Have you ever contacted customer support with a simple request, only to find yourself trapped in a loop of scripted chatbot responses that never actually solve the problem? It's an experience many of us know all too well.
AI has made customer service more conversational over the last few years, yet there is still a gap between answering a question and actually resolving an issue. That gap is exactly where today's conversation begins.
In this episode of Tech Talks Daily, I spoke with Mike Szilagyi, SVP and General Manager of Product Management at Genesys Cloud, about a new chapter in AI-powered customer experience. Genesys has announced what it describes as the industry's first agentic virtual agent built on Large Action Models, or LAMs. While Large Language Models have dominated the conversation around AI for the past few years, they have largely focused on generating responses, retrieving knowledge, or answering questions. What they have struggled with is execution.
Mike explained how Large Action Models take the next step. Rather than simply telling a customer how to solve a problem, these systems can plan and execute the steps needed to complete a task. Imagine contacting an airline after a sudden flight cancellation.
Instead of navigating multiple menus or repeating information to a human agent, an agentic virtual assistant could understand your situation, check alternative flights, apply airline policies, and complete the rebooking process across several systems. In other words, the AI moves from conversation to action.
We also explored how Genesys approached the design of this technology with enterprise governance in mind. From explainable decision paths and audit logs to guardrails that ensure every automated action can be traced and understood, the goal is to make autonomous AI trustworthy inside complex organizations. Mike also shared insights into Genesys' partnership with Scaled Cognition and how integrating specialized models helps deliver reliable execution in real-world customer service environments.
Perhaps most interesting was our discussion about the human role in this evolving contact center landscape. As automation begins to handle routine and multi-step workflows, human agents are free to focus on situations that require empathy, judgment, and expertise. That shift raises interesting questions about how organizations design customer experiences in the years ahead.
So how will customers respond when virtual agents move beyond answering questions and begin resolving problems on their behalf? And once one brand delivers that experience, will it quickly become the expectation?
Useful Links
[00:00:03] Welcome back to the Tech Talks Daily podcast. Now, have you ever reached a contact center, explained what you need, and then realized the bot can only answer a certain number of questions? If you drift from that script, it cannot actually fix the problem. And this is the kind of gap that I suspect we've all encountered. And I want to discuss this with the SVP and General Manager of Product Management at Genesys Cloud today.
[00:00:32] Now, Genesys has just announced what it described as the industry's first agentic virtual agent that is built on large action models, or LAMs for short. And they are designed to resolve customer requests end-to-end. And one of the things that grabbed my attention right away after hearing the announcement there is it's not just another chat experience announcement.
[00:00:56] It's actually about moving the conversation to execution, while also still keeping governance, explainability, and auditability front and center. So today, my guest will break down the difference between large language models and large action models. Do it so in a language that we can all understand, and help us all learn why LAMs matter when a customer needs something done. Let's say like rebooking a last-minute flight cancellation.
[00:01:26] And I also want to learn more about how Genesys is thinking about enterprise, guardrails, trusted action, and where humans still matter most in the customer experience. But enough from me. It's time for me to officially bring my guest on and introduce him to you now. So thank you for joining me on the podcast today. Can you tell everyone listening a little about who you are and what you do? Hi, Neil. Thanks for having me today.
[00:01:55] My name is Mike Szilagyi. I lead product management at Genesys Cloud. So I'm an SVP here at Genesys. I've been a part of Genesys Cloud since it was a white paper many, many years ago. And I've been leading the product management function since its inception. And one of the many reasons I was excited to get you on today is you recently announced what you describe as the industry's first agentic virtual agent that's built on large action models.
[00:02:21] So for listeners trying to separate signal from noise in the AI space because they're probably hearing agentic and AI and in so many different ways at the moment. So what problem were you determined to solve from the outset? And why is the right moment to introduce this capability right now? Yeah, great question. It's it's you know, we started the same place that everybody started when open AI released their large, large language model.
[00:02:48] Everybody in our industry started at the same point. They saw the promise of what that model could do. And what it's what it really does is it it thinks about languages. It thinks about understanding conversation, intense, and it can make sense of knowledge as well. So it can parse a lot of knowledge and generate answers. So that's the first place that everybody attacked with the large language models, including us. So it was really good about understanding what customers wanted.
