Scaling Agentic AI Globally: How Joseph Kim Is Building the Next Enterprise AI Platform

Added: 2026-05-05

[00:00:00]Okay, so we got a new episode of Legends and Leaders and today it's great to have Joe here. Joe, you're the CEO of Druid, which is a company that is focused on making AI agents for enterprise that handle different tasks that are task-specific, um, you know, help with different integrations that occur, making sure data moves seamlessly.

[00:00:17]There's lots of ways AI can help in enterprises and you're at the forefront of that. So, I'm excited to have you here and to hear your story.

[00:00:23]Well, thanks for having me here, Ben.

[00:00:25]Thanks for having me.

[00:00:26]Awesome. So, Joe, first of all, you know, what got you interested in AI in general? I mean, I feel like everybody has their AI moment. I'd love to just hear about how you got interested in this. Yes, sure. So, um, it was a There's a a fairly interesting story, actually. Um, cuz the right before joining Druid AI as its CEO, back in September of last year, I was actually running another company called SimuLogic. It was a public company that had recently gone private, uh, and in the security and observability space for enterprise software.

[00:00:59]And we ended up doing a joint project between ourselves and Anthropic at the time to have some AI-related functionality in the product.

[00:01:10]And when we were going through that project, it became very, very clear to me that what was happening in the AI space was going to change almost everything in terms of B2B software. And so, at that time, I started to look out for additional opportunities for me to jump in two feet first into the AI space because things have been changing so fast. And ultimately, after looking at a few, you know, companies out there and looking at the technology, uh, ended up being that close, actually. Druid was by far the most mature product and platform in the market to be able to do agenda capabilities for customers. So, that's how I actually ended up at Druid.

[00:01:52]Uh-huh. But, how did you end up with the the role leading the company, though?

[00:01:56]Well, it's um, well, I mean, I was already, you know, leading another company. Um, and so, I was actually looking for something equivalent, you know, obviously, if I was going to jump in.

[00:02:06]And so, that's how I ended up there. Uh, and so, at Druid, you know, they had a lot of co-founders that were leading there. And then, the company like Druid had a co-founder, you know, the the main co-founder at the company who was running as CEO and then actually getting ill and and passing away. So, it was a a fairly extraordinary situation, but um, and I wish I would have gotten to meet him, honestly, and have lots of good memories with him. Um, but, uh, you know, fell in love with the technology and the team and so, I ended up joining the company.

[00:02:37]So, AI for enterprise, though. Like, what is the potential of this space? I mean, it seems like there's massive potential in every category, but how do you see AI helping companies most?

[00:02:47]And you know, it's um, the paradigm in terms of, you know, what's happening and and especially when you're building AI agents is like completely different than when you're building enterprise applications.

[00:02:58]Uh, and it's hard for people to really grasp, even today, uh, because a lot of people are using AI agents, they're not necessarily building or training them, right? But, if you think about how enterprise applications are built today, ultimately, it's a fairly deterministic kind of endeavor. Meaning, when you build an application or when you build a website, it's going to function exactly the same way all the time for its intended purpose for the end user. Okay?

[00:03:25]When you're building AI agents, uh, it's not about building like a website. It's almost like building a little being, uh, technology being that you build very quickly with some levels of guardrails and rulesets that you might give it. And then, you're going to hold a whole bunch of building out for that specific being that you created, in this case an AI agent, to be able to fulfill a goal as often and and as best as it can. And you're extremely over and over and over again for it to get better and better at its job. And so, from that standpoint, it's actually very, very different. Um, what so what we found, uh, where customers actually started getting into this endeavor, what we've seen in terms of their transformation, uh, is that over time, all of these complicated things in terms of how you would handle data.

[00:04:15]Right? So, if you think about all of the systems of record that a customer may have today, all the structured data and unstructured data, there are a lot of constructs on top of there for you to be able to get now information out of that data.

[00:04:27]Okay?

[00:04:28]And when a lot of agents are being used, LLMs being used, a lot of those middle world technologies are starting to go away because you can get sensitive data much more quickly using agents and generative AI technology. And so, we're seeing things in a in kind of an interesting way. Things get a little bit simpler where people can get access to information, useful information, much more quickly than they used to before.

