Latest / You've Been Heard / 382- Securing Employee AI Usage w/Devs.ai

Transcript

382- Securing Employee AI Usage w/Devs.ai · You've Been Heard

• 0:12
  Speaker A: All right, really cool guys. We have Aaron Bailey, General Manager Devs AI on today. And I don't do a lot of vendor interviews all the time. I'm doing this because we've talked so much about real AI use cases and people are always asking me for that. And we're getting hounded by AI vendors, harassed all the time and everyone's like there, people are trying to tell me, you know, people are always telling me that I need these things and I need to buy this AI thing and they're trying to sell me stuff that they know that I need that I know I don't need. So of course that's why we have you on, to sell some stuff that we don't need. But in reality, I think the, in the AI world, the thing that's the most important concern right now, at least what I'm hearing is security data, specifically data security. And I think everyone's paranoid that their end users and employees are doing stuff on AI that's, that's, you know, leaking data to the world. And I want to be able to automatically offer, give, refer someone a real enterprise level solution that's secure. So that's one thing that's going to be your first task to demonstrate or explain why or how is Dev's AI an enterprise solution? Why is it secure? Kind of like maybe you get just the general overview and then I will continue. I have probably like a ton of.
• 1:41
  Speaker B: Questions, but go totally, yeah. So I mean just very high level overview. We have three core value props. The foundational value prop to your point is based on the assumption, I mean you said that people are paranoid that people are using kind of unapproved AI tools and they shouldn't just be paranoid like they 100% are using unapproved AI. That's not necessarily a bad thing. I mean you, you want folks to augment their product, find new ways to augment their productivity. About 75% of employees are using unapproved what we would call free AI tools. So the free version of ChatGPT or anthropic or of like little point solutions like your, your meeting recorder, that's free or whatever. So our foundational value prop is that we offer a place to consolidate all that AI usage onto one platform with proper governance and make it compliant. So we offer what we call private AI workspaces, secure environment for your team and your data so that the employees don't have to worry about security. Just your IT guy can be the one who worries about security. And frankly we, alongside your IT guy will worry about security. And so the couple things that you get access to are, you know, on top of that workspace are access to every major LLM. There's some cost controls I think that are helpful there, and then a no code agent builder, which I think we'll talk about hopefully today. But if you were just to talk through like the, if you were to subscribe directly to the enterprise version of each of these providers, you'd be paying $230 per month per seat. The reason I say the enterprise version, like some people are on the teams version of OpenAI or some of these other providers. And when you talk to the OpenAI Enterprise team, at least last time I did, they said, they basically say, hey, this is the only version that truly doesn't train off your data. When you go into the T's and C's, I actually have a slide that covers at least the free version of the T's and C's here. The way that they define content, they can train off that content. But on the teams plan, plus, you know, they, they have, they say we don't train off of any of your business data. It's like, okay, what's, who's defining business data? I think it's a very important question. But on the enterprise. So that's why I say the enterprise version of these, $230per month per seat. We offer all those Same models for $30 per month per seat. And so that's like the dev's AI value. And then from a security standpoint, I'll demonstrate this obviously. But I mean, this answers 90% of the questions that come up when folks ask about, like, what do you mean by secure? What do you mean by private?
• 4:00
  Speaker A: Right.
• 4:00
  Speaker B: We obviously have agreements in place with every major LLM provider that they will never use the data on our platform to train subsequent models. You know, even if they were lying and they did want to train off that data, we host some of our more popular models on private Azure instances, for example, OpenAI models. So all the OpenAI models we have on private Azure instances today, you, as far as I know, and I haven't used every single AI tool out there, but as far as I can see, we are the only platform that actually enables admins to go in and see all the data that's been uploaded from their employees and manage that data. So delete that data anytime. We don't keep any copies of it. Obviously people, people get worried because they see Deepseek here, which is a model provider based out of China. Correct. But they do open, they release open source model. So we just download those and run those on US based servers. That's kind of important. But from a compliance standpoint, SOC2 compliant GDPR, ISO 2700 on prem deployment options really for people who have like DoD contracts and that sort of thing, we are really trying to.
• 5:02
  Speaker A: That's really, really cool. So I mean so if they just want to roll it out themselves, they can. But the SoC2 that the ISO 2701 is going to make some people feel comfortable.
