1 00:00:00,000 --> 00:00:03,880 Welcome to the deep dive your shortcut to being well-informed today 2 00:00:03,880 --> 00:00:07,600 We're tackling a platform that really sits right at the intersection of customer 3 00:00:07,600 --> 00:00:11,760 service open source flexibility and well cutting-edge AI 4 00:00:11,760 --> 00:00:16,420 We are sinking our teeth into tile desk. It's a really powerful platform 5 00:00:16,420 --> 00:00:20,360 That's I think fundamentally changing how companies build these sophisticated 6 00:00:20,360 --> 00:00:25,310 LLM enabled chat bots and live chat solutions. It really is and it's crucial 7 00:00:25,310 --> 00:00:27,080 because it hits two major demands 8 00:00:27,560 --> 00:00:31,030 You know for enterprises at the same time. You've got the need for advanced 9 00:00:31,030 --> 00:00:31,640 automation 10 00:00:31,640 --> 00:00:34,360 definitely, but also that 11 00:00:34,360 --> 00:00:38,480 Non-negotiable requirement for data control. It's an entire ecosystem really and 12 00:00:38,480 --> 00:00:42,080 the fact that it's open source that elevates it way beyond just a standard 13 00:00:42,080 --> 00:00:46,650 Sauce product. Absolutely and before we dive deep into the architecture and all the 14 00:00:46,650 --> 00:00:47,720 agentic AI stuffed 15 00:00:47,720 --> 00:00:50,840 How this boasts we want to give a quick nod to our supporter today safe server 16 00:00:50,840 --> 00:00:52,920 Yeah, safe server is a great partner here 17 00:00:52,920 --> 00:00:57,960 They focus specifically on hosting exactly this kind of flexible powerful open 18 00:00:57,960 --> 00:00:59,360 source software 19 00:00:59,360 --> 00:01:03,600 They really support you through the technical side of digital transformation 20 00:01:03,600 --> 00:01:08,450 So if you're looking into deploying say custom secure AI solutions like tile desk 21 00:01:08,450 --> 00:01:10,280 in a robust hosting environment 22 00:01:10,280 --> 00:01:13,520 Well, you can find more info at www.safeserver.de 23 00:01:13,520 --> 00:01:18,570 Perfect. So, okay our mission today. It's tailored specifically for you the 24 00:01:18,570 --> 00:01:21,360 listener. We want to unpack tile desk from the ground up 25 00:01:21,360 --> 00:01:24,810 We'll explain how it combines, you know classic live chat with these sophisticated 26 00:01:24,810 --> 00:01:26,160 large language models 27 00:01:26,160 --> 00:01:30,860 And crucially how you can actually deploy it using enterprise-grade tools like kubernetes 28 00:01:30,860 --> 00:01:31,520 and docker 29 00:01:31,520 --> 00:01:35,340 We want to make it accessible even if you're maybe just starting out with your 30 00:01:35,340 --> 00:01:36,720 conversational AI strategy 31 00:01:36,720 --> 00:01:39,920 Okay, so let's start with the basics the foundation tile desk 32 00:01:39,920 --> 00:01:44,670 It actually began its life as an open source live chat platform built really robustly 33 00:01:44,670 --> 00:01:45,940 using node.js and express 34 00:01:46,240 --> 00:01:50,270 Standard stuff and it was designed for that asynchronous real-time customer support, 35 00:01:50,270 --> 00:01:52,320 you know designed to scale but honestly 36 00:01:52,320 --> 00:01:55,120 It's grown way beyond just being a chat widget now, right? 