Kore.ai on Enterprise AI Adoption: Multi-Agent Orchestration, Bounded Autonomy, and the Shadow AI Problem

Most enterprises are still shoehorning AI into structures built for a different era, the horseless carriage problem. In this episode, Shawn is joined by Cobus Greyling, AI Evangelist at Kore.ai, and Carl Katz, VP of Global Technology Partners, for the show's first-ever dual-guest conversation. Together, they cover why multi-agent orchestration is outpacing prompt engineering as the critical enterprise skill, how to govern a workforce of AI agents before sprawl sets in, what the shadow AI economy is costing organizations in attribution and compliance, and how technology leaders can make the business case for AI experimentation without a clean ROI to show for it. If you're trying to move urgently but thoughtfully on AI, this one is for you.

What You'll Learn:

• Why multi-agent orchestration is becoming the skill that matters more than prompt engineering and what that shift means for your team
• What "bounded autonomy" is and why setting the right autonomy level per use case is the difference between a useful AI agent and a liability
• How the shadow AI economy works against your attribution, compliance, and AI governance efforts
• Why 80–85% of enterprise AI pilots fail in-house and what the ones that succeed do differently
• The horseless carriage trap: why plugging AI into existing processes without reimagining them is a recipe for wasted investment
• How to frame the AI experimentation business case for your board and finance team when clean ROI isn't available yet
• What an agent control plane is and why you'll need one before your AI agent count gets out of hand
• How to think about the build vs. buy vs. partner decision as agentic AI moves from pilot to production

Transcript

SPEAKER_02 0:12

Hi everyone and welcome to the Amplitude of Tech Podcast. I'm Sean Kordner, Chief Marketing Officer of Amplix. Today we had Cobas Grayling and Carl Katz on the podcast, our first double guest episode. Both of them are from Core AI. We talked about the horseless carriage problem of trying to shoehorn a new technology into existing structures and the tension between moving urgently but thoughtfully with AI. This was a great episode. Hope you enjoyed it as much as I did. All right, Cobus Grayling and Carl Katz. Thank you for joining the podcast today. Happy to have you. This is our first podcast where I've had two guests on, so I'm feeling a little bit outnumbered and unsure of myself. Feeling a little defensive. I'll try not to gang up on me too much, but I'm really looking forward to having this conversation with you guys. But I thought let's start off with some introductions. So maybe starting with you, Cobas.

SPEAKER_00 1:08

Thanks, Sean. So my name is Kubas Krilling. I'm an AI evangelist with Cor.ai. I'm South African, born and bred, still in South Africa. And so something I enjoy is just uh like continuously researching and writing on how things are unfolding in the space of AI.

SPEAKER_01 1:26

Excellent. And then Carl, how about you? Hey, Sean. Um happy to be here again. Carl Katz, the vice president of uh Global Technology Partners for Core.ai. Been here for over two and a half years, working with different types of partners globally to evangelize the core platform and ecosystem with partners. So yeah, that's me. A lot of evangelism happening over there at Core.

SPEAKER_02 1:53

It's important to evangelize when you're talking about AI, right? Yeah. Uh perceptive viewers of this watching on YouTube will see that my camera just changed up. This podcast has been plagued with technical problems. This is our second time trying to record, and Riverside had an outage, which is the platform that we use, and now my camera appears to be having an outage. So you can enjoy the view from my MacBook camera if you're watching on video. So apologies for that. So I thought maybe we could just kind of start off with people that might not be familiar with core, and you know, we don't like these podcasts to be a commercial per se, but if you could just give us the quick, you know, 30,000-foot view of what core AI does.

SPEAKER_01 2:30

All right. Well, I'll do that. So uh core to AI is a um is an ecosystem, basically. So we provide the businesses, uh, enterprise businesses with the cutting edge of AI products, including generative AI, agentic AI, and advanced AI automations and orchestrations. We're founded in uh 2014 uh by Raj Canaro. And I would like to say that we are the oldest startup in the industry. Uh and I say that because we're very nimble in our approach and we're always developing newer technologies. We work extensively with the customer and employee experiences and creating really great, empathetic, and human-like conversations that are well advanced, you know, for our customers. Uh so we have a global uh presence. We operate in 130 different languages and dialects, and we have um offices uh throughout the world.

SPEAKER_02 3:31

So that's some about core. I think your marketing department would be proud of that response, but uh what do you really do? Let's talk about just kind of some use cases that people would understand.

SPEAKER_01 3:41

Yeah, so we really focus on several verticals. We we do really well in financial services, banking, healthcare, uh, and retail. Uh those are core industries and core verticals. And uh, you know, we actually have uh really faster time to market than a lot of AI companies because we have accelerators within those verticals that allow us to, when we sell our services, allow us to turn service up quickly because we have extensive experience in those different verticals. So if you're selling into a bank, we already have the call flows, the integrations uh complete, and even the agentic flows already created and pre-created uh to facilitate a quick engagement and a quicker time to ROI. Uh so those industries are our primary industries. Of course, we work with other industries as well, but creating that customer experience um from the actual virtual assistant to the agent is what we uh accelerate at it. And so I would talk about really our agnostic ecosystem. I'll give that over to uh Kobas to talk a little bit more about.

SPEAKER_00 4:54

Great. Thanks, Carl. Yeah, Sean, uh yeah, just to maybe put it into like you know deep, you know, practical terms. So we we try and be uh plat we are platform agnostic, but also model agnostic. So I think what you're seeing currently is that um, you know, and that's something NVIDIA Carl, I was just speaking before we went on record about um NVIDIA GTC. And so something NVIDIA um is really, really advocating is multi-model orchestration. And recently, I don't know if it was on purpose or not, but OpenAI revealed the um you know the mechanics behind their deep research API and also Chat GPT. And you know, so for us, ChatGPT is just this very simple user interface, but what they actually revealed was that multiple models are orchestrated under the hood, and you know, multiple models play you know different roles. So that's something we've been focusing a lot on as well is model orchestration. And so not one model to run the whole environment. And to be able to do that, you need to give your customers access to open source models, obviously from a price point perspective, and there's also specialized models. So we've put an environment where you can have access to, I think, close to 200 open source models, but it's also integration to hugging phase that gives you access to potentially thousands of models, and that really gives you the freedom to build a highly granular um agentic environment where you can orchestrate these different models within a workflow. And I mean, we really see that as the way forward where you don't have one single model that solves for everything, but you have specific models that run, you know, specific places in this agentique workflow, if that makes sense.