[00:03:15] And it was really good to go into organized knowledge bases and create specific answers for those types of situations. But when we got to the third phase, which is actually executing something when you wanted to do something, that's where we started to see some real challenges with large language models. They weren't very good at formulating plans and steps repeatedly. And they weren't really good when the number of steps started to grow as well.
[00:03:43] So the hallucinations that you may see in the language side, we started to see in the action side. Now, some competitors and others have tried to compensate for that by leaving the agency in the voice side and then connecting it with a step by step action model. That's the problem we're addressing with our large action models. We actually are using that same technology that used LLMs to determine speech and knowledge, but applying it to actions. So it's the same.
[00:04:13] There's a model that understands API calls. It understands semantic input of those API calls. It understands the output. It understands whether it was successful or not. And it was really good at connecting the intent to the action. Okay. So this for us is really completing what's necessary in the contact center, which is going from an intent through knowledge, understanding what the customer wants, and then actually executing on their behalf reliably.
[00:04:40] And one of the things I try and do on this podcast every day is demystify a lot of the big tech trends that business leaders are hearing right now. And over the last three years, there have been many conversations and a certain amount of buzz surrounding all things large language models. But today we're talking about large action models. And so I don't leave anybody behind here.
[00:05:01] In plain terms, how should business leaders understand the difference between LLMs that generate language and LAMs that execute work? Tell me more about that just so we can have everyone on the same page. Yeah, sure. Large language models, I think everybody knows they're optimized to generate words. Yeah. Large action models are optimized to generate outcomes. And I think those are the big differences that happen.
[00:05:30] You know, an example of that is if you're calling in to change your flight or do something like that, a large action model can actually rebook it for you. When it understands that you want to rebook and that it understands the situation that you're in. It can actually go out and get that information and then it can call the right systems, follow policy and confirm completion of that and then document the whole interaction as well.
[00:05:56] So not just telling you how you might do it, which is what the LLMs do well, but actually do it for you in a reliable way that meets policy of the businesses that are executing them. And anyone listening that has recently contacted a contact center, whether it is to change a flight, for example, and they've encountered, I don't know, a traditional chatbot or virtual agent. They can answer certain questions, but if you drift from the script, things can quickly go wrong.
[00:06:25] And as a result, customers end up escalating to a human. So what changes when an agentic virtual agent can reason, plan and carry out multi-step workflows across the CRM, billing and different service systems? How is this changing things? Yeah, I think there's two parts to that. When would they escalate to a human?
[00:06:44] And we saw in the original bots and even LLM assisted bot world, we saw that you're still were creating step-by-step instructions and there wasn't much autonomy that was happening. And agency is really about having some autonomy, being able to reason, understand and make decisions. And so I think that's an important part because it doesn't require the designer to create all the steps that are necessary. So it can follow the human conversation. It can adapt appropriately.
[00:07:13] And that's what we do as humans. When we're talking and having a conversation, you can adapt to the change in what the customer is saying. And so I think that part of the LLMs have gotten really good. The second part is if it goes beyond knowledge, if you're beyond an FAQ or something like that and you need to take action, I think that's where the LLMs fell down. Because they were handing it over to the old system of sequential rules-based decisioning.
[00:07:40] And when you get to that point, maybe the customer already gave you the information for step number two. But because you're going through a sequential thing, they'll have to repeat it. And they can't understand. The sequential part of the fulfillment can't understand what's been received already. So you apply the same kind of reasoning and skill on the action side with a large action model. It knows maybe it already got Neil's information about the flights. It already knows where he wants to change it.
[00:08:09] So the only pieces of missing information I need to go get and then I can actually fulfill it. So I think it's that reasoning and that being able to have a conversation both on the front end side as well as through execution is what's going to change the results of these LAM-based agentic virtual agents. And I'm fortunate to go to a lot of tech conferences around the world.
[00:08:35] And what made your announcement stand out from my side was that it wasn't an announcement around a shiny new tool or agentic AI. It was the fact that you emphasized right from the outset governance-first architecture, explainable decision paths and auditability. So how did you design this solution so enterprise leaders right from the beginning can trust autonomous action inside complex and regulated environments?