• 5:13
  Speaker B: Yeah. And the last piece here, and then I mean I won't go through all these slides, but the last piece here, similar to the delete date anytime piece is again, as far as I can tell we're the only ones that allow admins to see chat logs across your employee base. You can export those chat logs, run compliance tests on them, those sorts of things. Two main things that are a benefit here are one, you know, if someone, an employee uses an agent to download to scrape and download your, your customer list quits the next day, takes that down the street to a competitor, you're going to want to log that nefarious act. The other piece here is like as you build, as you build agents that.
• 5:48
  Speaker A: Are being logging of nefarious activities. Okay, just, just to summarize real quick, how many LLMs do we have ability? What are the major ones that you have? Private Azure instances I guess OpenAI any others?
• 6:01
  Speaker B: So let's just open up the model so as an admin you can come in and see all the different models, toggle them on and offset defaults which is why I bring this up and you can see the ones that have this kind of private.
• 6:13
  Speaker A: Excellent.
• 6:14
  Speaker B: So these, so the, the ones that don't have private doesn't mean they're not private. They're just operating under our agreements with the. Okay. And so you know, I don't think any of these model providers frankly have a very good incentive to like lie to us. That would be, it'd be devastating if they broke even one of those agreements. Okay. And so you know, I think we, we, we should trust the agreements that they have in place. But again for, for people who are, who want kind of that extra, that extra assurance, those private models I think.
• 6:47
  Speaker A: And devastating from the standpoint of APP direct in general is like I don't know what we were evaluated as of recently, but at least 1.2 billion six years ago, not to mention 90,000 plus customers, you know, 600 plus vendors etc so it would be quite devastating.
• 7:06
  Speaker B: Yeah, no, I mean, yeah, so. And it's not just kind of chat models. Right. We also have image models on here as well. And that's not really covered in the slides that I showed before, but.
• 7:16
  Speaker A: Okay, we just go back to that one slide. I just want to. People like you've been looking for this like an enterprise secure LLM model. I mean this is kind of more of an all in one with actual real, you know, I guess you could say SOC2, ISO 2701 GDPR for the European people. I don't think we care about GDPR in the United States. You know, secure access, being able to delete chat logs, being able to, I'm assuming, download chat logs as well, avoid in case nefarious activity happens. So that's it. Like that's really, really good.
• 7:59
  Speaker B: Just show off the chat logs too. Yeah. Export button there, you can cut it up by user, by agent or by model.
• 8:06
  Speaker A: I'm assuming rollout is ridiculous. Now for my demo, we have a lot of different kind of like just models and agents that are in there that are kind of like out of the box agents. Can we delete those and not have those in if you don't want them?
• 8:17
  Speaker B: Yeah, so that's right. So if you go to features, I mean, what. Yeah, so what Phil's talking about, for those watching, if you click this Explore Agents button, we have like a whole marketplace of third party agents that, you know, agent developers can come and build and launch those agents on, on our marketplace. But if you're rolling it out to your employees, for sure you're going to want to toggle off that public catalog.
• 8:39
  Speaker A: Let me just go through a couple, like one use case that came up the other day. So let's say we've got law firm that has to sift through some weird 3,000 page legal documents and provide summaries for them. Is that something they could do in here? Is that something that we would need to do more from a coding standpoint?
• 8:55
  Speaker B: No, no. Yeah, so you. Yeah. So a couple different ways that you'd want to do that. If you want to spin up your own kind of personalized legal agent, you can do that. I mean our default chat will take that as well. So you can just click drag. I mean we take all, we support all different document types and so if it's PDF, if it's a Word doc, most likely, if it's illegal, it's Word Doc. And so you can just copy paste or not copy paste, but click and drag that into here.
• 9:19
  Speaker A: I'm just trying to think of some like real use cases. Some real kind of like down and dirty gritty like use cases that that dev's AI has been really able to. I mean other than just glorified, we're better than Google. Like you should use us like employee for no, totally.
