Hi everyone. my name is Marish Bal and I am going to talk about agent automation and how it's going to change, the world of business process operations. today. Excited about the topic and a pleasure to be in front of you. Let me tell you a little bit about, myself. I'm currently based in Texas, had an opportunity to work with, various firms, global system integrators, as well as startups. In India, Australia, and here in the us. like most of you, I started off as a developer. Probably I'm, I was not as good as you guys are. and switched over to consulting very quickly after my MBA and then, had, stints in account management. As a sales hunter, here in the US for, it's been about close to 12, 13 years that I've been, in sales and, and, business development. The current role that I have is probably a mix of everything that I've done in the past. I'm a GTM lead for one of the key generative AI agent tech, automation offerings at, AWS Excited to tell you. More about it and what, this, this technology is going to unfold in the sphere of business process operations. On a lighter note, I feel, it's time. From the point where we are, to a point where toasters start talking to you and potentially, start giving us relationship advice. but it's not too far too. so let's be prepared with that. now I. Going to a little serious problems. Right? what's happening in enterprises? I have a chance to work with lot of, customers and, companies that, that are large. And what I have seen over the past few years is that most enterprises have tons of applications, and some of those applications are productivity apps. Many are ERP apps. Off the shelf apps, custom apps, you name it, and they have it. unfortunately many of these applications don't talk to each other. And how do we solve that problem? we absolutely need to integrate these applications. So how do we do that? The solution is people. Let's throw people at the problem and let people act as a glue and let people run the applications and the processes that layer on top of that. But what happens because of this is the fact that, the people get frustrated. They are stuck in mundane, repetitive jobs, which they don't like. and they really don't have any time for process improvement. They don't have time to talk to their customers and, think about what more they can achieve for them. And as a result, everyone becomes an, less productive. So how do we solve that problem has been, one of the key issues that's been grappling the large enterprises all over the world now. There has been a solution. We have been thinking about automation, and when I'm talking about these, processes, I'm talking about the, the processes that enterprises run. So I'm talking about the white collar processes. I'm not talking about the industrial processes and industrial manufacturing. I'm talking about processes like. An invoice process. For example, I'm talking about sales order processing. I'm talking about claims processing in healthcare and in insurance. I'm talking about those kind of processes. Now, automation has always been a solution and there are various flavors of automation, When I say various flavors of automation, what I mean is there is robotic process automation. Many of you probably have heard of that technology. it's, it's a great technology if you want to solve a simple problem and you need some rules based engine to automate a business process, fantastic. but then RP has its challenges. It is fragile in nature. It doesn't lend well itself for. Complex processes where there is unstructured data requires a big infrastructure footprint. API integration is of course a domain of, developers when it comes to adopting API integration. One of the biggest challenges is the legacy apps don't even expose API. So how do you deal with that? document processing? All of us know that in the last few years. Document processing has increased and has unfolded like a beast. and what I mean by document pro, what I mean by document processing is processing of. PDFs to clean insights from those PDFs, from images that are circulating across, enterprises. And then there are, low-code, no-code technologies. Data integration has been there for a while. machine learning models have been existing and now, LMS are also being used. interestingly in back office, front office, and middle office processes, which is where most of the productivity is locked now because of all this, it's a challenge. How do you make it simple? And that's when you know, we have agents to the rescue. What are agents that, this is what, talk of the town is, everyone wants to talk about agents, but what are these agents. In the simplistic terms, and most of you have probably, dabbled in this space and you are trying to, work with agents. For me, I have a very simple example. This is a beautiful, graphic that I found on, vectorize.io, thanks to the creators. what this graphic shows in the simplest of terms is there is an LLM as a foundation of an agent. And think about LLM as a brain. there is, an orchestration, which means there is an input that you need to provide. And that input can come from a web app, can come from chat, it can come from a mobile app, wherever. And then there is the last layer where you need to provide tools for the agent to actually accomplish something. Just like how, you need a ax to cut the wood. For the lack of a better example here, in this case, you would need tools. And the tools in this case can be, a web portal. It can be an application, it can be a knowledge base, it can be some business logic that's embedded, can be a SaaS platform. It can even be user interaction. This is how the agents are structured. Now, it sounds fantastic on paper, but is it easy to deploy? that's a question mark. Again, you have, APIs, who's going to build those APIs? Are there common standards? No, vector dbs. Some of us may be, really ninjas at Vector db, but what about the others? the commoners. so it's a complex technology to implement, easy to understand, because all the magic is anyway done by the LLM reasoning. But then there are a lot of issues that we need to understand before we go about implementing the agent architecture. And the next thing is, are these agents sufficient enough? Can you just implement agents across the organization and. Let them, do their trick and automate very complex business processes that are, executed by hundreds of people. That's another question