[00:04:54]So, building and deploying AI agents, it's an interesting space. It I mean, it seems like all companies have tried to make it easier for people to do that on their own without technical knowledge, but there's some stuff that just seems to be really complicated that you do need that. How do you view just the accessibility of AI agents for enterprise and the needs maybe to make them more accessible? Yeah, you know, it's it's it's a really good question.

[00:05:16]You know, it's you know, I kind of think about it as a sort of where the market is at in terms of its understanding. Cuz a lot of these technologies actually have very similar interfaces.

[00:05:28]Right? So, when you go to ChatGPT, you ask a question and it's giving you answers or if you're using Gemini or, you know, any of these technologies, or if you're using something that you built for an enterprise, it still looks kind of like this chat functionality, like you're talking to like a person, right?

[00:05:44]And so, uh, it's so interesting to me when you think about how these technologies are used for B2C, you know, type of use cases versus B2B use cases.

[00:05:55]But, in reality, if you are using this for business, the the bar for security, the bar for fulfilling compliance is really, really higher.

[00:06:04]And in terms of that versus I'm not just doing a search on Google or asking ChatGPT for something. And so, internally, um, what we've been doing as we're having discussions is to make a fairly simple framework to be able to describe to customers on when they actually can use something like generative AI versus when they need something a little bit more than that.

[00:06:27]And so, how the framework kind of goes is that if you are using technologies, these LLM technologies, to generate ideas.

[00:06:37]Okay? And ideas could be in format of a, I don't know, a paragraph, a sentence, a summary, a picture, a video, etc. It's okay not to use things that are so stringent like agenda capabilities.

[00:06:49]Okay? And ultimately, it's because the bar doesn't have to be at that 95, 97, 99% accuracy, have all of these security guardrails, etc. Because it's going to be okay.

[00:07:02]So, if you get a suggestion, the person can easily determine within a few minutes on whether you see the idea or not.

[00:07:08]Okay? Now, that threshold starts to cross when you are using these types of a system not just generating ideas, but you want it to make decisions and take actions for you.

[00:07:21]So, all of a sudden, you've got your AI doing things on your behalf. Once that happens, then you do you do need some guardrails. You need to share data and additional levels for you to be able to get accuracy.

[00:07:33]And so, if our technology as an example, then, when you ask a question, like you're going to, you know, some functionality that Druid has built, when you ask a question, well, for the first answer comes out, it is not uncommon for the system to go through minimum 55 double checks and checks before it gives you the first answer.

[00:07:56]Okay?

[00:07:57]Maybe for a generative AI, it takes for, you know, getting an idea, it may be something that you need that level of capability.

[00:08:05]There's one more system on top of agentic, I would say, which is more around conversational AI components, which are more than just making decisions and actions, but how do I use these systems to converse directly with a human person. Okay? So, it's more than just making decisions, it's like, I want to be able to make this unique experience for you so that you have a pleasant journey to interact with a specific business. And that then takes more technology than what you might get out of just agentic capability, things like generative voice and, you know, other kind of skills that are even beyond, uh, you know, just, um, you know, building an AI agent, per se.

[00:08:47]So, when it came time to just figure out like where to focus on with enterprises, Joe, like what did you feel were the biggest kind of areas that you guys could could tackle and could help with that were, you know, like you mentioned, in this more specific, more precise like domain?

[00:09:01]Yeah. I I think the best thing that we could do, because it's because the platform is so broad, you can use it for so many different use cases.

[00:09:11]And in enterprise software, I heard the saying that, uh, it's impossible for you to build a god box that does everything. But, it's impossible to build a god box. You can't tell a customer saying he can do anything for you once you're in trouble.

[00:09:23]That that's not that's not a good way to, you know, be able to add value. And so, what we found, actually, is that most enterprises still have a ton of different difficulties in terms of trying to get workflow automation. You know, all of these technologies that already exist, they're still even buying more technologies to get to some level of automation that they haven't gotten to yet.

[00:09:46]That's what AI can tackle pretty well.

[00:09:48]So, if you have a workflow today, let's say it's just I don't know, collections.

[00:09:52]Like, every business need to collect money.

[00:09:55]Uh, if you automate the collections methodology, there's a lot of companies and a lot of station tools that you can use.

