Welcome to the Tray.ai session here at the AI Deployment Summit where we'll explore the far reaching impact of AI agents for your enterprise. Notably, we'll look at the opportunities and the challenges for getting the most of your investments in agentic AI. And to make it real world for you, we'll even get a demo. And we've got two great speakers this morning with us, Paul Turner, integration strategist, and Luke Smith, sales engineer. Gentlemen, welcome.

Great to be here, Vance.

Thank you so much for having me.

And we're really glad to have our speakers with us this morning. Paul, for his part, advises enterprise customers on AI, AI agentic, and integration strategies. In fact, he has more than 20 years experience in SaaS integration issues with a recent focus on AI agents and analytics.

And for his part, Luke Smith, the sales engineer, works a lot with customers on making automation and AI and agents easy to adopt, and he has special training for using AI to automate complex processes, which, lucky for us, is going be on display today during the demo.

In the session this morning, From promises to outcomes: Overcoming the hidden challenges of deploying AI agents, we're going to have both Paul and Luke talk about how AI agents have captured the attention of many companies and how they promise to revolutionize businesses and deliver big value. Notably, there are some challenges. And Paul and Luke are going to take us through some of those, including integration complexities all the way to scalability nightmares.

Paul and Luke will also share insights on both the opportunities and risks for agent initiatives and explore the architectures and capabilities you'll need to overcome any roadblocks you might run into.

Perhaps most exciting, we'll finish with a use case and a demo of a Tray's Merlin Agent Builder in action. So great speakers, great content. One last word. You can download today's slides.

Just click the big red button under the new screen. You'll also see some other valuable resources and even access to sign up for a demo that you can take a look at. And we love our attendees to interact with our speakers, especially in these AI topics. So to ask Paul or Luke a question, just type in the submitted question box.

And so with that, guys, let me turn it to you and tell us about the idea for promises to outcomes, overcoming hidden challenges of deploying AI agents.

Thanks very much Vance for the intro. I'm feeling really old at this point. Twenty years now. I was just looking at my bio, and it's actually over 25 years, so it's been a while. Thanks for joining everyone. Really excited to share with you some industry perspective on opportunities around agentic AI.

Got some cool survey to show you. We can you spend some time on where we are seeing demand coming from the business, but also landmines to avoid as well. Things to watch out for as you start this inevitable journey. You're gonna be deploying agents for a good part of your life for over the next couple of years.

It's very common key delivery considerations, I'll share with you customer success, some real life success. I'm going to hand it to Luke, and he's going to share with you a demo, and you can see it all come together.

Quick background. I'm a former software engineer, computer science major.

Spent time in roles and strategy roles, comes like Oracle NetSuite and Workday, specializing in data analytics and business processes. And Luke Smith, my partner in front of me here, he's our principal solutions engineer. He came to Tray from Microsoft, fellow technologist with a background in cloud infrastructure. So we spend a lot of time all things integration, machine learning, AI, analytics, business process, all those kind of things.

So no doubt agent AI is on your radar for this year.

Hopefully, you've not reached agent max as we stand this year.

Gartner is predicting huge transformation opportunities around productivity over the next three years with the dawn of agentic AI, shift towards autonomous processing as agents become more pervasive, and the reasoning becomes more advanced. So they're predicting 15% of day-to-day work decisions will be made autonomously. Put that into hours, about an hour of decisions you make every day, you're going shift to an agent.

And we're already seeing huge advancements. You've got OpenAI as the operator all the way through to the efficiency gains in large language models, such as from DeepSeek as well. So the pace of innovation is really accelerating, and it's already fast, right, to start with.

Let's give you an example of where companies are placing their bets.

So, at Tray, we conducted a survey of over a thousand respondents. We asked them what problems they're looking to solve. It's a lot of the human task heavy functions are in focus from IT service desk, 60% here. Things like ticketing and provisioning, all the way through the customer service, as well as things like automating sales and employee onboarding processes. So it's typically manual heavy processes.

If you're in IT, you're task delivery, and no doubt you're seeing demand across the business function as well. You have a lot on your plate. A lot of big productivity benefits, but also a lot of broad based demand across the business.

