How to deploy agents—without the chaos

Hello, and welcome to this episode of our Analyst Angle series. I'm your host, Shelley Kramer, Principal Analyst here at theCUBE Research. And today we are going to dive right into a topic that is top of mind across organizations as they realize the next step on the enterprise AI front is agentic AI. And they're planning for significant investments in AI agents. I'm joined for this conversation today by Tray.ai CEO, Rich Waldron, who is tackling that problem head-on. Welcome, Rich. I'm so glad to have you.

Thank you for having me. Great to be here.

Absolutely. Absolutely. So an industry focus on agentic AI is beginning to boom, and of course, with good reason. When today it seems as though every app is AI enabled and the tech stack cannot be any more complex than it is, integration can be a bear and competing stakeholder priorities are also a reality in this equation. And siloed approaches and single purpose products inevitably lead to challenges with governance and vendor lock-in and, you know, I noted that Forrester predicted that 75% of firms that build agentic architectures on their own will fail. That's an attention getter. And that's where a scalable integration platform can play an outsized role. And that's exactly what we're going to explore today. So Rich, I'm so excited to have you.

Before we dive into this conversation, though, I would love for you to share a little bit about your career backstory and kind of your journey and how you made it to Tray.ai.

Yeah, sure thing. So I did a computer science and business degree a long time ago now, probably over 20 years ago at this point. And kind of off the back of that, I didn't really know which direction I wanted to go in, that I want to go more down the engineering route. I enjoyed the kind of product side or business side of life a lot more. I had a couple of years stint at a startup based here in the UK, doing a number of roles, but ultimately settling on product management. And that, for me, was kind of the jump off point to go out on my own and start a company with my two friends, Dom and Ali. And sort of the, I always joke, the kind of biggest asset we had at that time was naivety, because kind of not knowing what we were getting ourselves into, having that kind of like unbridled entrepreneurial spirit to get us started was one of the things that sort of sent us hurtling into this integration challenge head on. And really, ultimately, was the primary driver as to why we took sort of a different path when it came to solving a problem that many before us had tried to solve through different eras of technology.

You know, I love the honesty of that answer and the naivete. My favorite dad line, and I use it often, is, we don't know what we don't know until we know. And, right? You go into this, you bring your enthusiasm and your ambition and your drive. And then, you know, and the other thing I'm always grateful for is that I happen to be wired in such a way that I love solving problems. So, you know, so sometimes that naivete is a good thing because, you know, you sort of go in with a clean slate. You realize you don't know what you don't know. Then you know. And then you start figuring it out. So, I love that.

So, I know that Tray.ai recently did a survey called the state of AI agent development strategies in the enterprise. You surveyed over 1,000 enterprise tech leaders and practitioners. And the research showed, not surprisingly, that data management, security, and complexity all play a role in the ability for enterprises to capitalize on the real benefits that AI can deliver. Rich, I know that many enterprises are recognizing AI agents as, you know, really the value driving next step in AI innovation. But your study showed that 86% of those survey respondents shared they require upgrades to their existing tech stack in order to be able to deploy those AI agents. What do AI agents require that organizations don't have today in their existing structure?