[00:09:00] Because it certainly felt refreshing to hear this, that you were taking that, you know, belts and braces approach here of IT. Yeah, I think that comes from just our pedigree of being in this space and knowing what enterprise customers are challenged with all over the world. You know, Genesis has that experience and has that pedigree. So when we decided, when we were designing our virtual agents, that's certainly something that we started with. We started from a governance-first mindset.
[00:09:28] So it's designed in and it's not bolted on. And I think that's one of the things that if you saw some of the customers or some of the companies that came to market with some of the early LLM, you know, they really, they've really found themselves in some difficult situations with hallucinations, maybe giving the wrong advice. And so, you know, we started with governance as the first and foremost thing that we needed to do. And so we do things. We also know that there's really good compliances out there.
[00:09:53] So ISO 42001, it's a whole framework on how you manage AI and governance around AI. And so we do have that at a station. We're one of the first, I think, 20 that have it in the world. But it also drives decisions that you make as well along the way. And for us, so we create model cards for every model that we use. We're model agnostic. We use many models inside of our stack.
[00:10:20] But we create model cards to say what kind of data is it executing on? What is it doing with that data? So from there, we've just been able to build on it. With large action models, we knew that everybody was going to want to know why did we pick the action that we picked? And what was the input to that action and what was the output? So all of that stuff is traceable and transparent in our product. So it really comes down to just, you know, governance is designed from the beginning. We knew that we wanted to have enterprise-grade governance across the platform.
[00:10:51] We also put guardrails and explainability in there and audit logs so that everybody that has responsibility in enterprise to know what the technology is doing can get the answers they need. And if we put the tech and the AI to one side a moment and focus on the pain point and the solution, are you able to walk me through maybe a real-world example such as a last-minute flight cancellation that we mentioned a few moments ago?
[00:11:15] And just talk about how Genesys Cloud, a Genesys virtual agent can handle that scenario right from end to end. Because I think it would be really useful for listeners to understand that before and after approach. Yeah, yeah, sure. I think that, you know, the important thing is when you have a last-minute cancellation, there's a sense of urgency around it. You know, there's a sense of franticness around there. So the customer is already in a sensitive mode and wants to get answers very quickly.
[00:11:43] So when you think about what in the old world you were probably engaging with a bot or a virtual agent that didn't quite necessarily understand the pieces of information that you were going to bring to it in the order. So it's going to ask you step-by-step everything that you needed to provide it so that it could actually make that change for you. But, you know, we all know that there were problems in collecting the information and executing on that and getting action.
[00:12:11] So in the new world, you know, an Gentic virtual agent, the first thing it would try to do is just understand the customer constraints. So what's the timing of it? What type of fare class is it? Loyalty status? Special needs. Like we can put all of that stuff in front of the virtual agent as personalization information that it can use to drive this. So before you even contact them, it knows all this stuff. Then it starts to plan the options. So are there alternative flights? Is refunds, is refund a choice? Is a voucher a choice?
[00:12:41] So we can put all that policy and guardrails in the design of that solution as well. So once the customer actually gets to the point where it selects an option, it has to provide very minimal information because we have so much context. Then it can actually execute. It knows about, say in the reservation system, it knows about all the tools, all the semantics of the tools, all the information that it needs to be able to process something successfully. And then it knows the outcomes of what it's going to provide back.
[00:13:11] So all of that stuff can happen with a higher degree of flexibility. But also it's going to have the everything, it's going to know everything that it needs to do to successfully execute. And it's going to work through that in a reasoning and planning way instead of a deterministic way. So I think the difference to me is a shift from answering questions to really resolving the disruption.
[00:13:38] And something else stood out to me was that Genesys also partnered with Scaled Cognition and integrated the APT-1 model to support deterministic trusted action. So what role do you see partnerships like this and specialized models play in making autonomy safe and enterprise ready? Yeah, I think, you know, anytime you say safe and enterprise ready, it's always best to be done in a community environment.
[00:14:04] And I think we can look back in history and we can see how the World Wide Web was deployed and how everything became secure and protocols and compliance all help that. So to me, security and scale is best done in a community environment. So we've always taken that approach with Genesis Cloud. It's very much a platform. It has APIs. We're model agnostic. So that's what was so easy for us when we saw the challenges that we were having with the execution side of an LLM-based virtual agent.