But the reality is that agentic AI is much more than deploying a large language model. There's a lot of considerations. Here's the kind of feedback from some of the folks we surveyed. Beyond the initial requirements, some of the challenges, integration. How do you integrate AI with your stack? How do you integrate agents with your front ends, right, with your website or with your portal or with Slack or Teams, those areas?

Your data governance and compliance are huge considerations. Especially if you're using public cloud AI services models, you want to ensure that your company's policies as well as regulatory rules are followed, especially if you have teams out there that are building autonomously within the line of business. And, obviously, as you scale, you may have tens, hundreds, thousands of concurrent requests that you're responsible to the business, especially as agents are acting more autonomously, and so you have to manage that latency as well as performance and scale too.

Shari Lava, IDC research director, really focused on that the issue of integration. Integration is essential to get the most business value out of Gen AI. You bring the models and the agents together with your business data, and that can be challenging, especially with your AI stack changing. But integration is more important than ever as you're rolling out agents. And if you don't do it right, you run the risk of, everything from data security all the way through just simply hallucinations and errors.

And with agents, there's more than meets the eye. It's not just about delivering agents, but there's all this stuff, quick wins, business ask for knowledge agent or a line of business conversational agent. But when you drill down and look at everything that's beneath the waterline, everything from what might work great in a prototype doesn't maybe work out production, getting sucked into the data integration headaches, security compliance challenges, change management, scope and guardrails around it. But even that boring old stuff like SDLC and testing, debugging, tracing issues, instrumenting at all, it's easy to get caught up in the fast edge. But you really have to think about the stuff below the line that can increase complexity. That's important to make sure that you get the outcomes and success.

And, also, you need to make sure it's flexible.

The state of the AI agency, Gartner puts it in the it called the AI agency gap. The agents are moving more towards more agency, more autonomy.

The investments you make today have got to be flexible enough to meet demand as more high agency demands around agents exist. Otherwise, you can be left in technical debt. The platform you invest in doesn't necessarily get you from point A to point B one or two years down the line.

When you think about building agent, you know, you really have four options.

You can build it yourself in internal engineering project, but it takes a lot of effort and creates a lot of maintenance, harboring the LLM. It's not really repeatable when you think about all those initiatives I shared across business. You're building each one of them as a one off.

It creates a lot of effort, a lot of rework. It's not particularly a great use of developer time. You take an off the shelf agent. There's a lot of apps out there.

You can buy an onboarding agent, a customer self-service agent, or a knowledge agent. You can buy them as an app. But if you think about that as a one off, each one of those is a stack to administer, an application to administer, separate security, administration, customization, all those areas.

There's what what I call ecosystem agents, things like from Salesforce or SAP or Microsoft. But as we all know, those apps and platforms from those vendors often only work within the apps and ecosystem they provide. But if you think about agents, they're reaching across your entire stack way beyond, for example, what you might have for a Microsoft or SAP, especially with the data they're reaching into. Or you can take a platform approach, which I'm gonna delve into today, which is AI ready iPaaS.

And this is really one platform to build and deploy your agents. It's really flexible across all your agent use cases. Use low code development to build and map in new data and add new tools or skills to your agents. We'll delve into that later on.

And they already have an infrastructure designed to scale so they can handle the workloads, for example, in Tray's case and scale up with a high level of parallel processing and supporting a large volumes of queries and agents running whether, they're running conversationally, semi autonomously, or autonomously.

Another benefit to AI ready iPaaS is you end up with a whole separate stack for agent tools. This is really important because you'll have existing non agent demands, everything from building traditional data integrations and application integrations, automating processes, order to cash process, for example, you pop in your data warehouse, building and publishing APIs, all the traditional stuff. Even adding intelligence to existing processes. Maybe you want to add document processing to your order entry. And then as agents as well. But you don't want to end up with one platform for the old and another platform for the new or different platform for each because that creates IT headaches. So moving to one platform, less to administer, less developer enablement, one platform to monitor and scale.

Gartner really concurs with this, and that's what they're seeing is by 2029, 80% of enterprises will have shifted to a more consolidated approach, orchestrating business processes and agentic automation.