Well, yeah, I mean, the survey itself was fascinating. And it echoed kind of the journey that I've been on this past 12 months where I've been out in front of hundreds of our customers meeting CIOs on a monthly basis, out there kind of getting on the shop floor to a degree with prospects and understanding some of the challenges that they're facing. And where I've landed is it's an acceleration and in many ways an evolution of the kind of long held integration challenge. You know, we at Tray have been in business for over 10 years now. And the challenges that we've always seen in organizations were that they were always trying to get their data centralized in some form. You know, they've got all these siloed applications. All of the most useful tenants of information are sort of cluttered all over the organization. And some of them are in really nice cloud services that are super easy to access and useful to do something with. Some of them may be in, you know, archaic on premise systems that are much harder to get a hold of. And dependent on, you know, how long you've been in business, the kind of organization that you're running, you're going to have a real kind of mismatch of applications of data and of these structures that exist within companies. So when you think about how this applies to the AI agent challenge, which is you've got this incredible technology that can reason across a huge, vast number of data. It can take all of these different individual pieces of information and kind of give you the best or the most intelligent response or the most suitable action to go and have the impact that you wish. Well, for you to harness that technology, you have to have been able to get your arms around all of the most useful information in your organization in the first place. So I think a lot of companies are recognizing that, you know, they've got to be able to get all of this data streamlined either into one place or interconnected into the agent in the first place so that it's got something to respond to. A quick example of that would be if you take your sort of standard knowledge agent, which is be able to respond based on the knowledge that you feed it. Well, you know, the trick is kind of in the name there, right? The better the knowledge that you can feed it, the more history or the richness of the data that it can work from, the better the overall agent experience is. And I think for a lot of organizations that that sort of tech upgrade is mostly centered on how do we actually do the data integration piece? Because once we've got it there, actually, there are a few different paths we can go on to be able to unlock agents ourselves.

Yeah. So solving for that on the front end of it is incredibly important.

Oh, for me, it's probably one of the most important challenges because you can have the smartest AI in the room. But if you don't, if you're starving it of data and you're starving it of the thing that it needs to go and respond to, then you've already kind of hamstrung its capability.

Yeah, absolutely. I know that your survey showed that almost half of enterprises, about 42%, reported that they need access to eight or more data sources to deploy AI agents successfully, among other widespread integration challenges, including security. So we talked a little bit about some of the challenges specific to integration. What are some other pressing challenges you're finding enterprises are struggling with right now?

Yeah, I think there's a number of challenges that enterprises are facing when it comes to actually getting these things deployed. And a kind of big proponent of that is actually changing the way or the mentality that they have when it comes to getting things out in the wild. To really benefit from agents, I think you've got to adopt a kind of prototyping mindset. You've got to get yourself into a position where you're willing to kind of try and iterate extremely quickly to be able to determine how valuable is this thing anyway? Because I think there's this huge rush to go and adopt the technology. And we've heard all the great stories and the success stories. But in the same vein, it's got to be directed at the right use case. It's got to be directed at something that is ultimately going to have a genuine business impact. And I think what a lot of organizations are finding, and in our findings over 40%, that takes numerous sources, that takes numerous places to be able to gather data from and ideally act on to have that impact. So from an organizational perspective, you've now got the entire company kind of staring at you in IT saying, hey, we've all got these ideas. There's all these things that we want to be able to go and do. How do we get them into production quickly? How do we start to see the benefits of them? But in the same vein, well, in order for us to do that, how do we go and interconnect all these different data sources? How do we actually capture what's going on in the business and then critically act on it, which is another big challenge within the tree there. So I think overall, there's an approach, there's a mentality piece. I think a lot of companies feel quite overwhelmed because it's picking the right thing to go after at the right time. And then lastly, it's the technical piece that we've spoken to, right? We haven't even got into building agents yet. We're just trying to get the foundation right so that we can go and use the technology in the first place.

Yeah. And I think that, you know, to me, you mentioned that people feel overwhelmed. And I think that that's a really important point here, because when you step back and you think about it, most of these decisions, most buying decisions today are not made by just one person. It's made by a committee or a group of people. But the reality of it is when we choose to, you know, when we are inside an enterprise, when we work together to decide on, you know, a vendor that we want to work with and when we invest in a technology solution, the reality of it is that our reputations are on the line. You know what I'm saying? And if this is not a well thought out decision, you know, it's a decision that comes with stress. And so that's really, I think, where really understanding what this landscape looks like, really understanding the key components of getting it right at the onset and then finding, you know, a trusted vendor partner who can deliver on that front. And I think that that really is kind of like, you know, the holy grail of what everybody is looking for.