[00:10:01]You can actually build it from them and train the version to close that workflow automatically for you. Manual steps and even some automated steps by hooking it to existing processes.

[00:10:12]And by doing so, what we found is there's existing processes now we can almost completely automate for our customers. And these are processes that they're already familiar with that they know that they have to uh, find some level of efficiency. So, that's a core problem that we've been able to tackle.

[00:10:28]For any existing manual or automated process, we can build agents to automate that for you. So, that's kind of like point number one. The second piece that we found a ton of customer uh, value has been on a conversational AI piece. And and so, let's use that same exact collections workflow.

[00:10:47]So, instead of automating the workflow, that workflow is going to get a human in the picture at some point.

[00:10:53]So, let's say I'm a credit card agency that I'm in and, you know, you you know, we're trying to collect some money from you. Instead of automating the back of the workflow, we can actually fill in agent that interacts directly with you to give you the most pleasant experience without having to really have to look at a whole bunch of stuff or press one for this and three for that, etc. But, being able to get an understanding from you how to close that transaction as fast as possible, make it pleasant for you, and have that as an interface so that more and more of these transactions can be collected in this case, the collections process, from the customer as quickly as possible.

[00:11:27]Okay? And so, these kind of use cases related to practicals, you know, can be I'm using collections as an as a example of it, you know, maybe I'm in a doctor's office or something like that and I'm just scheduling an appointment.

[00:11:40]Well, you can now do that through a conversational AI agent. Or maybe you have an agent reach out to you, not you calling to it. Maybe it's also inbound and outbound related activity to go, "Hey John, I know that you had a uh, appointment coming up. Can you make it just to make sure that everything is, you know, kosher on on that front. And if not, then also be able to reschedule uh, appointments, you know, right there as it's doing the outbound call. Cuz ultimately, the more things that are happening in the background that you can post against scheduling or collections, etc., with the uh, with that initial in the specific calendar, the more money that the hospital or the business can collect.

[00:12:21]So, is there a lot of like bespoke lot like bespoke like making of agents for these companies and that's like majority of the business or is it kind of like the one-size-fits-all tool um, that covers enough stuff for them? Yeah, you know, it's uh, it's also a really good question. I think on the uh, kind of automating the workflow side in the background, what we have found is sort of like what I consider like the 1.5 AI agent technologies versus the 2.0, the true agentic capability.

[00:12:51]The 1.5 is more like there's some agentic features that you may have that already exist within your workflow technology. So, there's a lot of workflow orchestration companies that are more traditional that have features where they can do some of that capability. Okay? But, in essence, it's still same old traditional workflow automation. And then there are companies more like ours, which is that we're going to build an agent that's going to close that workflow for you, not just it not just as a wrapper.

[00:13:22]Um, and those are actually fewer companies that can actually do that, you know, real agentic AI development.

[00:13:29]There's fewer of those than than what we've seen in the market.

[00:13:33]The uh, other end of the spectrum, you know, certainly, you know, from a competitive side that we have seen is again more traditional conversational components like uh, if you were to go into a you know, call you know, an airline or something like that, whatever you call into to schedule or find information, as soon as you get first press one for this, press three for that, press three for this, etc. There's a lot of those kind of those technologies that are sort of imitating, you know, sort of chatbot functionality, but a lot of them don't have any real agentic capabilities internally cuz that's not how it would normally go. It's just bang, do you have a question? What would you like to uh, ask? And the engine is supposed to figure out exactly through uh, you know, natural language processing and everything else, exactly what you are looking for. And so, you can find some of these uh, you know, kind of competitors that are more traditional than with that of the wrapper around it than something that is truly agentic or conversational.

[00:14:28]But, but it's still fewer than fewer than what we had first imagined when we were competing in this space.

[00:14:35]Joe, where do you think the value lies when it comes to, you know, AI companies like like yours? I mean, I know the value kind of has shifted of how people look at it from like an investor perspective. You know, it used to be like they were thinking the models were the value, then the models became like open source. Where do you think the value lies now with companies with AI?

[00:14:53]I think actually more and more uh, especially in the B2B space, Ben, um, I think what you're going to find is the the value's going to come from um, more of the um, you know, it it's more of the ROI for each individual use case. Okay? And so, it's going to be things like time saved, you know, etc. on the efficiency side.