I know we're sharing materials afterwards, but let's just drill into what AI ready platform looks like. It starts with composable development. At the top here, how you build agents. You can build them using low code, visual development, and define them, and you connect them with your stack.

You can take a code first approach with APIs for your developers. They can define, instantiate, and integrate code first or just using out of the box system. Now iPaaS capabilities are really key to ensuring they are flexible enough. It's the second layer.

They can conduct tasks, for example, booking a meeting using orchestration, or you can integrate data essential for RAG to avoid hallucinations, so things like data transformations and tables and lookups.

API management ensures that when you build an agent, you can retask it to not just integrate with your front end, maybe your developers, if you consume it using, for example, React, all the way to plugging into your portal or your productivity tools. And connectivity ensures your agents run across the entire stack with hundreds of connectors and extensibility to add more connectivity, for example, to your ERP system or to your HR system.

At Tray, we incorporate a technology layer called Merlin Intelligence. AI serves us in a number of different ways. There's intelligence service that we build, so automated development.

So, like, a GenAI copilot assists with the building process.

What we call an AI Palette with access to things like your large language models that you may want to use, like OpenAI or Anthropic or Cohere or Bedrock using tools like IDP, for example, to ingest unstructured data.

We're going to delve into the demo today, agent builder, helps you create agents and equip them with tools with skills while ensuring tight management of scope and then plug them into mainstream collaboration tools like Slack or Teams. You're going to see that in the demo. And all of this is with a single governance management scalability layer.

One system makes it really easy to manage all of your different requirements in one platform.

We provide a lot of packaged integration processes, templates.

You can go over to our template store. Things around guardrails and safety and Slack integration, knowledge agents. We provide a thing called Merlin Guardian learning that provides things like data obfuscation and tokenization to make sure that you can basically mask data you're moving in and out of the large language models that you use.

And we also provide things like step isolation, which really helps with agent testing prompt engineering, and things like text analysis, all these kind of services you can use to build better agents.

What do you build with all of this? Everything from conversational agents, handle things like customer queries on your website, or help conversationally qualify leads, or handle HR or help desk questions in Slack. For example, to review contracts or invoices and then route them to humans for approval, or maybe evaluate resumes and candidates using AI. Through to fully autonomous agents that handle complex tasks like customer support from instant to resolution, pulling across channels to autonomously analyzing, taking action on contracts, for example, just a little spectrum of agentic.

We know it's sometimes hard to know where to get to started, but one of the things we also provide are accelerators to built our platform.

From a knowledge agent that connects to apps and data. For example, ask questions, how to fix an issue, for example, to custom IT ticket agents that plug in Slack. So we provide these as prebuilt apps you can quickly deploy. Because they're building the platform, you can fully customize them as well.

One of the big use cases for agents is having our structured data, like from invoices to resumes. You want to enable your agents to do that to ensure your agents have a methodical approach. With our platform, we include IDP, intelligent document processing.

You or your agent can form natural language queries to extract data from documents and use it in the downstream process. So you can, for example, use an agent for an invoicing process or a payables process. And we also include vector storage in our platform as well. This is really important because your agents can be dealing with unstructured data and you'll be using RAG. It's really important for knowledge agents, search, and recommendations in those areas. You can use basically vector processing within the platform.

Let's drill into Merlin Agent Builder and how it works from the platform. It starts with an API trigger. So for example, could be triggered from Slack or Teams or your website. An agent consists of multiple tools, things it can do, but they're basically skills. So for example, the agent looks at a request, it reasons across the tools you've made available to it and then executes the right tool, so maybe look up or search or creating a ticket. It creates a response, which might be a conversational answer or an action. And the query coming in could be a natural language prompt.

But what kind of tools are available? Well, tools you can equip your agent with, actually, from searching the web for you on your behalf, looking up customers or employees in your CRM or your HR, interacting with web pages, scraping web pages, so operator style kind of engagement.

There's also a lot of flexibility to build yourself. For example, platform, seeing an agent that's creating for your database on your behalf, and we'll handle all the orchestration around that. Merlin Agent Builder comes with more than 20 pre-existing tools. You can build tools for yourself, from scratch if you want to, you aren't locked in. Mentioned earlier, a lot of our customers using agent accelerator. There's a great starting point across the IT ticketing support on those areas.

