actually confusion- Mm-hmm ...

that createsAnd frankly, a lot of those mandates don't go too far because there's no outcome behind it. " Yep. So that's something that we saw a lot in the past, especially the last couple of years, in the exuberance of first models, agents, and now GenAI and now agents.

But more and more, what I'm seeing is customers asking the question of what specific outcomes can we drive, and let's start with the outcomes, and then try to achieve those outcomes. So those are the successful customers. And whether you're using Automation Anywhere with AWS, and GCP, or NVIDIA, it doesn't really matter.

It's like, what outcome do you want to drive? And that's, I think, the big shift that's happening. Yes.

And the final thing I'll say is, there is definitely the concept of risk. Yeah. And I think vendors or customers are looking more, "Okay, how do I manage risk as well?

I want to move fast. " And that's why, as a company, we are not only focusing on the innovation, but we're also focusing on making sure we have the right governance structure, the observability, all the responsible use so that customers feel safe. Yeah.

Trust. Yeah. Trust.

It comes still back to that word trust. It's funny, one of the things that I'm seeing, I talked to a lot of people who've been to so many conferences in the last month or two, it's around that trust thing in that first we would just use it. " That's right.

Right? And actually, an NVIDIA executive say, "You know what? For some uses, it's actually cheaper to use humans- That's right ...

" And I think some execs hear that, and it messes with their trust. Yeah. They're like, "Maybe I should go slow.

" Yeah. It's going to play out. Yeah.

Claude Code, for example, amazing, but now with the limits, the cost is increasing in terms of usage because it's a killer product. It is a killer product, but- But the cost is increasing ... so you were talking about the dot-com era, right?

I remember I started what was a hosting company and then an ASP. I sold and helped build an ASP, went public, and all that. This is exactly what happened.

You had so many people, they would just write a business plan with a dot-com at the end. They'd get funded. " Yeah.

But you know what? You don't make it up on volume. You're losing 20 cents every time you sell it.

And so there was that reconciliation. I think we're going to have a similar thing with AI. It's got to make economic sense.

Yes. I agree with you. Anyway.

Thank you. Adi, a pleasure. Pleasure.

Thank you. We're going to be back. We've got more coming here on Techstrong TV.

Stay tuned.