37 00:01:55,120 --> 00:01:59,320 So if I'm looking at this, is it just like a better version of the old chat box? 38 00:01:59,320 --> 00:02:03,070 I already have on my website or is it something more? It's really an entirely new 39 00:02:03,070 --> 00:02:05,880 category. I think of it as a conversational app development 40 00:02:05,880 --> 00:02:09,440 platform less like just a chat box and more like a 41 00:02:10,680 --> 00:02:15,120 Complete no-code visual workbench kind of like tools like voice flow maybe but it's 42 00:02:15,120 --> 00:02:18,220 purpose-built to empower you to create these highly sophisticated 43 00:02:18,220 --> 00:02:23,440 LLm enabled bots bots that can handle complex multi-step tasks 44 00:02:23,440 --> 00:02:26,880 It's really about building automated applications that just happen to communicate 45 00:02:26,880 --> 00:02:27,920 conversationally 46 00:02:27,920 --> 00:02:31,700 Okay, but for any company, you know worried about automating customer interactions 47 00:02:31,700 --> 00:02:36,430 The fear is always losing that human touch or worse providing those frustrating 48 00:02:36,430 --> 00:02:39,240 dead-end AI responses. So for enterprise adoption 49 00:02:39,240 --> 00:02:42,600 What's the single most essential feature? Is it still being able to hand off to a 50 00:02:42,600 --> 00:02:46,080 human? Oh, absolutely that human-in-the-loop capability is 51 00:02:46,080 --> 00:02:51,080 Completely non-negotiable. It has to be there tile desk provides these out-of-the-box 52 00:02:51,080 --> 00:02:52,520 chat bots that work seamlessly 53 00:02:52,520 --> 00:02:57,700 side by side with human agents if the AI gets confused or maybe the customer 54 00:02:57,700 --> 00:03:00,360 expresses something that needs real empathy or it's a 55 00:03:00,360 --> 00:03:01,760 high-stakes problem 56 00:03:01,760 --> 00:03:06,720 The interaction can instantly transition. Yeah, really smoothly to a human agent, 57 00:03:06,720 --> 00:03:07,280 but here's the kicker 58 00:03:07,280 --> 00:03:12,000 Okay, the human agent gets the full transcript and all the context from the AI 59 00:03:12,000 --> 00:03:12,760 interaction 60 00:03:12,760 --> 00:03:16,300 So there's zero frustration the customer doesn't repeat themselves and you get 61 00:03:16,300 --> 00:03:17,440 reliable support 62 00:03:17,440 --> 00:03:22,060 So the customer isn't starting over from scratch, right that alone fixes like 90% 63 00:03:22,060 --> 00:03:23,680 of bad chat experiences 64 00:03:23,680 --> 00:03:27,560 Probably now let's talk about managing communication across different platforms 65 00:03:27,560 --> 00:03:29,960 that multi-channel challenge most businesses 66 00:03:29,960 --> 00:03:33,160 They have to build separate bots for whatsapp for their website for Facebook 67 00:03:33,160 --> 00:03:33,640 Messenger 68 00:03:34,200 --> 00:03:38,760 Managing that content parity is just a nightmare exactly. That's the pain point 69 00:03:38,760 --> 00:03:41,080 tile desk addresses really brilliantly 70 00:03:41,080 --> 00:03:45,730 I think with its innovative multi-channel approach you design and configure your 71 00:03:45,730 --> 00:03:49,160 chat bot replies only once you use rich media elements 72 00:03:49,160 --> 00:03:52,160 You know images carousels buttons right there in the design studio 73 00:03:52,160 --> 00:03:58,330 Then the platform's underlying engine it just auto adapts those responses to the 74 00:03:58,330 --> 00:03:59,040 unique 75 00:03:59,200 --> 00:04:04,510 Formatting needs of every single target channel web whatsapp Facebook Telegram SMS 76 00:04:04,510 --> 00:04:05,960 even email Wow 77 00:04:05,960 --> 00:04:10,420 So the key takeaway there is efficiency build it once deploy it everywhere and the 78 00:04:10,420 --> 00:04:12,160 format just sorts itself out automatically 79 00:04:12,160 --> 00:04:14,360 You're not rebuilding like six separate brains 80 00:04:14,360 --> 00:04:18,560 You're just giving one brain the ability to speak six different format languages. 