Agents also plug directly into Slack, Teams, and other tools as well. And Luke's going to show this conversational experience directly into Slack for something, for example, like an IT self-service chatbot. We can ask questions around things like deprovisioning or provisioning, asking for resources, or just asking for knowledge as well.

So the key here is powerful agents, one powerful platform. You can build the tools, the skills that the agent is visually using low code. Sometimes with the alternatives out there, you're building code or it's really hard to build new tools, skills you're relying on the vendor.

Connect anything to your stack. We actually provide hundreds of connectors out of the box. It's extensible with the development kit. Often, for example, with ecosystem vendors, you're really locked into Microsoft or on Salesforce centric set of apps.

Developer productivity, this is really key. You don't want developers to have to use a whole different experience for both agents. We support integrations, product automations, agent development in one IDE vs having to learn something brand new. Document processing, for example, agents handling invoicing or reading resumes, candidate recruitment, those kind of things, it's included in our platform. Security management, one single governance layer. And elasticity, Tray is a serverless architecture, which means we basically will spin up the resource based on demand. You don't have to provision resources or Kubernetes and those kind of things.

An example, and Luke's going to drill into how this looks, actually is Headway. It is a rapidly growing mental health care system.

They were having to deal with increased demands from their employees. It was putting a lot of overhead on their IT team, things like handling service requests and request for provisioning.

They deployed an IT ticketing agent to autonomously handle IT tickets, the user integration, not built in tables, the RAG, okay, to ensure that the ALM is tightly integrated with their corporate data and corporate knowledge, natural language questions coming from their employees, and basically answers questions as well as create service tickets on demand, as I mentioned, directly plugged into Slack.

So one ID for every use case, AI infusion palette, easy to basically use AI services, call out a lot of language models, and the agent builder built directly into the platform.

So to find out the look, this is a kind of key takeaway. No need for separate silos for your agents. You can roll out all these agents, everything from marketing to sales, across finance, HR, service, and prospect self-service to employee self-service to IT ticketing, for example, on the IT side. We've had that single platform for building, integrating, orchestrating, connecting your agents, k, intelligence and the management, open perspective, tend to a broader range of LLMs, and also across your entire stack as well, which is, you know, really helpful to make sure you have a broad reach with your agents.

And with that, I can add a little bit of fun bit, which is Luke. Luke's gonna take you through a demo of how this looks. So over to you, Luke.

Awesome. Thanks, Paul. So let's jump into the demo here. So I'm gonna take you through a piece of functionality on the Trade dot AI platform known as our Merlin agent builder.

There'll also be several pieces of functionality on the platform to help you when you are building out your agents and corresponding in your agent tools as well. Now an agent will have access to multiple tools, and, generally, a tool is things that your agent can use in order to achieve a particular outcome based on the input provided. And so in this case, I've got an IT support agent that's got access to a few tools, things like being able to reference company policies and documents, being able to take action in systems like Okta and Jira and things like that. And so there's a wide variety of tools that your agent might have access to.

Now at the heart of Tray is a powerful low code workflow builder. Now what's really cool is that the agent tools are built using this same interface. So this provides both the power and flexibility for tool implementation whilst retaining the benefits of low code. So in this case, I have an agent that can do quite a wide variety of things, but each of these tools are a corresponding underlying tray workflow.

Now this first IT support agent that I've got is currently integrated into Slack as the main interactive point, but I'll also show you an example of one that's integrated into Intercom as well.

And when you are starting with an agent on tray, you will find these agent accelerators that you can make use of. These are starting points that you can leverage for common use cases associated with agents, like knowledge agents, IT ticketing agents, support ticket agents, and so forth. What's really cool as well is you'll be able to use templates for the tools that you're creating as well. So when you are adding new tools to increase your agents' functionality, there are some tools in our template library that you might be able to make use of to give you a good starting point, whether that's looking and interacting with Jira tickets, maybe looking in your CRM system, perhaps doing web searches.