SPEAKER_02 6:49

Yeah, that makes total sense. I I think it's more relevant now than it was a couple of weeks ago just because of the changes that Anthropics made with Claude and token usage, right? So they they made some changes there. I don't know if I can articulate exactly what they are, but what I do know is I'm hearing from people that all of a sudden they're hitting their weekly token limits in a day, whereas you know they weren't hitting those limits before. So I think it makes it more important to not be locked into any one model because it reminds us that the underlying economics of the companies that we're using in our AI solutions uh could change at any minute, and that could totally disrupt the economics of your model.

SPEAKER_00 7:27

Yeah, I mean, just on that, the other day I made a list just for myself on the models that's being deprecated. So if you look at the big model providers, there's a constant flow of models that's being deprecated. So, I mean, if you premise your application too much on the behavior of a specific model, then you're in danger. I think there's another dimension to that is something that's been documented as model drift, where models actually change under the hood. So the behavior of the API changes. And just to as a third point to what you said, Sean, about the token usage. I mean, like I use clawed code extensive extensively. I think Oak is 4.6, and I hit my my daily limit frequently, and then I hit my weekly limit, right? Um, before the end of the week. So those are all challenges, and I think people are really looking for alternatives. I mean, I've been looking at something like the NVIDIA DGX Spark, um, where you can run a model locally, and they are highly capable open source models you can run locally. So I think really, I mean, the not to go off on a tangent, but I think the new the newest buzzword is harness engineering, you know, um AI harness. And I think that's something we aspire to, to to be able to give that flexibility. I also think of it, uh I also like the term bounded autonomy, right? So, I mean, when Carl mentioned, you know, core comes from natural language understanding models, NLP models. There you had a problem of scarcity. You really had to be innovative to give like a natural conversation to your customers. And now you have the problem of abundance. You go we went from fast to feast, really, when it comes to the technology. And so I like this whole idea of harness engineering and bounded autonomy and try and control the workflows um and and and especially the user experience. And I think Carl alluded to it, but especially because we find ourselves in the financial sector and health services, so it's highly regulated environments. You need that level of uh when it comes to the applications, but just to book in that, definitely to have the the ability to swap out models and orchestrate multiple models as you choose to do.

SPEAKER_02 9:49

Well, I think you can also think about the outcome as being a driver for having multiple different models too, right? Because different models have different strengths and different weaknesses, right? So you really need to kind of understand what is the function that you need that model to play in the overall workflow of the AI. Also, each model has different, you know, costs, uh, cost implications, right? So something that's doing a lot of deep thinking that you know that you want to use that for, let's say, deep research and not necessarily for you know just rewriting um uh a paragraph in in one of the um deliverables or the outputs of the task that it was given. And there's cost implications, you know, depending on which type of model you're using, right? I I think that also kind of speaks to the skill set that's needed in what we could maybe call the next generation of AI, because you're right, large language model and generative AI, that came out of nowhere almost, it seems like, uh for most of us at least, and uh and it blew up, and it's almost like we got ahead of ourselves or out over our skis because that's not actually the end state of AI, right? Because it doesn't do anything, it tells you things, but it doesn't do anything for you. And so now I that you're seeing the adoption of agentic AI because that's where the rubber meets the road, that's where things can actually get done, and and maybe cost savings or productivity gains could actually be realized. And so it's gonna require someone to architect these things. If you're working in an enterprise that's building their own models or their own type of solution using different models, they need to know this. They need to know what are the available models and what are the strengths of those models and what are the economics of those models. And you need to have explainability and observability built into that to make sure that you don't have that model drift and that things are operating the way that they're supposed to be. And so that's a long way of me saying that. Um, and and I spoke to uh an AI expert that we've had on the podcast yesterday, and he was saying to me, from a skill set perspective, he thinks that too much emphasis was put on prompt engineering early, and actually where the skill set needs to evolve to is in orchestration.

SPEAKER_00 12:06

Yeah, that I mean, I think the orchestration layer is definitely something that's been um I think largely neglected. And it and I fully agree with that whole idea of um having too much focus on on prompt engineering um and and not all on all the different aspects that goes into building an enterprise um solution. But I just wanted to, just on on what you said about the skill set, Sean, like something I really find interesting is that you know people know what good AI looks like. You know, I I keep on thinking of that MIT piece that coined the phrase the shadow AI economy. And we're like the official AI tools that that's being supplied by people, supplied to to workers, they opt not to use that, but they choose to bring their own AI tools to work. And hence the shadow AI economy because you're not using tools that's you know officially supplied, but it's um the tools you choose to use. So I mean, I think there's a when you speak about um in the skill sets, there's definitely uh some kind of void there, but yet people know what good AI looks like, and they know the AI they they want, they choose to use. Um and then maybe secondly on that, um you know, I also think there's there's kind of like an attribution problem currently when it comes to enterprise AI. Um, you know, I was listening to an interview a while back and it from A6 on A16Z, and they spoke about how internet uh advertising had an attribution problem. Like how do you attribute user behavior in terms of clicks to your actual advertising? Um and the moment they got the attribution right for online advertising, the the business model was there and advertising could be sold. And then they equated that to the shadow AI economy that's happening currently, and AI use are not really attributed accurately when it comes to enterprises. So that that's something I I found interesting. Uh just the whole idea of people they've got set ideas on how they want to use um how they want to use AI. And the tools have been they've been given officially aren't they you know the tools that they'll they choose the tools of um that they prefer.

SPEAKER_02 14:34

So what you're saying is that businesses don't necessarily know which AI is producing whatever outcome or result that they're seeing in the business, right? So in the ad attribution model that you're talking about, there's different standards that you can use to try to figure out where your revenue is coming from so that you can attribute them to the right channel because that lets you know where you should continue to invest. That's how you make the ROI statement. And so the the ongoing challenge in the advertising marketing space is um we don't really know. There's no science to this. There's the first click uh model, there's the last click model. Uh, what we do know is that when people make a buying decision, um, it takes something like uh 14 or 17 uh interactions with that brand. There's a lot of things that go into it. So where do they hear about you first? What are the subsequent 17 touch points? Which one of them pushed them over the threshold and got them to buy? There's probably not a clean answer. It's the aggregate of all of those interactions that they've had with the brand and with the advertising. Um, and and so I guess the the challenge in the AI, enterprise AI space is you may have rolled out a handful of tools to your team that are approved, and then you have your team bringing their own AI uh tools that you have no visibility to, and they're using some combination of the approved and the unknown tools, and who the hell knows what they're using more and what they're using it for. And uh but what you do know is that you're either seeing uh productivity increase or you're seeing uh costs getting lowered, or you're not seeing anything. It really kind of leaves a variable in the equation that leaves you kind of you know sitting on your hands, like okay, now what?