• 9:34
  Speaker B: And I mean we can. So I mean I can, I can walk through certain, certain agents that do well but I mean when we did customer interviews a few months ago, these are, these are like the main things that we saw on a per department basis that people were using devs AI for on a daily, weekly, monthly basis. And so you know, I'm getting, I can share this deck with you. You can share it with anybody, you know, you know, any end user who's thinking about who's evaluating different, different AI solutions. But, and then if you know, so that's from a department standpoint. But like what is AI actually being used for? Content creation, research, marketplace operations. This is really app direct focus. So you don't have to worry about that one help desk. I mean, you know, we have the ability to spin up agents and then embed them. We actually can spin up a widget for you and you can just copy paste that widget onto your website. So you have a 247 agent that lives on your coding.
• 10:27
  Speaker A: Coding, I have to hear.
• 10:29
  Speaker B: Yeah, so coding. So this one, we, we are not, I don't think ever are going to compete directly with like a cursor or a GitHub copilot. We, we definitely, you know, we have a GitHub integration and the way that our, we've architected our kind of rag pipeline, it makes it really easy to ask questions about a given repo or a coding kind of question. And we have Codex or OpenAI's, what is it called? Codex GPT5 that you can use.
• 11:05
  Speaker A: Let me ask you this way. Let's say my developer who I've been asking to jump on here, but he's busy right now so please jump on this right away if you can. Let's see if he's going to get back to me. Let's see. I don't know, he, he might, he might not. So let's just say he built. And he's more of a vibe coder and he hates that I call him that. Let's. So he's built like this crazy long. Oh, what's wrong with me? Like prompts, like he has multiple prompts all kind of like, you know that he builds in, let's just say Claude or he builds in, in open AI in Their kind of project section. Right. So you don't have like a project section here. Those parts are kind of like gone, right?
• 11:52
  Speaker B: No, so, well, well, yeah.
• 11:54
  Speaker A: Okay, so you don't understand, like, like if you use Claude natively, it has different features that are like. Not just the chat box or what do we want?
• 12:03
  Speaker B: That's right. Yeah. Yeah.
• 12:03
  Speaker A: All right. Not just the, like this, where, you know, so he has like this project now. Let's just say he does build the massive long prompt and, and all of this stuff in Claude. Is there anything stopping him from cutting and pasting that and putting it here as an age?
• 12:19
  Speaker B: Oh, no, yeah, for sure. I mean, no, yeah, that's.
• 12:22
  Speaker A: Understand. So like, even if, like you have one IT guy that pays 250amonth because he has to have this version of open AI, he has to have this, he has to have this and he, he builds out all of, he builds out his agents there because that's just what he's used to. He's used to that gui, he's used to whatever. There's no reason why you can't give every end user 250, 300, a thousand end users this version and just take the, the prompts and put them in here and build.
• 12:51
  Speaker B: Yeah, that's not hard. Yeah, same with memory. I mean, right? I mean we, we, we have. Memory is like a store bank in here. Like, I mean what you can do is just copy paste and say, you know, from open, from chat GPT or from the can, just copy paste and say, remember all these facts, okay, default chat. And it'll, it'll fill in. So I mean, that's easy.
• 13:10
  Speaker A: I think next biggest, like the biggest like thing that I get asked a lot that like supposedly cannot be done. Right. Or hasn't been done or we haven't found a good solution for yet is creating separate secure buckets of data, AKA maybe you have a single source of truth in your company that's this massive data bucket and we need to segment that and we only want to give specific end users access to specific buckets of data, AKA single sources of truth or sources of truth. How can we give devs, AI access to that single source of truth securely so that it can data dip from that without people being super paranoid that it has access to our data?
• 13:56
  Speaker B: Well, there's, yeah, so there's a couple, there's a couple pieces. One, we have kind of native data source integrations with the Google Suite and the Microsoft Suite. Yep. We also have, we're continuing to launch like a slew of MCP servers that allow you to give, read, write access to different, to different platforms. You can also call any third party, any third party MCP server. So if it doesn't live in Google or if it doesn't live in Microsoft, then you can call like your ERP or QuickBooks or you know, whatever. Okay. Your kind of local CRM. Okay, the other, the other point, your question, I think it had to do with permissions like groups and that kind of thing.
• 14:40
  Speaker A: Yeah. Because. Okay, so let's say you've got this massive single source of truth in your company. Right. But there's a lot of HR data. So HR data might have what you get paid, what I get paid. And you don't want Johnny, you know, the whoever attempt guy that's like, you know, has access to this agent to be able to find out when everyone gets paid.