81 00:04:18,560 --> 00:04:21,460 That's genuinely interesting for you know large-scale operations 82 00:04:21,460 --> 00:04:25,100 Exactly it standardizes all that complexity. Okay. Let's move to the brains of the 83 00:04:25,100 --> 00:04:26,560 operation the AI itself 84 00:04:26,560 --> 00:04:30,080 We're hearing terms like a gentic AI operating system. Mm-hmm 85 00:04:30,080 --> 00:04:32,160 It's one thing to answer a simple FAQ 86 00:04:32,160 --> 00:04:36,560 But tile desk seems to be aiming for more complex workflows like information retrieval 87 00:04:36,560 --> 00:04:37,720 operations stuff 88 00:04:37,720 --> 00:04:42,990 Yeah, and what's absolutely crucial for any enterprise AI is accuracy. You just 89 00:04:42,990 --> 00:04:46,080 cannot have bots hallucinating responses with customers 90 00:04:46,080 --> 00:04:51,070 It's a non starter. This is where the knowledge base integration comes in and it's 91 00:04:51,070 --> 00:04:52,040 powered by our agi 92 00:04:52,040 --> 00:04:55,630 That's retrieval augmented generation essentially the AI agents are connected 93 00:04:55,630 --> 00:04:58,920 directly to your company's internal documents your knowledge base articles 94 00:04:58,920 --> 00:05:02,480 Maybe even external data sources. The AI isn't making up the answer 95 00:05:02,480 --> 00:05:06,340 It's actually retrieving a verified source document and then formatting the answer 96 00:05:06,340 --> 00:05:08,200 based on that source right our eggs 97 00:05:08,200 --> 00:05:12,790 We've heard about our G before but the documentation specifically mentions a gentic 98 00:05:12,790 --> 00:05:14,280 aria that sounds well 99 00:05:14,280 --> 00:05:18,380 It sounds like it's tailored for large organizations with frankly messy data. It is 100 00:05:18,380 --> 00:05:19,200 spot-on 101 00:05:19,880 --> 00:05:24,880 Standard rag often hits this massive enterprise problem siloed knowledge 102 00:05:24,880 --> 00:05:27,640 information is all over the place a 103 00:05:27,640 --> 00:05:32,500 Single sort of monolithic RG system tries to index everything and that can lead to 104 00:05:32,500 --> 00:05:33,600 relevance issues 105 00:05:33,600 --> 00:05:37,480 The AI might pull a pricing sheet when it should have grabbed a warranty document 106 00:05:37,480 --> 00:05:41,650 For example agentic RG solves this it lets you separate your knowledge into 107 00:05:41,650 --> 00:05:43,560 multiple distinct repositories 108 00:05:43,560 --> 00:05:48,080 So you might have one repository just for sales PDFs another one for HR policies 109 00:05:48,080 --> 00:05:49,720 and maybe a third for technical 110 00:05:49,720 --> 00:05:53,070 Support manuals. Okay interesting and how do they work together then do they talk 111 00:05:53,070 --> 00:05:55,200 to each other you chain them together essentially 112 00:05:55,200 --> 00:05:57,960 The system applies search fallbacks and guardrails 113 00:05:57,960 --> 00:06:01,400 So it allows the agent to say check the technical docs first 114 00:06:01,400 --> 00:06:05,120 And if that doesn't yield anything useful, maybe then it tries the general FAQ 115 00:06:05,120 --> 00:06:05,840 repository 116 00:06:05,840 --> 00:06:09,730 It ensures your answers are verified. They're routed correctly and retrieved from 117 00:06:09,730 --> 00:06:11,460 the right specialized knowledge source 118 00:06:11,460 --> 00:06:15,730 It prevents that cross-domain confusion that brings up the challenge of search 119 00:06:15,730 --> 00:06:19,080 itself though users don't always ask questions perfectly, right? 120 00:06:19,080 --> 00:06:23,820 They use messy colloquial language. What about this hybrid search? Does that 121 00:06:23,820 --> 00:06:26,100 actually help solve that problem? 