There's a whole host of things, including an example of how to do that knowledge base lookup that we're using in the session here today. Now like I said, our first one is currently integrated into Slack. So let's head on over to Slack, and I'm gonna ask you a question of I have just received my new laptop.

What should I do? And so my agent's gonna go away and take that prompt. What's What's actually happening behind the scenes is we are making use of another piece of native tray functionality here, which is our vector tables. So what we've got is we've got a Google Drive folder that's actually got a ton of policies and process documents in there, including a laptop and device guide as well.

What we did is we've created a tray workflow that takes the information in those PDFs using things like our intelligent document processing and inserted that into a vector table, so effectively creating a rag pipeline with that. What that means is that I now have a tool that when there is questions specifically related around these policies, my agent has some information it can be grounded in to give it specific information relevant to the query based on the information that was in those policy documents. So if I head back on over to Slack, you can see that I've got a really nice formatted message here coming back from my agent, all of the initial setup steps that I'm able to do.

What's really cool is you'll also notice a reference to those two documents at the bottom there that I just mentioned, giving it some citations about where I was pulling that from. And you can see it's now grounded in that information that's available for me to use as well. So really cool, it's got that knowledge base available to it. I've given it information specific to the environment.

But let's take that a step further. So I've got an IT support thread here that my agent also listens into. And so I'm gonna say, hey. I am having issues accessing an application.

I see this. I'm actually gonna upload an image to go with this one. So it's currently a a problem with accessing, in this case, DocuSign, whereas an error message that I'm getting back. So I'm gonna give my agent this to work with.

And what's gonna happen here is under the hood is that there's a few extra things that are gonna come into play here. There's a few tools that we're gonna be making use of here that our agent will have access to. The first one is going to be around application assignments, but, also, we're gonna be interacting with Jira to potentially create a Jira access request for application access as well as being able to ground that in information around who the actual user is. So you'll see over on the left, I've got a whole set of tools that are gonna help my agent, and it's three or four tools that are gonna come into play for this particular use case.

And it's gonna be able to help it ground the response in that it needs. So let's take a look back at the response that it's come back with. And you can see in this case that it's saying, hey. I can see that you're having trouble accessing DocuSign.

Now just to highlight here, actually, I never mentioned DocuSign in the written text here. It actually abstracted that out from the image that I attached, which is really cool. But you also notice that it's giving me information around who generally has access. And so in this case, because there's a discrepancy between me and the engineering team and the teams that normally have access, I need to provide a business justification for accessing DocuSign.

So it's got some guardrails in there, and this is all because I've grounded it in information around application access in this case, so who can access what applications based on its role. And, again, that is another tool that is being able to be used by my agent to help it get to this response. So in this case, I'm gonna say, okay. As the IT support agent, I need access to sign in fact.

That's my business justification.

So now what my agent is gonna do is it's gonna take that response back, and what we should see happen is it's gonna be interacting with, in this case, Jira. So it doesn't have the capability in this scenario because I'm in the engineering team to automatically provision it because I've got some additional guardrails that I've placed around this to say that it needs that manager approval, and so it's going to create a ticket instead.

And so what's happening here is that this is an example of a tool that's going to be interacting with a third party system, in this case, Jira, to be able to create a ticket for that access request that I've just created. So it's not just having information that it can access. It can actually take action in systems as well. So you can see here I've submitted an access request for DocuSign.

I've now actually got a link, including reference to my manager for that as well, which is really cool. And so if we go ahead and refresh that, that was ITMS fifteen. It's the one coming through. So give that a quick refresh on Jira, and we should see that ticket being created for us in just a moment for ITSM fifteen there.

Oh, yeah. Perfect. Just popped through. So you can see that my agent has actually taken an action in Jira to create that ticket for me as well.

But let's actually take that a step further. What applications can I access?

So now what my agent is going to be doing is it's going to be using some information. So I've got in a native piece of tray functionality called data tables, our application assignment table. And these are the applications that typically someone in a certain team will have access to. Now I, as a user, are in the engineering team. And so it will have information in here around which applications I should be able to access by default. And if I do should have access to these by default, it's one that my agent is actually able to go ahead and provision for me if I don't have access already itself without needing to go through that process.