[00:15:16]So, if it's an agentic AI use case, typically, it's going to be all around, you know, one of the three traditional value levers. It's going to be time saved and it's going to be money money saved.

[00:15:26]So, instead of a person closing something or you have, you know, 50 different pieces of technology that's automating something for you and replacing it with one, that's going to be cost savings for you.

[00:15:38]Okay? On the other end of the spectrum, a lot of conversational AI components where people are the the AI agent uh, is filling out or communicating to close a business transaction, again, the examples that I used were things like collections or scheduling for a hospital visit or whatever you think might be.

[00:15:56]Those transactions tend to add value on the top line.

[00:16:00]Right? Because the more that you can collect, the more that you can um, you know, schedule a doctor or nurse or whatever the case might be, that's revenue that is actually coming in for that specific institution. And so, the two primary drivers that we have seen uh, is on both of those. We call them internally customer experience versus uh, employee experience and we do both of those things.

[00:16:23]Uh, from the initial use cases so far on the justifications that we've done, uh, the the part that is the ROI that is easier to justify is the customer experience stuff, the things that actually add stuff to the top line.

[00:16:39]When you think about like how to integrate AI agents into companies, you know, like everybody's concerned about data and data privacy and there's all these, you know, kind of precautions in place at companies. I mean, how do you think about like the best ways to integrate agents that get all the information they need cuz usually they need a lot of information. And also not compromising the company. Yeah, yeah, yeah. And and this is one of the things that I actually even ask customers if they're going down this journey to discuss um, the specific uh, you know, AI company that they're working with how do they deal with security? And ultimately, I think from um, you know, you got to look a little bit about security. I think from the market side is do you think that all of these uh, same compliance and and security regulations are going to stay and maybe even go beyond where they are right now?

[00:17:28]That is our perspective certainly at TrueEra. Or are they going to just forget about it and hopefully, you know, people are okay with the wild wild west?

[00:17:35]That's still happening, but that's not the perspective that we have right now.

[00:17:39]And so, because we have such good security uh, regulations in terms of, you know, the controls and logging, etc. in our technology, um, you know, our customers usually don't have to worry. And usually for new projects, we ask them the same thing.

[00:17:52]It's like, "What have you done and asked for specific uh, you know, vendor or whoever you're working with, what are the compliance checks that they've already gone through?" Because you need to have certain level of uh, legal awareness before you do ISO 27001 SOC 2 Type II, GDPR compliant, HIPAA compliant, etc. What do you have all of those things at the moment? We're even trying to go further than that to go, "Can we get uh, like software actual device certification at this point?"

[00:18:19]Because people are trying to install our software at that level.

[00:18:23]And so, um, and that's what we're seeing more and more stringent because of the sensitivity of the data. Now, if uh, you wanted to go into much more specifics around, you know, things like RBAC and you know, do you have your own, you know, do you do you inherit RBAC from existing systems?

[00:18:40]Uh, what do you do?

[00:18:41]You know, does your you know, information that's residing within TrueEra will have its own ability to control access, which we also do. Um, I think things like that, I think you know, customers can also ask, I think uh, to get a much more comfortable.

[00:18:58]Uh, but I think more than that is just taking a step back and going, "What is just the general approach on security?

[00:19:03]Do you have existing compliance certifications already?" And usually, that's a good place to start uh, for uh, for most people.

[00:19:12]Where do you think that the next big boom is going to happen from in AI? I mean, we've had these big moments recently that have occurred. Um, you know, major things that major habits that have changed like the way we access information or certain automations that have been introduced. Where do you think the next big one's going to come from?

[00:19:29]I if I start to look at use cases right now, then the uh, big thing that happened last year compared to the year before is probably around voice.

[00:19:40]Mhm.

[00:19:41]So, for that uh, previous to that, when you're looking at agentic AI capabilities, a lot of the the agent uh, the AI agent sounded a little a little more like the automated kind of voice systems. It was so very robotic. And also in the last year, the whole video kind of pulled on around like generated voice.

[00:20:00]Mhm. So, if I were to put on a voice with like the system itself, now it uses LLMs to generate natural voice that sounds like a person in different languages and dialects. And in a few minutes, I'll be able to instruct you versus something else that understands the vernacular of that specific vertical as well. That is a predominantly where things have been moving last year and heading into this year.