• 15:00
  Speaker B: So. Yeah, so that's right. So we basically as, as the user. If I'm building. So if HR is building an agent, then yeah, they're like if admins for workday are in there creating an MCP connection to workday, then it'll use whoever is basically logged in and. Well, actually there's. I'll take it one step back because there's really two ways to do this. One is you can upload. So all of our HR data is in Confluence. Right. And so if I make this, if I make this connection right now. Confluence.
• 15:40
  Speaker A: Yep.
• 15:40
  Speaker B: It's only going to upload data that I have access to. Okay. Okay. So it's permissioned at the user level. So if the CEO who has, let's just say he has access to everything, he builds it, uploads data here, then it's going to scrape all that data. And then if he makes this available in his organization's catalog because it's by default it's not open to everybody, but if he makes this available in his organization's catalog, then you know, the entry level hire would have act, you know, through that agent would have access to everything. That's right. So you're going to have to make sure that you're dealing with. Okay. Only, only ingest this data.
• 16:18
  Speaker A: Gotcha. Yeah, so that's one of the use cases that, that came up on a, on a podcast a while ago is it would be great if we had a sales rep that's at an expo and he has a bot. We have a bot. So that when someone asks him a question at the expo, he can immediately turn up. It'll search like all of the new, whatever sales data you Know, parts, manufacturing parts, all this different stuff that's, that exists in this single source of truth, let's just say, and be able to pull accurate data for a potential customer live while he's at an expo. In other words, he might get a question about something that's was on the roadmap six months ago and he doesn't know is it, you know, did we actually go live with this or not? And he'd be able to pull up and actually produce, not only that, be able to produce a PDF or some sort of document kind of like on the fly, based on particular information that this customer just asked him.
• 17:18
  Speaker B: Yeah, yeah, that, I mean that's, that's on the fly.
• 17:21
  Speaker A: Sales slicks, you know, that have never been developed type of thing based on actual information.
• 17:25
  Speaker B: Yeah, so there's two. Yeah, exactly. There's, there's and there's multiple companies that kind of do this and it manifests in two ways. One is, yeah, I basically have a sales coach that as I'm on a call with a prospect I can like type in different questions and, and that can generate like based on our sales methodology or based on kind of the battle cards that we have with our competitors, they can help me kind of come up with points to answer questions that come up. The other way is kind of on this customer support side of things where people have a catalog. There's a call center with, you know, 20 to 50 people in there that answer questions about, you know, people will call in about, you know, their vehicle and they say, hey, you know, I'm, I need to replace my steering wheel or whatever. They can query the entire catalog and say, okay, here are your four different options based on your make and model. And you know, here's what shipping will look like, you know, for these four different options, pricing difference. So that kind of, that kind of thing is, yeah, very, very useful. I mean those sales, those sales, customer support people also kind of act double the salespeople right in that call center.
• 18:34
  Speaker A: So, so to keep this short and sweet, I mean just kind of, I'm trying to think of how we can break this up into different parts.
• 18:41
  Speaker B: But.
• 18:44
  Speaker A: If you had like, I don't.
• 18:46
  Speaker B: Know.
• 18:49
  Speaker A: Five seconds, five seconds to tell anybody about this, what would you tell, what would be the first, what would be the most important thing? If you had five seconds to tell someone to something about Dev's AI, that's going to buy you another 30 seconds, you know, what would that be?
• 19:04
  Speaker B: Yeah, well, I mean generally the nice thing about so private AI workspaces So we're combating shadow AI. So we are amongst the most like secure places on the Internet to play around with AI and we are LLM agnostic. Right. So you save a bunch of money if you have a bunch of subscriptions already to Grok, to Anthropic, to chatgpt, you can consolidate all that usage on a compliant platform. That's the reason it buys. And then the reason they renew is because of our agent, like our agenta capabilities capabilities. So people will come and identify redundant workflows internally. They'll have their internal kind of AI heroes or what we call AI enthusiasts. They'll pair those with those redundant workflows. They'll begin to automate those workflows internally and then that's why people renew. So people buy for the security and for the LLM agnosticism. No vendor lock in. Right. And then they stay for. Because they've automated a bunch of.
• 20:01
  Speaker A: All right, so the first one was the most secure place on the Internet to.