So let me go ahead and take a look at what the agent came back with. You can see that it's based on my role. So, again, it's got information about me as a user, as a junior systems engineer, and then actually the applications I can get access to. So in this case, let me ask it for, can I get access to GitHub cloud?

And so what we should see is that over on Okta, my Okta is looking pretty quiet at the moment for application access. But because I'm in the engineering team, I should have access or can get access to GitHub, and my agent has the capabilities to go ahead and provision that for me as well. So a similar example where we're taking an action in the system instead of creating a Jira ticket in this case, so we're gonna actually be interacting with Okta to provision the application in Okta directly. So it's come back.

You see that it's provisioned me GitHub access, which is great. If I go back on over to Okta and click on refresh, looks like I now have access to GitHub enterprise. So I've given it guardrails for scenarios where I shouldn't have access based on my role, but for those that I do, it's able to go ahead and actually provision that access as necessary as well. So that's really nice.

And one of the key things here is that this agent is currently exposed as an API endpoint. So when we look at the underlying workflow, it's an endpoint that I'm able to call. And so what we can also do is easily change the underlying infrastructure or the system that this agent is exposed in. So in this case, I've got a product on Shopify, and I've already got a intercom hooked up to it.

But my agent is now listening for intercom requests as well. So because it's an API endpoint, it's really easy for me to, you know, change that source system or where that agent is gonna live, whether that's just listening for API calls, whether that's embedded in certain applications. There's a lot of places we can use this agent. But in this case, what I'm gonna do is I'm gonna go on ahead and ask it some really specific information about a particular product that we've got.

So just with an intercom, I'm gonna say, hey. I want some specific information about your product and a piece of functionality related to that. Tell me some information on the mortar application for the advanced bricklayer machine.

Now what's really cool with this is just as before, what we've got sitting within Tray is under our files at the very bottom section here, we've got some rag sources. And what is in here is a couple of PDFs, which are our instruction manual specifically for these products. So exactly the same as before, what I've actually done is I've created a knowledge base with our vector tables with those product PDFs to be able to give us some grounded information that my agent is able to use. And so now when I look back at the response coming back, you can see that it's giving me some really specific information around that in order application.

I can also take that step further and say, hey. I want to return an order. And what it's able to do is also, in this case, interact with Shopify to be able to pull back that order information as necessary.

So you can see that with these tools and with Trade's extensive connectivity library, we're able to pull in information from multiple different places, whether that's unstructured data sources like the PDFs, whether that's interacting with your source systems. And we're also able to take action on these particular sections as well. You can see it's pulled back the return information for my order, which is great. You can see that it's giving me the actual order information from Shopify.

And in this case, it says policy not good enough. Let me just say why I wanna return it, and then one. So I'm gonna say, hey. This is what I wanna return.

In this case, it's gonna go into Shopify and initiate that return. So just before how we interacted with Jira, how we interacted with Okta. In this case, we're now interacting with Shopify with our agent, and it's able to take that request and action that accordingly based on its capabilities that we've given it as a tooling. So it's gone ahead.

It's created that. If we take a look at that underlying order, we can see that it's gone from fulfilled into return in progress.

So as we can see, Tray has great connectivity across applications, and this allows you to easily create tools that your agent can use to do a variety of tasks. Each of these tools is our low code workflow builder, making it easy to expand the available tools over time. Combining that with native features like vector tables, it provides an incredibly powerful platform for you to start building out your own agents on. Hopefully, that's given you some thoughts. Really look forward to potentially taking a look at some of the things that you build with our Merlin agent builder functionality. Thank you so much.

Great. Hey. Thanks for the demo, Luke. Hope you really enjoyed that. It's fun to see and fun to see all in action.

And with that, I'm gonna hand it back over to you, Vance.

Wow. Paul, Luke, great session. Really great overview of, new AI enabled architecture, let's call it, for getting up and started with both AI and AI agents. And who doesn't love a demo? Really fantastic demo. Thanks, guys, very much.

Yeah. Thank you, Vas.

Thank you so much.