SPEAKER_01 16:26

Yeah, I just want to add different I just want to add from my experience. Uh yeah, businesses are in different phases of their AI journey, right? They they uh you know uh you have some businesses that are actually considering AI for the first time and implementing uh AI solutions, even though they're in, you know, and they could be in highly regulated environments with compliance concerns that by the way we can meet globally from that perspective. Uh but you have other companies that are really advanced in their AI journey and that are actually initiating AI and creating even like small language models to facilitate their engagement as opposed to you know leveraging the large language models that might be a little too obtuse for what they require. So, you know, the benefit of the core.ai orchestration, you know, an automation platform is that we can facilitate multiple large language models. We're agnostic, as mentioned. We can facilitate multiple large language models into our ecosystem. And depending on the type of query within a conversation, we can pivot. So if you want to use, for example, you talk about token costs and such, you know, from like, for example, healthcare loves Azure, you know, they love the Azure LL LLM. Then you have others, then you have a query that might be something on a website or some other Excel spreadsheets somewhere else, but you might want to use Deep Seek, right? Because the token cost is significantly lower. So we can we can facilitate that type of engagement on our platform to where token cost is not really an issue, um uh as much as let's say just leveraging Chat GPT or OpenAI 5.2, whatever it is, uh, for all your queries, then then it gets a little outrageous, honestly. It gets a little nuts there. So we can work with you, uh the your the custom your customers to create uh an overall AI strategy and mitigate costs for token usage uh utilizing several elements of our platform.

SPEAKER_02 18:31

Yeah, I think um that brings me back to the buy versus build conversation, right? Because I think a couple of years ago that was the entry point was buying, and we saw it mostly in the CX space. It was low-hanging fruit for business to well, to back up, you know, like you were in a situation a couple of years ago where boards were saying, we need AI, go buy us AI, right? And uh nobody knows what that means, least of all them. And so that's I think what made it uh attractive to go the buy route because you don't really have to know everything there. Uh there's probably less risk by doing that as well, but it satisfies that uh that mandate to buy AI or to do AI, as uh as the uninformed would say. But I think what we're seeing now is pilots are starting to get built, and most of it is agentic, and most of it is around workflows supporting processes that already exist. So, you know, I I wonder a platform like yours, or maybe not yours specifically, but just in general, that kind of uh, you know, you're I'm I'm paying for a platform that has these agentic workflows kind of pre-built. Is that a is is that a half-step or an evolution towards the build? I guess I don't I don't know personally, tactically, what it means to leverage a platform that already has agentic AI workflows built. I imagine that there's some level of customization there. So are the people that are your clients or customers, are they getting more hands-on experience? Are they learning something in that process that can then be applied for them to build in the future?

SPEAKER_01 20:13

Yeah, that it's so much, it's so complex. And I I don't know if you're aware of this, but 80 to 85% of all pilots fail within the enterprise business space. That is a that is a true number. And the reason they fail is because they, in many cases, they don't know what their requirements are per se. And when you get involved with the agentic flows and stuff, it never never quite meets uh requirements. And then, you know, most of these companies that build their own are not AI companies, obviously. They do something else. But at its core, our platform brings together multi-agent orchestration with a full spectrum of development tools that businesses just can't allocate themselves. I mean, we leverage no-code, low-code, and pro code. So teams across the organization can build deploy AI agents at scale. Core.ai's ecosystem that's powered by advanced search, data AI, and also robust engineer AI engineering tools and really full-sack observability all throughout the journey to monitor and optimize the performance. This is not something that we can um that organizations can readily leverage, right? Especially from an enterprise perspective, because this needs to be built out. That's why being a 12-year-old company with this type of, you know, with this type of technology and the advancement of technology, advancement of the technology truly differentiates us from other AI companies or organizations that are trying to build it on their own. We have one airline customer that try to do that. I won't say who they are. I have an airline customer that try to do that on their own. They were seeing uh there were no guardrails on the conversation, which is an issue, by the way, obviously, hallucinations and other types of issues, a larger the large language. Model of the more latency associated with that language model, right? And also the more opportunity for hallucination within the model as well. So, like we have an airline to try to do that, and they say, hey, we're using open AI. COBIS knows who I'm talking about. We're using OpenAI, we're doing a great job, you know. We're an airline, we're going to build this great system. And guess what? They weren't giving bereavement fares, the rates were wrong, the experience was bad, it just wasn't a great experience, and they got sued, actually. Uh, they lost, and now they are a happy core.ai customer. So there's an example of where, you know, we're I guess working together, right, with businesses to understand and then bring them on to our into our ecosystem is the game changer for them. And that's where we accelerate.

SPEAKER_00 22:47

Sean, if if I can just add to what Cole just said, um, you know, in you, Sean, you mentioned the build versus buy, um, you know, part of it. Um, so definitely something that exists in the market is like this disconnect between marketing and what's happening on the ground. Like you spoke about when the rubber hits the hits the road. Like, so there's definitely that disconnect. There's this marketing hype technology that exists, but what's really happening on the ground? And if you could look on archive, there's a whole new genre of studies that looked at what developers are experiencing, you know, in building agentec systems. What are the bigger, biggest pain points? So they went to they use platforms like Reddit, Substack, which is still being used, uh, GitHub, just to quantify and categorize what developers are struggling with. And the biggest impediment for developers they found when rolling out technology is technology churn. They just need to run constant updates, patches. The technology churn keeps them busy all the time. Um, and apart from that, there's lots of granular, especially when it comes to search and vector databases, they struggle with. And this these studies just pointed to one fact, and that is that it is very hard to keep up with technology at scale and together with that deliver solutions. And you know, when when Carl spoke about this environment we're building, I really see it as this abstraction layer. So we try and abstract away all that technical overhead and all that technical PT and have that under the hood and just surface AI primitives with which people can build uh solutions with. You know, so I mean there are companies that want to tinker. Uh we one of our biggest customers uh I really like building good prototypes in-house, and that's a good way of getting a grasp on technology and a complete understanding of what's happening. But you know, these studies just pointed to the fact that you need to abstract away that that overhead of technology churn and hand that over to someone that focuses on that and then have the the no-code primitives or the abstraction layer you can make use of to build the solutions. Um I think it's it's becoming near impossible for a non-tech company to to keep up with the technology churn, you know, and always have access to the latest technology that's available.