• 20:07
  Speaker B: Yeah amongst. I mean I haven't been all over, you know, I haven't been on every AI website but I, you know, I, I definitely would say among the, among the safest places on the web to, to play around with.
• 20:17
  Speaker A: Okay, okay. Okay. Next most important things. I'm sure there's a lot of ways to build. There's a lot of ways to build different agents and tools. You guys are one part. So we. What part of the AI equation do you guys fill and what part do you not and do we need to. Do we. Do people need more?
• 20:40
  Speaker B: Yeah, so that's actually very. Yeah, good question. So I mean there's, there's a few solutions that we, that we have today. Right. So AI chat, the, the AI agent builder, enterprise search, which is basically connecting your data and then being able to, you know, ask questions about it. This can automate a lot of like.
• 20:57
  Speaker A: That'S that single source of truth we were kind of talking about.
• 20:59
  Speaker B: Exactly. We actually have launched an internal vibe coding app builder internally. The differentiator would. I mean the idea is we want to help folks, IT folks and then marketing folks build two types of sites. One is like an enterprise grade security secure kind of B2B apps that, that basically would, would replace your current, a lot of your current tech stack. So We've replaced about $250,000 worth of SaaS apps internally with our vibe coding.
• 21:32
  Speaker A: Please explain. Please explain.
• 21:34
  Speaker B: Yeah, so I don't think I actually.
• 21:37
  Speaker A: Have access permission to talk to that.
• 21:40
  Speaker B: Yeah I have a slide actually. So at Thrive. So actually Let me see if I can pull it up here. Let me.
• 21:45
  Speaker A: Well, just give any example, like how so, I mean, yeah, so we.
• 21:50
  Speaker B: Yeah, so a good example might be we have an okr KPI tracking platform that we use to basically roll out KPI's to the entire team and like, what they are, what their OKRs are every year. That, that is not a hard app to build. Right. And, and so we were spending, I think it was 80k a year on that platform and we had reporting, basically.
• 22:14
  Speaker A: Some kind of like reporting or what.
• 22:16
  Speaker B: Yeah, so, yeah, that's right. And there was like this encouragement, like high five kind of aspect to it. There was a bunch of different little features that didn't take us out. So it was probably two to three weeks to build that, that we actually did have a software engineer go go over it and review it because we used a kind of variety of Vibe coding apps just to get a sense of what the market was, was offering. And so that's, that's kind of thing, number one. Thing number two, we, we created a marketplace of all the different offers, all the different Spiffs that are available on a per provider basis on our, this is on our, the marketplace side of the business. Not really. You know, our companies, AppDirect, where, you know, an advisor could come and see all the different providers that have Spiffs running. So that was another one. We launched a whole new website for one of the companies that was recently acquired by AppDirect. There was a PR release that was going to go out the next week and it was actually our CEO who kicked off the project of redoing the entire website via a Vibe coding app and then pass it off.
• 23:22
  Speaker A: He played around with it.
• 23:24
  Speaker B: Yeah, Yeah. I mean, well, yeah, yeah. And he's not technical. Right. Okay. So, yeah, So I mean, I have a whole list if you give me a sec to pull it up. It's good.
• 23:35
  Speaker A: It's just, it's kind of like, what if people are like, I don't have time to like Vibe code or do this. I can put this in our hands. We need to do this. Do we have some people that are just good at helping custom build these agents?
• 23:49
  Speaker B: Yeah. So we do have, you know, we do have a catalog of different professional service providers that we can introduce people to that they can, you know, buy their time on an hourly basis. But I mean, the, the real benefit of, I mean, this is what I would implore your audience. Phil.
• 24:06
  Speaker A: Yeah, go shoot.
• 24:08
  Speaker B: Like, they need to adopt this. They need to adopt AI. They need to adopt it in their, in their workflows. I think a lot of people are worried that AI is going to take over a lot of these jobs. And this has maybe been said too much at this point, but it's not that. It's not that AI is going to take jobs. It's that people who know AI are going to take jobs of people who don't know AI. And so, you know, this is. We've, we've tried to make this very self serve. There's a bunch of people creating content around devs and you know, we have a bunch of content that we can share with people. But overall I think it's, you know, it's important that people try to do this on there.
• 24:44
  Speaker A: What's the box? Website.
• 24:46
  Speaker B: So this is the company that we acquired.
• 24:48
  Speaker A: Okay.