[00:14:33] We ran into Scaled Cognition. They were thinking the same thing. We were able to take their model and put it into our platform fairly easily. So, you know, in this industry, first mover advantage is huge, but things are moving very rapidly. So, you know, soon we'll start to hear other companies that have large action models and the evolution is just going to be continuing along the way. So I think partnerships are super critical. I think building blocks and protocols are super important.
[00:15:02] And I think we're going to see those in the AI world. We already are with things like the A to A, agent to agent protocol and MCP protocols. That's, you know, the right building blocks are starting to emerge. The right models are starting to come out. The models will get less expensive over time. They'll evolve. So I think it's a great thing to do with partnerships and community in mind. And I think it was around 18 months ago, I hosted a panel at a Genesys event in Denver.
[00:15:28] And the theme of that talk was the last best experience we have anywhere quickly becomes the standard expectation for what we want everywhere. And I certainly see that in myself. So from a customer experience perspective, where do you see the biggest shift? Once virtual agents move from scripted conversations to goal-orientated execution, it actually resolves issues. And when companies provide that service compared to those that don't, how do you see all that evolving?
[00:15:56] I think, you know, when you look at the customer service industry today, even the companies that have the best, are known for the best experience out there, they're still not completely meeting the needs of their customers. And by that, I mean, they're not available 24-7, 365 usually. Usually there's hours of operation if you need to talk to somebody to resolve it or interact with someone. And they don't have all the language support that they need.
[00:16:25] So there's lots of reasons why. Maybe they're not on the channels that you want to contact them with. So in a lot of ways, I think they're underserving their customers today. And I think the virtual agents help eliminate that. They take away the ability to have to restrict service to your customers. So now that you can actually understand what a customer wants and you can actually execute that across languages, across channels, and across time,
[00:16:55] then you start to see that that's going to drive a great experience for any brand out there. So I expect that there'll be loyal. Customers will have more loyalty to brands that have these types of services that are available to them. You'll start to see them be able to solve problems, solve them, not just bring them up, but actually get them solved with large levels of customer satisfaction and actually being able to get that done. And then you'll see them just completely move across the front office to back office.
[00:17:24] So I think we're just going to go up in complexity on what these things can do. There's always going to be things that I never want to try to predict when and if they'll take over for humans. There's always parts of the contact center that need a human touch. And we're focusing on that as well, just driving and elevating the things that humans do in customer service as well. But for the things that a majority of us need on a regular basis,
[00:17:50] I expect that we're going to see a lot of customer satisfaction and loyalty driven by these virtual agents. And the speed of tech change continues to race ahead at breakneck speed. And not only that, the year seems to be racing ahead. Already we're entering the third month of 2026. And as we see automation inevitably take on more routine and multi-step work,
[00:18:14] how do you see that balance evolving inside contact centers between agentic AI and human agents, particularly when it comes to things like empathy, judgment and accountability? Because it's not about replacing people. Both are needed here, aren't they? That's right. Yeah, that kind of goes back along the lines of what I was just alluding to before is that, you know, most human agents, you know, they get tired of the monotony work. So I think this is a win-win for both the company and the agent at this point.
[00:18:44] Like the virtual agents will start to take on the more simple work that, you know, typically could inundate them with things to do throughout the day. Now, we tried to move a lot of this to self-service channels, web-based self-service or mobile app self-service. But, you know, those, again, are very programmatically rigid and difficult to deal with. So they escape into the contact center.
[00:19:08] But yeah, I do expect that there'll be a nice blend of human agents and virtual agents in the future. I think that human agents will start to handle the things that people do best with, you know, things that require maybe empathy, emotional attachment. You know, you think about one of our customers is a suicide hotline. It's hard to imagine that there might be an AVA that you might ever want to have a virtual agent handle that type of situation.
[00:19:38] It needs, you know, there's too much risk associated with it. So I think there will always be a role for humans in situations where things are complex or highly emotional, or there's a strong sense of urgency about getting something done, you know, like in emergency situations. But also when things are complex, the complexity levels of some financial institutions, you know, require a human touch. Or maybe that's just the type of service they want to provide as well,
[00:20:05] if it's more of a premium service that brings a human to bear. So I think it's going to give companies opportunities to really put the humans where the humans matter. And something else I try and do on the podcast is give my guests an opportunity to right a few wrongs that they may have seen when scrolling down their LinkedIn news feed, or maybe on Reddit or anywhere in the media. I mean, what do people most misunderstand about your industry?