SPEAKER_02 25:34

Yeah, I think I have a sense that um for the most part we're still in the very nascent stages of this for enterprise adoption, right? And it's just getting started. Carl cited a stat, I I think you're probably alluding to the MIT research um study that they put out, and I think it was 95% of pilots fail to produce an ROI. But that's a contentious study because I I think what it fails to take into consideration is a lot of those pilots weren't intended to produce an ROI. They were people learning, and so it was it was an experiment and it was an exercise in wrapping their hands and head around what was going on. But I've seen some demos of some people that are leveraging agentic AI in a more advanced way than I think your average, certainly your average individual, let alone your average enterprise, uh, would be able to do. And and what I've actually seen and demoed is you know, just a guy building this in his basement on a Mac Mini, and he's got a uh army of agents that are writing code for him, right? So he just um he doesn't even type, he talks to Claude, and Claude kind of orchestrates it, and it works with hundreds of other different models, and he says, Um, okay, what I want to do is this, this, and this, this is the objective, this is the data that I want you to access. And it just went through, and I think you you called it uh bounded autonomy. I'm pretty sure this is what you're referring to. So he's got guardrails built around the whole thing so that as it goes through each step in the process, it tells him where he's at in the process. He's got a pane of glass that shows what the AI is doing and why it's doing it, and then he's got another pane that is logging it so he can explain it later if he has to, right? And it goes through each of these steps until it produces code, tests code, and commits code. And then he is basically done, and then he's got human and loop uh in each of the steps so that he gets an opportunity to review and approve or send it back uh for revisions if he wants to. And so what I thought was interesting about that is this he can tell it to integrate with business applications. He doesn't even have to do the integration, he can tell it to integrate with Salesforce or an ERP or something like that, right? So what he said to me is I can see a time in the future where software as a service, as we know it today, dies because it's it's not the portal, it's not the pane of glass that you really need to access. It's not the user experience of that software, it is the data that that software houses, and that data becomes a piece of a broader picture of a you know an agentic workflow. And it could be interchangeable in that you migrate from one CRM to the next CRM. There's no real learning curve because people aren't engaging with the data that way, right? So then you don't have vendor lock-in. Like we're we're a HubSpot shop. I can't leave HubSpot no matter what they do to me. And believe me, they try to make me leave HubSpot, but I can't because where am I going to go? And I'm gonna have a learning curve for my team to get up to speed on a new platform. And so the vendor lock-in becomes less of a thing if it's not about, you know, our producer of this podcast is Brandon. It's not about Brendan learning how to use HubSpot or learning how to use a HubSpot competitor. It's about just plugging that data and that functionality into our agentic AI workflows, right? So, you know, I don't think SaaS is dying overnight, but what does it look like in 10 years? I don't know. What do you take of that?

SPEAKER_00 29:20

Yeah, so Sean, I I've put I've put a I've got uh this construct in my mind. I mean we spoke about the future. So three sides to this. The one is like software is expensive to create traditionally or to produce. So software had to be durable. You had to license it, you you've got to have like future proof and on all the rest of it. Now, all of a sudden, software is becoming ephemeral, like that guy in the basement. You know, I use clawed code to code all the time now. It's just in CLI, right? So even the IDE is being being abstracted away. Like you didn't have a IDE anymore, VS code, you just do it in the CLI. So that's one side of it. Software is less durable, it's more ephemeral, it's generated almost as a utility when you need it, and then it can be discarded. So that's one side of it. I think when it comes to the enterprise and implementing software in the enterprise, there are two sides. So the one is the more recent, you know, it's closer to where we where we are currently, and that is trying to just give um employees to give workers, knowledge workers, AI tools. So I'll give you an example, and I think Carl knows this much better than what I do. But I was recently at a Gartner IT symposium and and a CIO came to our booth and he said he's in shock seeing how many people are working on their laptops during presentations, uploading docs to Chat GPT, to Cloud. You know, so a lot of our customers don't have access within their day-to-day you know, working environment on their laptops to AI because for obvious reasons they upload docs and stuff. So a big part of our business is giving enterprise customers alternatives. You know, so you know, we call it AI for work, so it's an alternative to Chat GPT or to Claude. You know, it's it's uh all the data governance, model sovereignty, data sovereignty, data flows managed. So that's one part of it. So you're currently within the current organizational structures, you know, Spanish bank recently, one of our customers took away all their all these AI tools because of the risk. And then you've you've got to give the workers similar tools or better that complies to all these things. So that's one side of it, like very granular, giving people these specific tools. But then, Sean, there's another side that I that fascinates me. So currently AI is augmenting enterprises, but somehow we'll have to get to the stage where enterprise is transformed. And so recently there was an interview with Jack Dorsey, um, you know, the founder of X Twitter, and Rulolf Buetter from Sequoia, you know, they spoke about the enterprise of the future, the AI enterprise of the future. And the best way for me to articulate that is, you know, when we used to have horse-drawn carriages and the internal combustion engine was developed, you know, this engine replaced the one or two horses. But the engine was just placed within the carriage, right? And it was used to propel the carriage. There was opportunity to redesign the carriage completely because the carriage is built around the horses. But it was just seen as a horseless carriage, and the internal combustion engine was just slapped onto the carriage and off they went. And it took, you know, quite a while for the actual carriage to develop and change around the engine. You know, the engine became an integrated part. And I think AI, when it comes to enterprise AI, it's very much in that state where you know Jack Dorsey speaks about enterprises hierarchies developed because we had to pass data up and down and filter data and manage data and manage people. And so that's why we have the enterprise structures we have. And so we're at this early stage now where we're plugging AI in, augmenting humans, but it's got to get to a stage where um a company or an organization is completely transformed in the way it uses AI. And I think that's a state, I'm not exactly sure what that will look like, but that's a state we we need to get at. Um, and they, I mean, they created a manifesto. It's actually a very good uh interview to listen to, and how they want to completely reshape block into this um, you know, AI-oriented enterprise. So I think, you know, in that sense, AI is gonna look much different than what what we know it today. Agentic AI, I always say agentic points to agency, a level of agency and freedom to act on your behalf, obviously within the guardrails you've put down. So I think that's gonna be very exciting. And like you mentioned it, Sean, like we don't know even what's gonna happen in the near future. But I think that's the exciting part where enterprises will be transformed. I think that the problem, a big part of that problem is distribution. Like you you mentioned HypeSpot, you know, you're stuck in HypSpot. If you think of other systems of record, they've got the distribution and the presence within the organizations. So they've got that advantage. Uh software as a service, people say it's dead, but if if you've got distribution, then that's you know, it's gonna be hard to remove that. And maybe just to book in this thought, I mean, that's why I think uh, you know, Elon Musk bought Twitter distribution, you know, to have that distribution, the distribution for open AI is ChatGPT. And obviously Microsoft has got distribution. So I think that traditional distribution will help you to get your AI in. But I'm very curious to see how this future will look like where enterprises are truly transformed and we don't sit with this horseless carriage scenario.