• 24:48
  Speaker B: Yeah. So I, I don't have a before, I just have an after that.
• 24:52
  Speaker A: I can show their website that you guys rebuilt with.
• 24:58
  Speaker B: That with. Yeah, exactly.
• 25:01
  Speaker A: Do we have, Are we, are we giving access to the Vibe coding piece yet or.
• 25:05
  Speaker B: No, no, it's only internal right now. Yeah, unfortunately. So I'm hoping that by end of year we can roll it out. So this whole website was Vibe coded.
• 25:17
  Speaker A: You can see this node. Js.
• 25:21
  Speaker B: I think it's a react site, but. Okay.
• 25:24
  Speaker A: Either way, it's not WordPress.
• 25:26
  Speaker B: That's right. It's like a very interactive site.
• 25:31
  Speaker A: Dude.
• 25:35
  Speaker B: Anyway, yeah, that was our CEO that led this kind of campaign to do this.
• 25:39
  Speaker A: Okay, so please let us know as soon as this is available.
• 25:42
  Speaker B: Your.
• 25:42
  Speaker A: Your, your Vibe coding aggregation. Is that we're going to, Is that what it's going to be? Is it going to be a Vibe?
• 25:48
  Speaker B: I mean, we're just calling it App Builder. And you can see we. I have it. I can like kind of show it off here, but it's, you know.
• 25:54
  Speaker A: Okay, please do we want to know because like, where is this on the roadmap?
• 25:59
  Speaker B: I mean, well, so it's, it's, as you can probably tell, it's very high priority. I think we, we want to make sure that the experience that we release is something that's inspiring.
• 26:11
  Speaker A: Okay, are you saying like really like literally like coding for dummies?
• 26:16
  Speaker B: Well, no, there wouldn't, there wouldn't be any coding really involved.
• 26:18
  Speaker A: I mean, Vibe coding for dummies.
• 26:20
  Speaker B: Exactly. You, you. Well, so all, all Vibe coding apps right now. I mean, you don't really need to know anything about coding. It's all natural language. So you're using your natural language too.
• 26:31
  Speaker A: I want this. Can you build this for Me. Okay.
• 26:33
  Speaker B: To code an app. That's right.
• 26:35
  Speaker A: Okay.
• 26:35
  Speaker B: Or a website doesn't necessarily have to be an application.
• 26:38
  Speaker A: Okay. Okay. Excellent. Other than that, you guys are the enterprise solution, the most secure place on the Internet as far as we know, to ingest all LLM models and not leak data to the public or learning or the give the models ability to.
• 26:59
  Speaker B: Learn free versions of those applications. Yeah, yeah, that's right.
• 27:03
  Speaker A: And then.
• 27:03
  Speaker B: Okay.
• 27:04
  Speaker A: And then build agents as well. I would love some really good examples of agents that, you know, that people have done that have been very, very vastly productive.
• 27:13
  Speaker B: Yeah. So I mean, the most common ones that people build are definitely the, that sales agent that we were discussing before. So. And then a lot of ticket automating. So like hr, for instance, like HR at large orgs gets the same type of ticket. You know, it's, it's usually five to 10 different tickets that they get that same type, you know, 100 times. And they're answering. So it's like, how many, how do I log pto? How many PTO hours do I have? I don't want to log into like workday or whatever their system is. So having an agent that can answer those questions, they can build deterministic workflows that answer those questions accurately.
• 27:54
  Speaker A: Okay, so. So just walk me through what they would do. They would, they would. How did they build this agent?
• 28:02
  Speaker B: Yeah, so I mean, this is. We could, we're running low on time here, but I mean, generally, you know, I'll just, I'll just pull up an agent here. I mean, step one is you're going to give it custom instructions. Step two is you're going to choose a model. So you know, 4.1 model is just fine, but I mean, you can choose any model and run evals based on. Yep. Step three is data. So, you know, are there, is there any proprietary or relevant kind of data that needs to be ingested by this, by this agent?
• 28:31
  Speaker A: Yep.
• 28:32
  Speaker B: Step four is tools. So, you know, this is probably the most important aspect. Like do you want to give a certain read write access via MCP or via API? Do you want to give it access to search the web, Python tool, spreadsheet tool, image generation, etc. But for that HR tool, I mean, usually you're just going to give it maybe API access or MCP access to.