Yeah. Yeah. I think there's a nuance to that as well, which is the, you know, the historical solution to that problem was go a bit slower, right? Pick a well-known trusted vendor, run a long implementation process. Right. But there's now this fear that, oh, we're going to be left behind. Yeah. If we don't move quickly, if we don't take advantage of this technology, or if our competitor figures out how to deploy a customer support agent that accelerates the way that they can handle their overall customer experience, and we don't, what does that mean for our business? So I do feel like it's put a lot of pressure on organizations to be able to start getting into a place where they are comfortable adopting technology in a quick manner. Obviously, safety and security is absolutely paramount. But the way that you draw up those guardrails is really, really important.

Yeah, I agree. And, you know, to your point, I don't think there's any conversation that's happening around AI that's saying, let's just slow down and take our time. We've got all the time in the world. That is not what the industry landscape looks like right now. So how can enterprises address these challenges and eliminate that complexity in the building of and the deploying of agents? Well, I think the first point is, once you've got the, you know, once you've figured out the area that you want to go and tackle, you've decided on, you know, this is the business problem that we're looking to go and solve. It's really what is the quickest way that we can get a prototype into production within the organization, right? How do we get our arms around the data that we need? How do we get something in front of most likely our employees or our team where they can start interacting with it? And how do we get that feedback cycle working quickly enough that we can determine, OK, we're on the right track. It's having the value impact that we wanted to. Do we now need to go and iterate on it or evolve it in any way? And, you know, from our perspective and from what I'm seeing at Tray, one of the big challenges with that is the implementation process of creating the agent is pretty heavy. You know, you don't really have too many options. It's either build a custom application, which comes with writing the code, scaling the machines, picking the model. Then you've got to go and build testing around it. Then you've got to put a whole kind of application framework around the structure or the construction of this agent in the first place. In doing so, you then have the burden of maintaining an application, which is certainly not a small deal. The flip side of that is, do you go and buy something out of the box? So do you go and turn on the AI from all of the vendors that are knocking on your door? And if you've got 300 or 400 vendors, all right, well, which one do you pick? And, you know, good luck. You've just become an AI referee trying to referee who's doing what, where and how. So that in itself comes sort of riddled with challenges. The approach that we've taken and kind of the capability that we've been so excited to launch recently is actually centering it around bypass and around integration, where you've already got that kind of capability born in, right? By definition, you're creating orchestration. So going and adding that layer where you can go and, you know, connect to an LLM, push the data into a vector database, take advantage of the technology, means that you're already at the point where you have something that's quick and easy to go and deploy. And when that feedback cycle comes in, you're in that pivot point. It's a low code implementation. It's a, actually, we need to go and add an additional service. Actually, we need to go and make a change. These are happening in, you know, in some cases, hours and days, rather than, you know, what we're seeing elsewhere, which is months and years. And that's not good for anybody.

Well, you know, speaking of the need for speed, your survey showed the majority, I don't know, maybe 60 plus percent of the respondents want to deploy agents in no longer than three weeks. And this is to your point, this is, you know, really shortening that cycle that we're kind of accustomed to. So what is Tray doing to make this possible for customers? And I know you've developed the Merlin Agent Builder and Tray Agent Accelerator. So walk me through a little bit of what those solutions are and how they're designed to help solve for this.

Yeah, certainly. The three week number makes me chuckle a little bit because, you know, starting back in IPASS over 10, 11 years ago now, you know, you always hear these horror stories when it comes to integration implementation. And, you know, the goal was always, hey, we've got to get this stuff out and deployed extremely quickly. And actually, technology enabled that. It meant that backend services changed. It meant that cloud services allowed a much richer and faster compute or accessibility through the way in which they built their APIs. And I think from an agent perspective, you know, we've lent into that same trend. The Merlin Agent Builder, which is a capability that we recently launched, gives our customers the ability to construct agents visually and instantaneously. And the fastest way for them to go and do that is we've built these packages called accelerators. And so an accelerator is a, think of it as like a template plus plus. It's, hey, here's an out-of-the-box IT support agent that you can one-click install, authenticate the services that you use with, and that will be ready to go in probably under 10 minutes. Now what you can go and do, because it's backed by the Tray platform, because it's backed by all of the tooling that we integrate with and all of the services that we kind of have at our fingertips, it means you can go in and actually customize that, change the services that you connect to, change the data that goes in. All of this is happening instantaneously. And suddenly that three-week goal actually looks like a long time because you can get something stood up that you can go and put out in front of a group extremely quickly because we're handling the deployment process, we're handling the error handling, we're handling the maintenance, we're handling all the connectivity. It's putting agents at the center of the challenge, which is actually, from my perspective, integration itself.