SPEAKER_02 35:39

Yeah, I love that topic. That's um it's a hard one to have, I think, when you're a technology leader who's trying to keep the lights on and and keep up with the day-to-day, and and do you really have the time to sit back and reimagine what the enterprise would look like if you could start with a greenfield, right? But it's probably important to dedicate some time to think about it, at least in uh specific ways. And like I'll give you an example. I right before this podcast recording, I was on our executive call. We have a weekly executive call, and we have some AI initiatives that are in the pipeline. And our CEO said, just remember, we're not trying to be bad faster. We're trying to be better. And that really resonated with me. He's got away with words. He's a clean-speaking Boston guy. But I love that, right? Because don't just put AI into the process as it exists today. Evaluate the process and understand if there's a better way to do the process because this is your opportunity. You don't want to inject automation that comes with some risk, right, into a process that's already bad or not creating the desired results. But you said something uh that I noted. You said you have the they had the opportunity to build the carriage around the engine once the engine was invented. And I think I would argue that it's not about the opportunity, it's about the necessity. Right? People aren't going to do anything, they're not gonna change unless the pain of staying the same outweighs the pain of that change. So when I'm speculating here, but when engines got so powerful, or at least they had the ability to be powerful, that the carriage started to break down around it, or when the thing ran into a pole or something and it just disintegrated, the pain of staying the same design as a carriage started to exceed the pain of designing a better carriage to go around the engine. I don't know. What do you think? What's going to be the catalyst for that level of change in your average enterprise?

SPEAKER_00 37:48

I don't know if Carl's got some thoughts, but maybe I can kick it off. Like, you know, I I think necessity or pain. I mean, I'm I'm thinking of a few years back I read Lou Gersner's book who says elephants can't dance. And you know, when he took over as the CEO of IBM, IBM was done. Like, I mean, it was written, it was like a fact that IBM will stop existing. And uh then the only solution he had was to re-engineer the whole business to remake IBM. And then he said he had to re-engineer all the processes to save the company. And then he said one thing he said in the book was like, you know, re-engineering the processes of IBM was like setting your hair on fire and putting it out with a hammer, you know, what is what what is worse. So so I guess like there's a I guess there's a there's an aspect of necessity. Uh I mean like, you know, again, thinking of that interview with Jack Dorsey, um, I think, you know, block can be like a blueprint of how to really rapidly change an organization and and having have have it managed with AI in in the center. I think most other companies will have an incremental change um in in the way they do things. Um, you know, when it comes to employing AI. But I mean, I I think there's people need to realize the power. I mean, I I mean, just recently, full disclosure, just recently, I started coding with with Claude Code. I was using Grok. I mean, I was just like blown away by the way Claude CLI can orchestrate my MacBook. I mean, the way it can log in, go to places. So I think you need that realization of what's the actual power. Um, and I mean, like, currently, like everything is measured in tokens. You mentioned it, like, there's there's a cost um in involved in everything. So currently, like, I mean, we didn't talk about electricity, we talk about utilities at home. So currently we just talk about AI and tokens, but that's gotta be, we've got to get to a stage where that's abstracted away, and we speak about functionality and tools we've built, you know. So that's like a next step we need to get to. But I I think some companies will, to your point, Sean, about necessity and pain. I think some companies like Block, uh, Jack Dorsey, it's just gonna like like really just go like uh extreme. But I guess most organizations will will have more like a stepwise gradual approach in transforming.

SPEAKER_02 40:34

So, Carl, in science, have you ever seen the movie Contact?

SPEAKER_01 40:40

Yeah, yeah. Jodie Foster?

SPEAKER_02 40:41

Yeah. So I don't know if you remember the specific scene, but there's uh the head of science, I don't know. I can't remember what his title was, but he's the head guy, right? Drumlin, I think was his name, and and he is talking to Jodie Foster, and and he's going on this diatribe saying that science should be about producing results. We shouldn't be spending money on uh we shouldn't be spending public money on scientific efforts that aren't going to yield a measurable good to the public that are funding it, right? And she says, oh what? So there's no room for pure experimentation in science anymore. And I think if you know anything about the history of science and the scientific method, you really do need to sometimes just experiment for the sake of experimenting. And then it's the aggregate knowledge of all of science that leads to new innovations. That's why we government funds DARPA, right? It's it's it's money that they spend to, you know, to do the moonshot experimentation, right? And that eventually does trickle down into the economy. So where I'm going with this is Carl, do you do you think that how do you think a technology leader can go to the finance department, can go to the board and say, I need money to play around with AI. I don't think there's going to be an ROI. As a matter of fact, I'm quite certain there won't be. I'm quite certain we're gonna fail. I'm quite certain that we're gonna make mistakes that cost us even more money than I'm telling that you that I need for this, but we need to do it anyway. How do you sell that?

SPEAKER_01 42:14

Yeah, I mean, part of it is keeping up with the Joneses, right? Um, your competitors are all getting involved in AI and offering um a more consistent experience. Things actually uh have evolved. It interests interestingly enough, you know, we started with onshore contact centers, for example, then we went near shore, then we went offshore, right? Complete. And then now things are pulling back in and going AI and then onshore, right? For those calls that can't be handled by AI. So really discussing the entire customer journey and how effective automation orchestration can be in that journey, right? With a more consistent result, right? And more consistent. At the end of the day, it's all about customer engagement, customer sentiment. So talking to them about their NPR scores, right? Talking to them about their uh their challenges from a customer engagement perspective, you know, obviously ROI is always involved in the conversation, right? And and costs, you know, associated with that. So I hate to say this, right? And we talk about human in the loop, but companies are looking to, and we're seeing that with BPOs even, right? Where BPOs are actually, because I manage the BPO uh organizations globally as well. We're seeing BPOs going into the AI module, right? They're leveraging core and other companies to facilitate that because they realize they need to get on board with that, not only from a cost structure, but for consistency, net promoter scores, and also um uh, you know, and also from a ROI perspective, right? So um it's really easy conversation. To have because everybody, as you mentioned before, everybody's hearing about AI, and different organizations are in different sections, uh, you know, are different are different parts of their AI journey, and they want to know more, but really they want to know what the other guy's doing at the end of the day. How is the other bank doing this? How is the other insurance agencies doing this? How is other healthcare institutions leveraging AI? So they want to learn and they want to implement. And in many cases, it's a land and expand methodology, right? Where I'm like, you know, I mean, uh, I'm not gonna go too much about it, but but at the end of the day, we are we are agnostic. So, you know, somebody says, I got Watson X, okay, great. Well, you know, we have a more agnostic platform. So we can start there and work in conjunction with Watson X, or or you know, I only want to implement sometimes if you want to like, you know, take a bite before you eat the whole cake, and you say, Okay, I want to implement this segment of AI. And it could be like the low-hanging fruit, like just a chat virtual assistant, and then they want to move into advanced agentic applications. We're not seeing companies that are starting with AI looking, going right and diving into the deep end of the pool and saying, you know what? I want all channels, I want these agentic flows, right? Here are my orchestration agents that I want to be implemented, right? Here are the flows that I want to implement within the orchestration. We're not seeing that too often. We're trying, but that from my perspective, being in the sales side, that increases the cycle significantly. We find that if you give them something, and believe it or not, a lot of times before the sale is made, we're actually going deeper into the conversation than they believe they would go. So you're right, there is pushback. But once the conversation started, okay, and they understand the benefits associated with it, with all the things I mentioned, it's an easier conversation to have, especially today. Everybody's using AI every day for something.