mark. So how do we take it to the next level? And that's where. Lamb, comes into the mix. What is lamb? Lamb is a large action model. what is a large action model? I try to keep it really simple, right? LLM is, the brain. and if I call LLM, the brain, the large action model is Braun and the whole, gist of a large action model is in its name that it takes actions, LLMs understand, understands and summarizes. another thing about lambs is that. Many of the lambs and a great example of that is computer use from anthropic operator from open ai. these are trained with specific integration with certain tools like browsers, to automate and execute certain tasks, based on natural language. So lamb absolutely sounds more powerful. But the problem is, if you tell the lamb to perform certain actions in a series of steps, like 70 or 80 steps, it may not be able to do it. A four or five step process like book, me, a nice Italian dinner in a restaurant in New York at 7:00 PM on, right before Thanksgiving, when the week is crazy. And then also book me a couple of tickets for a Broadway show. It can very well do it. It may also even ask you follow up questions and get clarifications. But when it comes to executing a complex business process, then it becomes challenging. And, then how do we do that? How do we even, crack that puzzle? Now, that is where, What we need to do is we need to think about orchestration. How do we execute processes with lambs, with LLMs, with document processing, with API, integration with, low code, no code, and put all of that together. And that is, I think, the secret sauce. And as developers, we need to think about how do we put that mechanisms together? Are there any solutions in the market? Because if we are able to do it, we can automate very complex processes. something of what I've written here, invoice processing, claims processing, and these processes sound simple, but in large enterprises. Fortune 500 fortune thousand companies, hundreds, if not thousands of people are engaged in these processes. you will not believe, there are banks and financial institutions where 300, 500 people, thousand people are engaged in doing just, anti money laundering checks, know your customer checks. Those are the processes where productivity is absolutely required. what is the value? McKinsey estimates the value to be garnered from automation of these processes using generative AI is close to 2.6 to 4.4 trillion. They, of course, estimate, estimate the total economic potential, but it, these are staggering numbers and this is all achievable. If you think about the US healthcare, there is the whole, the money spent on healthcare in the US is $4 trillion. Guess what? How much of that is waste? $1 trillion because of inefficient processes, and if we are able to use agents and generative AI technology to its maximum potential. We should be able to cut that down. And that is precisely what we want. So these are the things that we need to be, cognizant of that the impact is massive. We can do a whole lot with the Asian tech technology. it may sound complex, but the best part is the technology's getting mature. The price of, the foundational models is dropping. So it's a no-brainer that we got to use that in what we are doing. I just want to leave you as, developers to think about some of the key challenges of putting all these technologies together. The large action model, LLMs with a view to automate a business process. The first one I mentioned earlier, orchestration of services. while you have these multiple tools. How can you seamlessly orchestrate them to achieve the desired impact and the business outcome? Because if I am a business user who's, worried about, managing the benefits administration process in, my HR role. I really don't care what technology am I using. I care about the outcome and how it is orchestrated. What is the transparency in the process? Can I check how my work item is progressing? What is the output that's coming out to me? What are the analytics? Am I measuring the right metrics? That's what I care about. I don't care about technology. As developers, we need to make sure that we need to put, an orchestration layer, which can make these, technologies integrate together. Complex, no, no doubt. While LLMs mask a lot of complexity, but putting the governance, the responsible AI and the guardrails is absolute must. Without that, you cannot really build the trust that is required with your business partners as well as, the leadership and you will never be able to execute key processes. Think about it. Will you be okay with a 90% accuracy in a process that processes your monthly paycheck? You will not be, these are deterministic processes. I. You will have to have the trust that every single time the process gets executed and you get your check, the numbers are right, every single cent is appropriate, and that is where you need auditability, governance, responsible ai. And lastly, all these processes. While great, we understand that LLMs have a tendency to hallucinate. We need human in the loop for certain edge cases so that we understand, humans, verify the output and then let those transactions through. And lastly, the most important thing in this whole agent take automation initiative, if at all, you embark on that. Is the fact that we need to have appropriate change management. This is a cultural change. We need to advise, your leadership, what kind of, Approach. They need to take, they need to be open to technology. Some of the earlier govern governance principles, archaic, security policies, they will not work. We need to work top down, bottom up to change the culture of the organization to welcome agents into the workforce, and we should be okay to work alongside certain digital employees. who are going to assist us, make us more productive and make us deliver more for less for our customers. That is the message that I want to leave you with. if you want to connect with me, please ping me on LinkedIn. I'll be happy to chat. always love to contribute to the community. And, lastly I want to call out that what I presented is all my personal opinion. I have, had that disclaimer, right up front. none of this reflects my employer's, stance on any of these technologies or the markets. this is entirely my personal opinion. Thank you again for the opportunity and look forward to engaging with you again. Truly appreciate, this, thank you so much.