Right. Absolutely. You know, and you mentioned the importance of iPaaS, and really I know that iPaaS is really kind of a strategic asset these days. So how do you see the role of iPaaS evolving in AI adoption? And do you think we're going to see a greater reliance on iPaaS?

I mean, I would guess the answer is yes, but what do you think? Well, I think it's the greatest risk and the biggest opportunity for iPaaS right now. I think iPaaS has to evolve. And the exciting thing about this is if you kind of piece back the history of iPaaS and the purpose of it, it was to provide effectively a low-code way or a framework way to build integration quickly. So way back when you'd have one single ERP that needed interconnecting between different systems, you'd have on-premise systems that needed to talk to each other, you'd have on-premise to cloud systems that would need to talk to each other. So suddenly, with this AI boom, with this whole new kind of intelligence angle, every process that integration covers is up for grabs, right? It's being reimagined. And so I think the critical role for iPaaS is, well, you're already in the key seat. You're already the strategic player which provides that glue and that connectivity between an organization. And the key part is you're not just pushing data around, you're an action layer. If somebody wants to go and do something in a third-party service, they do it via their iPaaS. And so the angle that we've taken here is that you've got to lean into this extremely hard because there's a real opportunity for iPaaS to become the kingmaker for AI. It becomes the way in which that deployment is actually going to happen in the first place. The counterpoint to that is if you don't enable that capability, all of those processes that are being powered today by traditional integration will have to go somewhere else. And that in itself creates some risk. It creates a brittleness for an organization. They lack that central governance and the control and all of the things that iPaaS brings. So really, it is the opportunity of a lifetime to be able to go and tackle this challenge.

I love it. I love it. So one of the things that is a reality is that AI is driving up data volumes and it's requiring better management of unstructured data. And I don't know an organization out there that's not focused on and trying to get arms around data management, structured data, unstructured data, all of this. But specific to unstructured data, why is it so important for modern platforms to have unstructured data handling capabilities? How are you helping solve for this with customers?

Well, yeah, the unstructured data piece is kind of the thing that we've all been promised for so long and not been able to take advantage of. And when I think about the IoT boom that we saw probably five or six years ago now, the excitement was, well, suddenly we're going to have all this richness of data and knowledge and conversation and all of these data pieces and the volumes and the vastness of them are going to be enormous. And if we can analyze them and do something useful with them, well, you know, what an amazing insight we're going to get. You know, if we know every time somebody changes their central heating in their home or every customer conversation that has ever occurred across our entire business is categorized and is unstructured but available in some massive, massive database. Well, the exciting thing from an AI perspective is now actually going and doing that analysis, doing that reasoning can occur again instantaneously. And so actually to again harness the power of AI to take advantage of the acceleration of the technology, how you can capture and handle and push unstructured data into these workflows is critically important. What we've been doing from a trade perspective is we're the first iPaaS to offer a native vector database solution. That means our customers can collect unstructured data from any service, any source that they require and throw it into our vector database solution. If they already use one, we support all of the major vendors at the gate. So it's really easy for them to go and do that. But what you're enabling and what you're doing in capturing that information is you're giving the AI more context and more variation to go and react to. And that's really when it operates at its best. So I think this is a really exciting time for what has been an industry of hoarders for so long because we've been sat on all this rich and vast knowledge, but we've not really been able to do anything with it. So it is such a critical component to capturing what is available right now.

Yeah, absolutely. And can I talk you into sharing a customer use case story or two with us?