SPEAKER_02 46:21

So the the two things that I I pulled out of that the two themes that I think your response circled around, if not hit directly on, is we can't get left behind. And so if we don't get in the game now, we are going to get left behind because everybody, and that is powerful, and I think that's actually a that's a reason enough to get in the game and start spending experimenting and spending some money because this we don't know what's going to happen with AI next week, let alone two years from right from now, right? So there could be a leapfrog moment that happens here that is predicated on having a certain skill set, having certain governance frameworks, having certain education and and transparency built into your enterprise to be able to leverage that next technology when that moment happens. So if you're not accumulating that intellectual property or that intellectual capital, I mean now, it's too late. Like, what's that saying? Like, don't dig the well when you're already thirsty, dig it before you're thirsty, right? So now's the time to start digging. But then I think you you hit on something that's that's actually equally powerful, which is you as a technology leader have the opportunity to frame this conversation with your executive team and with the board, right? So start framing it around soft metrics that tell the story, develop the narrative that you need it to develop, not in a deceptive way, not in a manipulative way, but in a way that you're quantifying the soft, immeasurable value to the business and starting to report on that, right? Because ROI doesn't always have to mean dollars and cents. It does if you're talking to the finance department. But if you if you get to a point where your CSAT scores are higher because you've implemented AI in a in a really smart way, right? Or if you've built this skill set that allows you to eventually implement a gentic AI workflows in aspects of the business that do reduce overall costs, then you do have that ROI story to tell. But you can't get to that without having that understanding first. Alternatively, you could go and hire a CAIO or a CDO or whatever, somebody that's going to head up the AI initiatives in your business and own that, but that's a pretty expensive proposition in itself. And I think we're in a situation right now where it's like in cybersecurity, where you've got more demand for people with the skill sets than you have supply of people with the skill sets. And so those people that really do understand it are a significant expense if you can find them.

SPEAKER_01 48:55

Yeah. I mean, just a few years ago, we were talking about robotic process automation, right? That wasn't too long ago. I mean, natural language processing, understanding, moving into the language, all large language models and generative IT and now agentic IT, agentic AI. And very soon we're gonna be talking about artificial general intelligence, right? That's gonna be the next iteration, right? And that's coming out within a couple of years. And, you know, we're we're developing that as well as an organization. So, you know, you also want to be on the latest and greatest. You don't want to be at the back on the back end of the curve. You know, you want to be forward. And working with an organization, whether it be core or somebody else, hopefully core, um, you know, it is gonna allow you to, you know, to get involved in an ecosystem that's constantly evolving and constantly changing to benefit the customer experience. I mean, there's a lot of I mean, you know, we have hundreds of engineers working on our product every day to maintain and advance it. Hundreds of them. We're an AI company. What businesses are telling me when I talk to them is we don't want to be an AI company. I did meet somebody who was very astute and building his own AI platform, very large company, um, Fortune 10 company. I was speaking to them and they have their own AI division and they're doing their own agentic flows, and that's what this guy lived for, et cetera, et cetera. So you know, he thought he was doing a good job, but going back to the MIT study, okay, I think it was like 67% of all in-house builds pilots failed. While I think 30% of the uh the other 30% that were external did well or something like that. So if you leverage an external source, a company who specializes in AI, you're more apt to have a favorable experience than creating something on your own. And not only that, it's just it's future-proofing. You talk about compliance and you know, and and you know, just building your own ecosystem and such, it's not always forthright. So you want to keep track of the compliance changes within HIPAA, within GDPR, within these different uh compliance rules that are always evolving. I know Cob COBIS always posts something about the evolving metrics of compliance. I see that online, and that's always changing as well. So you don't want to be caught with not being compliant because there's some significant downside to that, obviously. So I think that I hope that answers the question long-winded.

SPEAKER_02 51:34

No, I I think it answers it as well as you can, right? I mean, there's no I have a lot of empathy for people in this situation because this is the constant fight that a CMO has over spending money on brand initiatives. What's the ROI on your branding efforts? I don't know. Nobody knows. I was on a Gardner webinar over the summer, and the topic was you know selling brand initiatives to the board. And I thought, wow, maybe they have an answer. Maybe they're gonna tell me how to show an ROI on my brand spend. And they didn't bring it up. And so at the end I asked the question and they're like, we don't know. That's Gardner. So no, like, no, nobody knows. You can't you can't quantify, it's it's very, very difficult to quantify return on investment with the brand. And I I think like Cobus had said earlier about good AI, you know, you know when your brand investment is working, when you see it working, but you can't really you can't really put your finger on it, right? So, but that means that you I mean that doesn't mean that you can't do it. And so I think that's the situation that a lot of the technology leaders are in right now. And and the the buy versus build conversation, I think, comes back into this because you can buy, you can find a partner. You know, finding a partner doesn't necessarily imply buy either. You can find a partner that will build along with you, and then that can be a transfer of knowledge as well. But I think the message is uh you need to get in the game. Well, COVID's someone sitting there listening to this and saying, okay, I need to get in the game. Would you say that most enterprises today are from an architecture perspective and from a data perspective and from a governance perspective, ready for this, especially for agentic AI? And if the answer uh I'm quite sure is going to be no, what can they do?

SPEAKER_00 53:25

Yeah, I was I love the build versus buy discussion. I was recently at one of our biggest customers, and I I asked them, asked someone there, what's your view on build versus buy? And they said that they they buy for table stakes and then they try and build for differentiation. And what was interesting, they didn't get to that, right? So that they didn't get to the build for differentiation. So what they do is they've got like an incubator, like a, you know, you know, people build prototypes and um, you know, systems in-house, and then once a year they've got an AI day, and people showcase what they've built and what they've learned. And I really liked that approach um because they've they've bought a platform, they are actively implementing you know AI solutions, but also they're building their own understanding. And I think a big part of like advancing AI in enterprises if people have a very good understanding of technology and how things um fit together. Like we we like to speak about value engineering and so so proving value, it's not always not always that easy. So we you know, we've got people that that look at ways of how value can be calculated and represented. So I think there's a big part of value engineering that helps with this problem of attribution. But um, you know, I it's great to have conversations with people that try to build something. The best way to build knowledge is to try and build something. You know, if people try that, they get a better understanding of how the technology fits together. That really helps a lot. I'm very apprehensive when people's understanding is too opaque, there's not a clear understanding of what they want to achieve. I'll give you an example. So a while back, I was speaking to a CIO of a big leisure company, and he said to me that he's made three mistakes in acquiring AI software, like three failures. And he said, like, so now he's speaking to that. We were just having a conversation, and he said to me, Well, I don't want to make that mistake again, so I don't want to be tied to technology. And I said to the team, like, you have to. You've got it is that I don't want to be tied to a framework. And I said, You have to. Like, I can name dozens of frameworks, different programming languages, different models. You've got to be tied to technology to have that solution. So it's almost like this fight or flight, you know, you went from one extreme to another extreme where you know it's just I mean, you you can't you can't not align with technology. So I think, you know, enterprises, if they build their own understanding of of what they would like to achieve, then um that really goes a long way. Um you know, in in achieving success. Uh if it's too opaque, I think if you just throw money at the problem, I think that's that's really um that's a recipe for disaster.