122 00:06:26,100 --> 00:06:29,970 It really does because hybrid search cleverly combines two powerful techniques. It 123 00:06:29,970 --> 00:06:32,080 uses traditional full-text retrieval 124 00:06:32,080 --> 00:06:36,600 That's your basic keyword search, which is fast and great for exact matches and it 125 00:06:36,600 --> 00:06:38,580 pairs that with semantic understanding 126 00:06:38,580 --> 00:06:43,560 Which is powered by the LLMs themselves. So if a user types something complex like 127 00:06:43,560 --> 00:06:48,140 My delivery was supposed to be here yesterday, but the tracking hasn't updated 128 00:06:48,140 --> 00:06:48,760 since Tuesday 129 00:06:48,760 --> 00:06:51,480 The system understands the intent behind it 130 00:06:51,480 --> 00:06:55,740 It's a delivery issue needing tracing even if the exact keywords are kind of spread 131 00:06:55,740 --> 00:06:56,640 out and imprecise 132 00:06:56,640 --> 00:07:01,480 This leads to much smarter result and internally fewer missed resolutions 133 00:07:01,480 --> 00:07:05,110 And the system doesn't just sit there executing instructions does it? It actively 134 00:07:05,110 --> 00:07:05,440 learns 135 00:07:05,440 --> 00:07:10,780 That's probably the defining feature the self-learning loop. The AI agent isn't 136 00:07:10,780 --> 00:07:11,440 static 137 00:07:11,440 --> 00:07:15,200 It actively listens to every single message exchange, especially those where the 138 00:07:15,200 --> 00:07:18,360 interaction eventually had to transition to a human agent 139 00:07:18,760 --> 00:07:24,080 That's key. It extracts crucial beta points common questions that stumped it before 140 00:07:24,080 --> 00:07:26,400 and importantly the effective solutions 141 00:07:26,400 --> 00:07:28,120 provided by the human agents 142 00:07:28,120 --> 00:07:33,010 This new high quality knowledge is then automatically and seamlessly integrated 143 00:07:33,010 --> 00:07:34,680 right back into the knowledge base 144 00:07:34,680 --> 00:07:39,200 Huh? So every single time a human has to step in and you know save the day 145 00:07:39,200 --> 00:07:43,890 They are without even realizing it training the AI to handle that exact scenario 146 00:07:43,890 --> 00:07:45,080 better next time 147 00:07:45,200 --> 00:07:49,120 That's continuous improvement baked right into the operational layer. That's pretty 148 00:07:49,120 --> 00:07:52,360 clever precisely and we've seen this incredible learning capability 149 00:07:52,360 --> 00:07:56,240 But great AI still needs sharp instructions to execute tasks reliably, right? 150 00:07:56,240 --> 00:08:00,050 Which brings us to how tile desk actually makes designing complex workflows simpler 151 00:08:00,050 --> 00:08:01,260 using their visual tools 152 00:08:01,260 --> 00:08:04,940 We're talking about the no-code slow builder or the visual designer. Yeah for 153 00:08:04,940 --> 00:08:08,640 anyone building automation complexity always kills usability 154 00:08:08,640 --> 00:08:10,480 It just gets overwhelming 155 00:08:10,480 --> 00:08:16,110 How does tile desk make managing these potentially huge complicated automation 156 00:08:16,110 --> 00:08:17,500 routines manageable? 