• 28:55
  Speaker A: Your database, some kind of.
• 28:58
  Speaker B: And then in this case, I mean, we had built these deterministic workflows. I mean, this is an AT&T quick quote agent. Yep, it's opening. One sec.
• 29:05
  Speaker A: What is it pulling Information from like a file of pricing or something or like.
• 29:10
  Speaker B: Yeah, exactly. A catalog. Yeah.
• 29:12
  Speaker A: Okay.
• 29:13
  Speaker B: A catalog at and T's catalog of, you know. Yeah. Of basically like what SKUs are available for what products and what.
• 29:20
  Speaker A: Right, right, right, right, gotcha.
• 29:22
  Speaker B: And so you can see like we built a workflow, right? So if someone says generate a quote, it's first that's going to show the products that are available at that address. It's going to take the user's input of what, you know, what product they.
• 29:33
  Speaker A: Want, and then Fiber Direct access or something.
• 29:37
  Speaker B: Pull a, an API and basically. And I can, you know, we can actually, we can demonstrate this if you want. Where it's like, it's something that would usually take four or five days. I can just now come in and say generate a quote. And I mean, we bounced from this like HR agent to this agent. But since we're at this agent, you can see like, I can just get.
• 29:59
  Speaker A: I mean, the HR one I think is pretty simple. Like if they know their top five questions that they get now. But once they're done with this. Oh, they built the agent and then how. What do they do? Just like, hey, go to the agents and devs AI guys and ask your question. Your hr.
• 30:12
  Speaker B: Yeah. I mean, you, you'd have your own catalog of your organization's agents that people can leverage. But you can also pull it into Slack. We're working on a teams integration. So you can pull it into teams. But I mean, you can see I just gave this my address. Now I'm getting this input field here where it's like, okay, I want dedicated Internet. And then the next step here it's going to pull.
• 30:31
  Speaker A: That's probably gonna ask you for address or something or what?
• 30:34
  Speaker B: I already gave it my address.
• 30:35
  Speaker A: Okay, okay, gotcha, gotcha.
• 30:36
  Speaker B: And so now it's gonna pull basically all the SKUs that are available for that product at that address. And so it'll begin to fill in those SKUs.
• 30:45
  Speaker A: Great.
• 30:46
  Speaker B: I mean, I won't go through this.
• 30:47
  Speaker A: Whole can it search fiber locator at the same time. Do we have an API in a fiber locator or anything like that?
• 30:52
  Speaker B: Well, I mean, I didn't build this agent, so I was.
• 30:55
  Speaker A: Okay, I know what to do.
• 30:57
  Speaker B: Yeah, yeah. But this is available publicly. You can go into our, into our kind of public agent marketplace. This one.
• 31:04
  Speaker A: Yeah. Super cool. Really? Really? I guess. Thank you for your time. The as now just from, from a coaching standpoint for it, like, how do they get HR guys to go ask the questions to HR do they just say, hey, anytime you contact hr, start here first at the agent at the age. HR agent or something like that. I mean, and then let's just say the HR agent can't answer their question. It's like, okay, does it open up a ticket for him? Or like, what. What are we doing there? I'm just curious.
• 31:30
  Speaker B: Yeah, I mean, well, so again, it's very. It's very dependent on the systems that they have and how they want to ingest it. So, yeah, it's like, like most people will use like a JIRA for a ticketing system, right? For. Or they have like a sauna or, you know, different. Just different kind of project management tools for. Or ticketing tools for. For this kind of thing. And so, yeah, I mean, do you want it to. Do you wanted to. Do you want the agent to create a new ticket for every question and then show that it's completed that ticket? Or would you rather just go in and download the chat logs? So there's a variety of ways that, that you can do it, but yeah. Beautiful.
• 32:07
  Speaker A: Okay, excellent. So again, people come for the security aspect, stay for the, I guess the agent building and the, the. The AI. We can fire 20 of our employees and downsize now because, yeah, it'd be.
• 32:25
  Speaker B: Less about firing points and more repurposing.
• 32:27
  Speaker A: Them to people that know how to use AI.
• 32:30
  Speaker B: Yeah, exactly. Cool.
• 32:33
  Speaker A: You will use this. All right, man. Really appreciate it. Thank you so much for your time.