SPEAKER_02 56:34

So I I agree. I think um the best way to learn something is to do it and just learn on the job. I started cooking professionally at 14 years old, and I eventually went to culinary school, but I had already been cooking for four years at that point. I learned because I was a dishwasher and I was looking over their shoulders, and they would let me start to prep things, and then they would let me, you know, mess around on the fry machine and then the saute, and and I I learned a lot faster than I would have if I had just started in culinary school. So I think you you start and you learn as you go. We had a AI expert on uh the podcast a couple episodes ago, and I asked him question if you're going to start tinkering, if you're gonna start dabbling, if you're gonna develop a pilot, do you treat it like a true experiment in that you develop a hypothesis and then you develop your parameters and then you test the hypothesis and then you evaluate your results? And so, you know, you go into it with an idea of what you think you're going to be able to do and what you think the outcome is going to be, and you have some sort of KPI that you're measuring on, you know, from one end to the other end, treat it like a science experiment. And he said that felt uh burdensome and probably unnecessary and to just get started. So fast forward a little bit, and I I had I had my team work on a list of priorities for us to implement AI, a way to improve our productivity in the marketing department here. So we all individually developed our use cases. We we started to look at what are the repetitive tasks that we do, what are the things that are time consuming but not valuable and not necessarily accretive to our department. And then we looked at the commonality of them, and then we did some chunking and turned those into big initiatives, and then we prioritized those initiatives against risk and investment and dependencies. What can we do without the IT department or the security people, you know, so that we could be autonomous. And we kind of stack ranked them. And so then I developed this list and I sent it to a different AI expert, and he said, This is great, but what you're missing is you're not treating like this this like an experiment. You need to have a hypothesis. And he basically went through the same thing, right? So I'm getting conflicting answers here, and the answer is probably somewhere in the middle because I can tell the two different personality types of the individuals I was talking to. But I don't know, what are your thoughts? What's the appropriate level of managing your experimentation versus just getting your hands dirty?

SPEAKER_01 59:11

I think, Cobas, that's probably more out for you, but I mean, um, you know, from my perspective, you know, experimentation, you really you need to have all the right tools and all uh you have to have the right people, you have to have the right consultants, the right business partners, right, before you consider any any type of experimentation because you don't want to get a suboptimal result, right? We do have a lot of um companies that have tried their own experimentation and play around with it, and then what they what they get is basic usage. Kind of like somebody coming in to use uh ChatGPT just to write letters or something, you know? How deep can we go down this rabbit hole to increase productivity and and help you gain a competitive edge and a more consistent customer experience? Why experiment when you actually have businesses out there that are leveraging the technology already in many of the same use cases you have in your vertical? And uh, you know, because when we discuss experimentation, a lot of times if people are doing this on their own with their own platform or trying to create your own platform, then it's more than experimentation. Then it's getting the right tools, getting the right, you know, uh you know, creating the your own ecosystem. Right? And then and then only then if when you have the right ecosystem with a com right orchestration and right compliances and and then you put in your use cases, only then can you actually have an accurate, you know, an accurate uh experience, right? Why do all that? And that's that's what Kobas was mentioning, you know, as opposed to buy birth versus build. That's the advantage of dealing with um with working with uh a company who's been around for a while and understands the um you know the use cases and what other companies are doing and say, have you tried this? Have you tried that? And you know, they're gonna build something and they're gonna say, Well, you know, I didn't have that information. Oh, and uh yeah, uh if you want to get that, I mean they can you can invest millions and millions of dollars into this, uh creating your own ecosystem, and it's probably still not gonna be right. So, I mean, that's my thought. Cole, was you think anything differently?

SPEAKER_00 1:01:30

Yeah, I I think call what you mentioned is true from like a macro enterprise level, but if I think of it on a like a more personal level, it's so I'm I'm I'm very outcomes-based. Like, so every day I've got tasks I want to achieve. And so the way I got into programming was like I would write utilities for myself, you know, Python utilities or or just bash utilities to transform information, gather information, um, just to make my life easier. And then I keep those utilities, those little programs, and just run them. And I I tend to use AI in the same way. Um, you know, like when you spoke about the hypothesis and like what a success looks like. I know what I want to achieve on a personal level every day. And so and I use AI to expedite that and to get closer to or get to that outcome faster. And whatever I use differs from day to day. I I think a big I wrote a blog on this recently, but you know, dark fooding your own AI. I think a big indication is how to what extent does AI companies use their own technology internally. And I mean I could I've like we do, but uh it's a shockingly low percentage of AA companies that's using their own AI. It's like a mo it's like a vehicle manufacturing company and and and no one drives their own vehicle, you know, they drive some other brand. But I think that's for me, Sean, that's like the main focus too, and I think that's a good flywheel for for like you know, like you mentioned, you're a small marketing group, you know, small group of people. That's a good flywheel to see which tech works. Can you how can you spin that out? That's just on a on a on a personal level. And and I love the way people compare notes, and especially on Reddit, what works for them and what don't does not work for them. I think one of the sort of I'm going off on a tangent there, but I think that's why the DGX Spark of Nvidia are so popular, that small GPU unit, because people run open source models on that, and it's you know, it it's no token costs, you know, it's just the price of the device, the hardware. Yeah.