[00:20:20]I think if in this year, the big thing is AI agents and agentic frameworks.

[00:20:26]They're everywhere. Every company is doing AI agents, whether they're actually doing it or not. We're going to be in the space for decades. And so, we have a lot of comparisons to others. But there's now hundreds of thousands of companies that are now jumping into building AI agents for people, whether they have full capability or not.

[00:20:44]I think because there's so many different frameworks and so many companies trying to do it, but next thing that I'm thinking we're going to see heading into next year is going to be how do you orchestrate these agents together?

[00:20:56]Okay. So, if I have an engine from Druid and I have an engine from, I don't know, Salesforce and I'm using Copilot from Microsoft, how do I get these things to work together and communicate together and orchestrate the capabilities together?

[00:21:09]Uh internally, we have a uh capability called Druid Conductor, uh which actually helps do this. Um but I think you're going to see many, many more companies going after the orchestration space uh next year. I think that's going to be the next our next big thing. And then eventually, when we get to that point, there's a lot of tooling that we're just missing as an as a uh industry, like if I'm orchestrating my agent versus some other agent that's trying to do the same job.

[00:21:35]And then, for example, how do I test and measure those two things to see which one's better?

[00:21:41]Like those kind of frameworks don't really exist right now.

[00:21:44]>> Yeah. I think you're going to have to orchestrate and compare these agents that do similar jobs. Um we'll see an entire uh additional best practices and frameworks and companies uh where you're going to be able to now do comparisons.

[00:21:58]Yeah.

[00:21:59]No, that makes a lot of sense. Um just the last question I have, like what is your long-term vision for all of this right now, Joe?

[00:22:06]I think for us right now, I you know, for a lot of the folks uh when you know, when they take a step back in my kind of own frameworking here, Ben, I really think about the universe in three chunks at least right now in AI. In the enterprise side, it's conversational AI, agentic AI, and then there's generative AI. Okay. I think on the enterprise side, there's two players, you know, within AI and then there's Anthropic and Google and you know, you can use Block if you wanted to do something like that. But there's lots of players on that end.

[00:22:35]We're going to continue to have a model folks, you know, work on the models and then infrastructure people work on the infrastructure. I think on our end, we'll continue to push the boundaries on what can be done on the agentic side.

[00:22:47]Again, like I'm thinking of all this orchestration capability as an example.

[00:22:51]And we'll continue to push the boundaries on what can be done with the conversational component with voice. And so, those things are going to be the two big areas that we're going to push on.

[00:23:00]And for me, um you know, ultimately, you know, where I want to be for the company to be at right now, Ben, um in the future is how do we get in front to make sure that customers understand when they're making these decisions on all the factors and and, you know, things that they need so that they're not going to make a mistake at the beginning.

[00:23:20]Okay, cuz right now, we have a lot of wins, but most of those wins have come after a customer has burned through like millions of dollars to fund their own research or picking the wrong vendor or whatever the case might be. Uh and then they're like, "Oh, here's the seven things that we needed." And then we're like, "Well, here's the seven things and there's just five other things that you should be thinking about."

[00:23:38]And then we're winning the signing as a customer. But I feel bad for the customer ultimately cuz they have, you know, they didn't get to burn through the the dollars as much as they could right at the beginning. So, but for me, it's like the part that I want to be able to figure out for the company is still getting ourselves out there and getting our frameworks and all the knowledge that we have and making it more open to the public so that you know, customers can make some really good decisions, even if they're not choosing us, but make better decisions around the things that they want.

[00:24:08]These are the things that I should consider. And then ultimately, uh you know, landing in a a winning a winning space for them uh instead of trying to burn through cash. And so, that's where I want for us to be. And obviously, we have to learn a lot of those things, you know, to be frank. But I'm thinking that this future component at the beginning is sort of uh you know, what we would like to push. Yeah. Well, Joe, that was all the questions I had. I appreciate taking the time and doing this. I think you're in a fascinating space and you guys are having a really great impact in the category. Saving money for these businesses is is really important in time as well. And um you know, I'm just excited to see how the use cases are going to grow. I'm sure there's going to be lots of areas that Druid can apply the technology to. So, looking forward to what's next and thanks for coming on.