157 00:08:17,500 --> 00:08:21,400 Especially when you're throwing AI logic into the mix it relies heavily on this 158 00:08:21,400 --> 00:08:22,720 concept of prompt chains 159 00:08:22,720 --> 00:08:27,180 This is really how you decouple the complexity if you try to write one massive 160 00:08:27,180 --> 00:08:31,860 Monolithic prompt telling the LLM to handle say five different steps all at once 161 00:08:31,860 --> 00:08:37,660 Well debugging that or updating it becomes basically impossible prompt chains break 162 00:08:37,660 --> 00:08:38,560 that logic down 163 00:08:39,000 --> 00:08:42,920 Instead you create a sequence of smaller smarter much more manageable prompts 164 00:08:42,920 --> 00:08:43,420 Okay 165 00:08:43,420 --> 00:08:47,700 So each little prompt module has just one specific job like check inventory or 166 00:08:47,700 --> 00:08:50,780 calculate shipping costs something like that exactly and the full builder 167 00:08:50,780 --> 00:08:52,700 Let's you do this orchestration with no code 168 00:08:52,700 --> 00:08:56,520 You just drag drop and connect these smaller prompt blocks visually this give you 169 00:08:56,520 --> 00:08:58,540 really granular control over the workflow 170 00:08:58,540 --> 00:09:02,660 Makes it transparent much easier to debug and ensures the AI follows the exact 171 00:09:02,660 --> 00:09:04,560 decision tree you designed step-by-step 172 00:09:04,900 --> 00:09:08,590 Hmm that structure seems pretty essential and we start thinking about the multiple 173 00:09:08,590 --> 00:09:10,300 specialized agents concept, too 174 00:09:10,300 --> 00:09:13,740 If you can break down the workflow you can break down the agents as well, right? 175 00:09:13,740 --> 00:09:18,560 You don't need one giant AI trying to be a genius at absolutely everything that 176 00:09:18,560 --> 00:09:19,300 really works 177 00:09:19,300 --> 00:09:22,740 Well instead you create AI agents with highly specialized roles 178 00:09:22,740 --> 00:09:28,440 Maybe one for sales another for technical support a third for onboarding new users 179 00:09:28,440 --> 00:09:32,480 And because the workflows are chained these specialized agents can actually 180 00:09:32,480 --> 00:09:34,840 collaborate and even switch mid conversation 181 00:09:35,260 --> 00:09:38,740 So if the sales agent identifies a technical issue cropping up 182 00:09:38,740 --> 00:09:43,220 The replace bot action allows a seamless handoff to the specialized support agent 183 00:09:43,220 --> 00:09:45,820 and crucially it passes all the context along 184 00:09:45,820 --> 00:09:48,820 Okay to ground this powerful design in reality 185 00:09:48,820 --> 00:09:52,140 Let's maybe look at a few specific use cases from the source material that 186 00:09:52,140 --> 00:09:53,940 illustrate this kind of integration 187 00:09:53,940 --> 00:09:57,950 The ecommerce assistant sounds like a great example of connecting AI directly to 188 00:09:57,950 --> 00:09:59,140 core business systems 189 00:09:59,140 --> 00:10:03,940 Oh, it's powerful because it moves way beyond just simple FAQ stuff. The ecommerce 190 00:10:03,940 --> 00:10:06,580 assistant integrates directly with platforms 191 00:10:06,580 --> 00:10:11,370 Like Shopify it syncs product catalogs in real time. It can display interactive 192 00:10:11,370 --> 00:10:14,100 product carousels right within the chat window and 193 00:10:14,100 --> 00:10:18,810 Crucially it allows the customer to conversationally build and modify their 194 00:10:18,810 --> 00:10:19,740 shopping cart 195 00:10:19,740 --> 00:10:24,180 That really transforms what might have been just a support bot into a genuine 196 00:10:24,180 --> 00:10:25,180 conversion engine 197 00:10:25,180 --> 00:10:29,420 And what about addressing that traditional time sync of customer service email ticketing? 