Really excellent, guys. In fact, I'm glad to see that you left us some time. Got a couple of questions here, both high level and kind of into the technology itself. So with your permission, let's go to some questions.

That sounds great.

Awesome. First off, many different ways, many different folks ask this question. Where do you see the biggest and easiest agentic opportunities here in twenty twenty five? Frankly, as one questioner put it, with so much to learn, we don't wanna run down too many rabbit holes.

Yeah. That's a great question.

Where we are seeing a lot of demand actually reflects the survey I shared earlier on, which is especially on the IT service side of things, which is, you know, when you think about where IT spends a lot of their time, it's a lot blocking and tackling work, right, especially when you have a very large set of employees to support. I think, for example, basic just knowledge requests.

How do I configure my laptop? Or how do I work with Zendesk, for example? All the way through to provisioning requests as well. Can I please have access to HubSpot, for example?

We've had customers when they've sized it out, you know, their IT organization is spending twenty to thirty percent of their time, a lot of that highly manual to the high volume requests from across the business. So that's a huge opportunity for AgenTek, integrating it with your, you know, Jira and, you know, Workday, you know, Zendesk, those areas, all your sources of knowledge, and enabling employees to be much more autonomously engaging with an agent around that. And it's great for IT. Right?

It frees you up from that to be much more strategic.

Really important point there, Paul, and thanks for the great suggestions.

Relates to another question. This is a little more high level. But just to summarize all Trey's architectural thinking and features, let me read you this comment or question. It sounds like Trey has an integration focused AI architecture that lets me AI enable practically all my legacy assets, apps, data, even processes.

Is that a fair summary?

Yeah. Yeah. I would say that's a pretty fair summary. I think what that summary captures is, one platform approach.

Right? I think it's important to think about that. If you think about what different technology does arrive on the scene over the last two years, it's been a lot. Just twenty five plus years, and it's important now to start to be much more strategic around the investments you make and not end up with proliferation.

And so by taking a much more whatever one platform to rule them all approach, you know, doing that your classical backlog, but also take care of your energetic initiatives, that that's huge for developers. You don't want to get spread across five, six, seven IDEs or having to weigh the pros and cons of each. You can work in one system. And also the composability is also important as well because sometimes the logic you build in for integration, let's say, for example, you build a nasty integration to your ERP system, you can reuse that for your agent as well.

So composability enables you to the work you might create for as good integration with all transformations around that, you can also take that work, and you don't wanna end up having to reinvent the wheel for your agent initiative when you wanna build, for example, your agent that's handled auto management. You can reuse all that work, build that, and so that's the composability aspect. It's all about reuse. It also eliminates multiple points of maintenance.

Right? You change one once, and it changes everything from the orchestration through the agent you've built using it.

It's a really powerful vision. It's a really powerful delivery or implementation. Let's now drill into the special features of Merlin Intelligence and the new agent builder. Great. Awesome demo, by the way. Just quickly, the question here, guys, says, which features of all that we saw today do you think will be most useful by customers?

So I would say that the integration with the collaboration tools within Slack is hugely valuable. But beyond that, the amount of flexibility you can drive into the agent in Slack. And it was everything from let's say the ITSM sense. It's everything from including tickets to even reasoning around a justification around a resource request through to knowledge self-service.

So the amount of scope you can drop into that agent is delivered through Slack is wide and broad, and it can reach across all of your stack as well. The other area is just a prebuilt aspect. So the secret with middle agent builder is that they were providing you with a lot of templates. You know, as I mentioned a little with the tools templates we provide as well as the application of the fire round things in that ticketing and knowledge we provide, but you can customize it yourself.

If you decide you wanna add a new tool, let's say you wanna add screen scraping, for example, you can add that. And let's say your HR system, you wanna patch that in, you can use our connectivity to add, for example, Bamboo or using your SuccessFactors or in your Oracle HR or if you're doing NetSuite or maybe Sage, if you're ERP, you're not gonna hit a wall, and you're gonna map that agent in. And that's only gonna surface through that same, for example, Slack front end. The other key thing is that multichannel is important, that you might, for example, wanna decide that Slack is a great experience, but maybe you move to Teams.