SPEAKER_02 1:03:46

Yeah, I I could argue both ways, probably, because to support Carl's position, you know, people aren't going out and building their own telecom infrastructure, right? They're they're leveraging a telecom provider, and you know, they're not building their own computers, they're buying computers, and so I get that. But at the same time, I I think the business, if it's going to seriously leverage AI in the future, needs to start building a foundational skill set and institutional knowledge for the business about AI. You I think doing these kind of that kind of experimentation yourself and and building yourself also kind of forces the issue of education and training and governance and data readiness, which probably applies both ways. You know, these are disciplines that that the business needs to develop, and the business can't develop them if it doesn't have the opportunity or the necessity, like we talked about before, to develop them. And I I view it as an investment, right? Brendan, uh again, on my team, he's a really smart, tech-savvy guy. He he knows his way around HubSpot, he knows his way his way around everything, right? He's a generalist. But by telling him to get involved in building some AI initiatives for the marketing department without third-party help, he's educating himself, right? So I'm giving him license to spend company time to mess around on Claude and start figuring things out on his own. And that is a skill set that he's going to develop and grow and leverage here, and he'll be able to leverage it in his career in the future as well. And I'm learning as a byproduct of that as well, right? So I'm I'm a Gen Xer, technically an Exennial, if you know that microgeneration. But, you know, I'm I'm 48 now and I'm finding myself being a little grumpy about technology. And so I'm being dragged and screaming into this. So I, you know, I I need a a forcing function for me to actually get my hands dirty. I can have these conversations. I understand the the business theory and necessities and outcomes that we're trying to achieve with these things, but if you ask me to go. Build something for you, I like I literally don't know what button to push first. I have no idea. You know, so I have no hands-on experience in this kind of thing. So I think a business does that. But that leads me to the last topic that I wanted to touch on here, guys, which is let's fast forward a little bit in your AI maturity as an enterprise. And let's say that the people listening here, maybe they've got no AI agents today. Maybe they've got a handful because they've been doing some of this experimentation. And that let's fast forward a year and let's say that they've got a hundred AI agents. Let's talk about char for AI, not AI in HR, but char for AI. Ultimately, if you if you have 500 AI agents running in your enterprise, who manages them and what does that work chart look like? And who's accountable for what they're doing? How does the human workforce start to work alongside of the AI workforce? And what does that look like from an organizational and a management perspective?

SPEAKER_01 1:07:00

Yeah, so that's a good point. And that's why organizations need to um have uh cardrails on AI usage within the organization. They just shouldn't let everybody utilize different AI solutions. Like internally, one of my partner sales directors is leveraging Claude and several other AI platforms like Notebook and things of that nature to facilitate um creating um, you know, creating these uh account overviews, right? Uh scenarios. And I I said, yeah, that's okay. You can leverage it for that. Um he's also leveraging our AI for work product, which is our customer experience product, uh, employee experience product to facilitate the information instead of like a chat GPT, et cetera. So they're all working together um to facilitate this output, right? But it's important to have the guardrails on what AI usage is happening at work, right? And organizations need to be in sync, in sync and lockstep in what AI they're utilizing, how are they utilizing it, what are the compliance standards associated with that, right? What are the risks associated with that? They really need to nuance it significantly because when you open things up, then it gets more, things get more convoluted and actually you're risking um you're risking the organization from a compliance perspective, you know. And also, you know, you don't want people using AI for certain tasks, right? Writing a letter is one thing, but actually creating doing my job description, right? Leveraging AI, you might not want to do that, right? Whether you're like a uh insurance adjuster or something of that nature. I mean, you can actually create AI to facilitate the output on a claim if you wanted to, right? Without human in a loop, right? And you mentioned human and a loop. I think human loop is critical, especially when you're judging the output of information that is going to cost the organization a significant amount of money or whatever it may be. It could be money, it could be something else, right? Human loop is critical, and people don't realize that that's an important part of what we do here at core. And um, and I think, especially as it pertains to like banking decisions, insurance companies, loans, and things of that nature, you just don't want to leave it in the hands of AI. So keeping track of how AI is utilized and what AI tools are being used in workplace and put on people's laptops and how they're leveraging it could be bad, but also it could be good because you could leverage and you could say, wow, well, Joe is using AI for this, and this could be a really good best practice within the organization. So organizations can learn from their employees because you know, different employees have different skill sets within AI and they can create amazing things to benefit the organization, of course, from a productivity perspective, but obviously you have to know, you know, place guardrails on that, you know, and no know their no limitations, no limitations. And by the way, from my perspective, when I walk in to a very large organization and they've built their own CRM, for example, that's a flag. Like we could do an API integration instead of a CRM if it's open open source, right? Open API, and that's great. But to me, that shows that the organization is trying to be progressive. But I will tell you that in many cases that that CRM is outdated by the time they build it. Okay, because something happens, some change happens in the organization that they haven't accounted for, and then using their own builders and they're constantly cycling through builders, you know, engineers, and it just just the changes. The CRM doesn't work. So the you'll see people moving to Salesforce, right, from their own tools. You don't see it the opposite way as much. Once they're a lot of times, these these CRMs are legacy CRMs that they've been using for years, you know, they're using these legacy CRMs for years, and that they're just afraid to get off of it. They know it's not doing what they need to do. The same for AI, right? You build this legacy tool, and in two, three years from now, it's not going to be what you want it to. You don't have the support, you don't have the technology group, you don't uh, you know, have the expertise to take it to the next level to satisfy what your requirements are now. So that's the way I would look at AI and legacy technology. Any thoughts, Cobas?

SPEAKER_00 1:11:47

Yeah, I I fully agree with with with what Carl said. And just when you on what you said, Sean, about um having multiple agents and this you know, problem of agent sprawl. So something that's happening now is it's just technology is evolving, you know, as adoption evolves. Like so, one of the things is this whole idea of an agent control plane, and you'll see a lot of vendors are speaking about this the agent control plane that handles this agent sprawl and it gives you inspectability, observability, and discoverability. So I think you know, like this whole problem of all these rogue agents, different products will come along like an agent control plane where you can have like a register of agents and you can monitor and manage those agents. You know, a second thing is what Carl mentioned about human in the loop. Um obviously you want you know confidence thresholds, and if confidence is low or if it's a a risky transaction, you want to involve a human. Recently, uh Andre Kaparthi spoke at a Y Combinator event, and you know, he was um a founding member of OpenAI for many years the lead on Tesla, you know, autonomy, autonomous vehicles, and he spoke about an autonomy slider, right? So he said, like we live in this world of all or nothing, no autonomy or full autonomy to AI. And he actually spoke about the autonomy slider, and depending on the use case, you want more autonomy or lower autonomy. And that came to mind when Carl spoke about you know the whole human in the loop idea. And we actually constructed this whole grid of different tasks and risks and domains, and in each one of those, you want a different level of autonomy and a different level of human involvement. So I think that's really important when it comes to implementing AI agents or agentic AI, you know, any software with agency where you understand the task that's being executed and understand the level of autonomy you want and the level of human involvement. And just to close that off with, you know, open claw or claw bot, you know, that's I mean, that opened a whole new attack surface when it comes to nefarious attacks. But at least that showed us that people have a desire to you know automate their lives. So that that's what's a good confirmation. But I that just came to mind when you spoke about these rogue bots running around unmanaged.

SPEAKER_02 1:14:22

Well, guys, uh rogue bots running around sounds like a good place for us to stop. So Cubis Grayling and Carl Cats, thank you for your time and expertise today. Thank you.