[00:20:32] Or are there any myths or misconceptions about your job or field of expertise that we can finally lay to rest today? Anything spring to mind now that frustrates you at the moment that we can put to rest? I've been in the contact center industry for a while. And for a lot of years, it was always seen as something that customers or brands or businesses needed to have, but they didn't necessarily want. Because it's, you know, it's a cost center.
[00:20:59] But I think we've seen over the years that it's cost, managing cost is still important. But also, it's starting, we're evolving to the point where it's a revenue center, where it drives loyalty, it drives customer retention. And ultimately, that can turn into revenue as well. So I think that, and what that means for me is that, you know, Genesis is in a perfect position to solve that crisis of loyalty. And we call it the experience economy.
[00:21:28] And I think often people are looking to maybe some of the hyperscalers that are coming up with the AI models, or maybe some of the startups that have applied it to customer service. But I would challenge and say that Genesis is in a perfect position to manage this technical transition, because we understand contact centers, we understand enterprises, we understand all the data that exists in the contact center, and we know how to apply AI to it. And we're very much driving that technical innovation with virtual agents.
[00:21:58] You know, nothing is more important to us than bringing virtual agents to market, because we think it's a win-win-win situation, win for our customers, win for our customers' customers, and a win for us. So we're very much aligned to, you know, driving that technology into this space to really prepare for the experience economy and drive loyalty and customer satisfaction in the industry. Well, thank you so much for sitting down with me today, demystifying this technology,
[00:22:26] talking about the real pain points and the solutions to that, rather than just talking about the technology itself. So thank you so much for that. And for anyone listening wanting to keep up to speed with anything we've talked about, especially some further announcements that might be on the horizon too. Where would you like to point everyone listening to find more information? Yeah, they can certainly start at our website at www.genesis.com. We have lots of information that we discuss sitting there as well, but also look into our public social channels as well.
[00:22:55] So out on LinkedIn, we've got a lot of information, a lot of blog posts that we share about our thoughts in the industry as well. Well, we covered a lot today from the impact virtual agents built on LAMs will have on customer experience, how virtual agents will work alongside humans in the contact center, and how Genesis Cloud, a Gentic virtual agent, how that will work, helping people understand the kind of value that will bring as well. So I will add the website details, all the social channels as well. I'd urge people to check that out.
[00:23:25] But more than anything, Mike, just thank you for bringing all this to life today. Yeah, thank you, Neil. Thank you for having me. Great questions and very much enjoyed the conversation. So if you take one thing from today's episode, I think it is that customer experience automation is entering a new phase. Because for years, chatbots, virtual agents, et cetera, have been decent in deflecting simple questions. But the moment the customer needs a real outcome,
[00:23:53] that handoff to a human was almost inevitable right from the beginning. But getting that handoff was often quite difficult. So what Genesis is describing with LAMs is this shift towards systems that can actually carry out multi-step work across the tools that business has already run, and doing it in a way that leaders can explain, audit, and govern. So a big thank you to Mike for also leading with a focus on trust and responsibility.
[00:24:20] Because autonomy inside a contact center is only as good as the guardrails around it. And the ability to answer a simple question after the fact, and understand why did it take that action? What data did it use? And what happened next? For me, that is the difference between another shiny cool demo, a tech conference, and a keynote speech, and something enterprises can actually put right into production without losing sleep. And I think there's also a human story here too.
[00:24:50] If agents can take the repetitive work that burns people out, I think it creates space for human teams to focus on the moments that require empathy, judgment, and accountability. And that balance feels like where the next chapter of customer experience will be won or lost. So as I said, I'll add links to Genesis along with their social channel so you can keep up with everything we talked about today. But as always, I'd love to hear your views.
[00:25:17] Are you excited about virtual agents that can execute end-to-end? Or do you still just prefer a human? And do you feel more comfortable with a human in the loop for anything else that matters? techtalksnetwork.com. Please send me your thoughts and experiences and insights over. But that is it for today. So I will return again tomorrow in your podcast feeds, waiting for you to hit play. I'll speak with you then. Bye for now.