Or maybe, for example, you wanna integrate into Notion or some of that. Right? That having the flexibility to publish its API and then consume that agent how you wanna consume it buys you a lot in terms of, ensuring that you can provide the best experience to employees and customers.

Wow, Paul. That was really great. I have a couple of use case questions here. It's almost like kids in a candy store. The more they learn about what you're doing, it seems like they've got a few little brainstorm ideas of what they'd like to do. So let's run through some of these pretty quickly, and then we'll wrap up.

You mentioned the idea of integration tools and templates. I think that might fit into almost all of these.

First one is with your integration tools, can Trey support new LLMs as they arrive? We're looking at DeepSeek right now.

Yeah. DeepSeek's a wild one. Right? So, yeah, that's actually one of the nice advantages of our platform is that, obviously, we provide out of the box connectors.

Again, as I mentioned earlier, OpenAI and, you know, better all can put here and those kind of things, so we brought back connectors. But we also provide a, you know, called a CDK, like, connect developer kit. And it is TypeScript based, and it also reads open API specification as well. So it's very, very quick to create a connector.

And then once you've mapped it in, it adds it to and you connect library.

And then you can start using that within your business processes. One of our design right away from day one with traits all by extensibility because we're very familiar with the fact that stacks change, especially even with the pre AI era, the cloud era. If you're offering running integration, you gotta be agile with change.

Really good point. And you're right. Deep seek is pretty exciting. Here's another exciting topic. And you mentioned it in your slides, Paul, hallucinations.

The question here says, some have mentioned the ability to access multiple models or multiple datasets can reduce hallucinations.

Does tray dot ai agree? And if yes, can your integrations help us avoid hallucination?

Yeah. One of the ways you really tighten up the responses is that the more integration the more data you integrate in RAC, it enables you to refine tune the results you're serving up. And what we've done with Trey is that, as I mentioned earlier, we provide our vector processing directly within the platform. And so you don't have to pull out to a separate vector database. For example, you can create the vector table directly within Shrey and then start to ingest the data and handle all of the embeddings around that. And so what that enables you to do is it makes you able to plot path to a strong rack based implementation and in turn helps with reducing the risk around hallucinations and also provides you with the agility to ingest more data and add in more data as needed as well. So that's a big part of the platform.

Paul, Luke, this has been awesome, mind bending even, and it was so interesting to see how extensible an integration platform that is now AI aware can really change the game for how we can not just build our AI agents, but also at the very beginning, design and do a POC. It's really been awesome. Thank you very much.

Yeah. Pleasure. Appreciate it. I hope everyone found it valuable.

Absolutely. Before we go, we always want to help our attendees get in touch with the technology and the services that our presenters provide. Give us some next steps that that you can suggest for how people can learn more about Trey, the platform, but also your terrific Merlin line of intelligent and agent builder technology.

So we actually provide a hands on workshop.

So it's our agent builder workshop. You join us the workshop, and we'll help you get set up with Trey. You can start from our home page, and we have our AI team run the workshops, and you can start building. So some of the things you saw today that Luke shared, you can start to see how it works yourself and have a chat and get beyond the demo and get right into it, which is what we wanna get to. Right?

That is so awesome. I know there's been a lot of talk about the AI Copilot, but, frankly, when you learn something new, there's nothing like having a human Copilot to help you. Right?

That's right.

Paul and Luke, this has been fantastic. Thank you very much, Paul Turner, integration specialist, and Luke Smith, sales engineer from Tray.ai. Fantastic session learning about the latest on AI agents and the whole AI enabled Tray platform using integration to help you bring AI to all your legacy assets. I just want to thank our speakers and thank our attendees for some really great questions.

Final note, Paul mentioned the training, which, of course, we'll have a link to here, but there's other assets from Tray we want to highlight to you. They're here right in the breakout room. Click away to any of those links, and you'll get them right away. And as you can tell, the innovation parade has just begun in 2025 at Tray.

Here's a slide that will take you to some more great drill down assets at Tray. Download the terrific slide deck this morning, and all these links will be live. So thanks again to our speakers and audience. Thanks again for really amazing questions.

Have a good day.