You certainly can. I mean, we're very proud of the customers that we work with and some of the conversations that we get to be a part of. Some of the ones that have really fascinated me, there are some truly enormous organizations out there that work with Tray who have got AI use cases deployed within three months of starting them. And that includes getting those initial prototypes out, getting them pushed through and then actually pushing them into production with their end users. And these are companies that are hundreds of years old who, for them, this is a whole new paradigm. And so that to me is the thing that really cements the seismic shift that we're seeing in the industry. When I think about companies that have got agents live recently, probably the one that sticks out for me is the team at Zuora, the CIO there, Karthik, and the VP of IT, Mark Gill, they're so exciting to work with because they're so focused on their employee experience. They want to create an organization where they can not only make the entire workforce more productive, but they can give them an experience, which means that they go and do their best work. And so they went live this week actually with a series of agents which are handling things like IT support, HR applications kind of across the stack. And some of them involve pulling from data services that actually were not particularly accessible previously, but held really critical data. So that's a kind of great example of a customer that has built something, got it live, and is actually getting it out in the hands of employees straight away. What they've got coming next is some really exciting marketing use cases that we're seeing and that have an agentic feel to them because they're reasoning on the inputs that are occurring in and then determining, actually, this would be the best source. This is the thing that I should go and push back. And this is helping that team produce better quality materials, but also be more responsive to the field. So on both those fronts, that's an organization that's kind of really innovating very quickly and one we're really proud to work with.

Yeah, I love that. You know, I was at Cisco's WebEx One Analyst Forum event here a few weeks ago. It all runs together. But one of the product leaders was talking in a keynote, and what he said really resonated with me so much. And you've really touched on this as well. And what he said was, you know, in all these conversations we have about AI, we talk about productivity, productivity, efficiency, efficiency. And he said, you know, I don't want to talk about that. And that's not what AI is about to me. Innovating with AI, to me, is about delivering better experiences, better customer experiences, better employee experiences. You know, and when you think about that, you know, happy workers thrive and do great things and happy customers also thrive and do great things and they stay, you know. So I really thought that that reframing in this conversation and moving away from all things productivity, which is really, I mean, it's great, but it's kind of boring in the big scheme of things when you can really focus on and shine a light on experiences. I thought that was really a great observation.

Yeah, you've just reminded me of another story, which is a large legal firm that we work with that's based in Europe on a quarterly cadence brings an entire new army of trainee lawyers. And the experience for those lawyers as they come in is for the first couple of years, they're effectively having to go and do a lot of manual lifting on behalf of the trained lawyers, right? Go and assess what's going on with this business, go and pull this data from this thing. These are smart people. These are, you know, at that early stage of their career, they want to be getting involved in what's going on. What they've been using AI for is they've actually been getting those trainees to become AI proficient. And then the work that they have previously been done, they're now automating using AI. So for the next generation that come in, well, A, they're not doing the same work that they had to. But secondarily, they're basically training an army of lawyers that are AI proficient out gate. And so when you talk about experience, we're putting the computers to work, we're using the technology in the way that it's intended, the connotation is only positive. And I think that that to me is the thing that resonates the best, because it is an amazing thing to behold. It changes the way we can all go and go about our work day. And I think that that is the thing to really celebrate, because it is going to have that impact on us all.

Yeah, I love it. Well, as we wrap, it's obvious that we're at a transformational time, and it's incredibly exciting. And while the development and deployment of AI agents is a high priority, I think the most significant need is the ability of an organization to create an AI-ready environment. Break down those silos, unify fragment approaches, streamline complex workflows, and provide a foundation. You know, it's kind of like building a house, right? You got to get the foundation right. Well, you want to build on AI, you got to get the foundation right in order to have AI success at scale.

Rich Waldron, thank you so much for joining me today. It sounds like Tray is bringing the right solutions to the enterprise table at the right time. I've so enjoyed our conversation today, and I appreciate you making time for it. Thank you, Shelley. It's been a pleasure. Absolutely. To our viewers and listeners, I'm your host, Shelley Kramer. Thanks so much for joining us for this conversation today, for joining us on theCUBE, your source for enterprise and emerging tech news. We'll see you next time.