198 00:10:29,420 --> 00:10:33,830 Yeah, the AI email ticketing feature looks like a huge efficiency game. It links to 199 00:10:33,830 --> 00:10:35,460 basically any corporate inbox 200 00:10:35,460 --> 00:10:39,870 But here's the smart part before responding it actually learns from your historical 201 00:10:39,870 --> 00:10:42,340 email threads and successful resolutions from the past 202 00:10:42,340 --> 00:10:46,940 This means its automated responses are rooted in real high quality past 203 00:10:46,940 --> 00:10:51,580 interactions that leads to more accurate often instant resolution of common queries 204 00:10:51,660 --> 00:10:55,660 Which minimizes the backlog for human agents and then there's this simpler but 205 00:10:55,660 --> 00:10:56,180 still critical 206 00:10:56,180 --> 00:11:00,890 24-7 scheduling feature it connected Google Calendar or Outlook Calendar handles 207 00:11:00,890 --> 00:11:05,340 availability checks confirms appointments all without any human intervention needed 208 00:11:05,340 --> 00:11:09,170 Okay, that's a pretty comprehensive look at the functionality, but the real value 209 00:11:09,170 --> 00:11:11,760 proposition. I think especially for technical or 210 00:11:11,760 --> 00:11:18,050 Regulated industries is that tile desk is open source. This gives users profound 211 00:11:18,050 --> 00:11:20,140 control over their infrastructure 212 00:11:20,340 --> 00:11:24,770 So how do you actually deploy tile desk at scale? What does that look like? Right? 213 00:11:24,770 --> 00:11:26,660 This takes us into the technical foundation and 214 00:11:26,660 --> 00:11:31,540 Tile desk offers flexibility here, which is good for initial development 215 00:11:31,540 --> 00:11:35,370 Maybe for trials or even smaller deployments the recommended option is Docker 216 00:11:35,370 --> 00:11:37,580 compose. It's relatively easy 217 00:11:37,580 --> 00:11:41,180 It's great for quickly standing up the necessary multi-container applications 218 00:11:41,180 --> 00:11:43,860 locally just to see how the whole system the server dashboard 219 00:11:43,860 --> 00:11:47,950 Database all interacts but Docker compose probably doesn't cut it when you need, 220 00:11:47,950 --> 00:11:48,300 you know 221 00:11:48,820 --> 00:11:52,980 5 nines reliability and massive user concurrency, especially with a real-time chat 222 00:11:52,980 --> 00:11:54,540 application precisely 223 00:11:54,540 --> 00:11:57,590 Yeah for production environments and definitely for enterprises that need high 224 00:11:57,590 --> 00:11:59,260 availability and serious scalability 225 00:11:59,260 --> 00:12:02,420 The clear recommendation is Kubernetes with Helm 226 00:12:02,420 --> 00:12:07,400 Chat is fundamentally a real-time service. If your server hiccups or fails active 227 00:12:07,400 --> 00:12:09,420 conversations just drop. That's bad 228 00:12:09,420 --> 00:12:12,320 Kubernetes handles all that container orchestration 229 00:12:12,320 --> 00:12:16,610 It manages automatic recovery load balancing rapid scaling of all the necessary 230 00:12:16,610 --> 00:12:18,580 components things like the core tile 231 00:12:18,580 --> 00:12:22,860 Desk server the dashboard the design studio the underlying MongoDB database 232 00:12:22,860 --> 00:12:26,520 Kubernetes ensures continuous service even under peak loads 233 00:12:26,520 --> 00:12:30,530 Okay, and the combination of being open source and having these controlled 234 00:12:30,530 --> 00:12:34,380 deployment options enables that really critical concept of data sovereignty 235 00:12:34,380 --> 00:12:37,660 Why is that such a massive differentiator these days because offering on-premises 236 00:12:37,660 --> 00:12:42,160 deployment means organizations get complete actual physical control over their data 237 00:12:42,160 --> 00:12:42,740 and their privacy 238 00:12:43,180 --> 00:12:47,630 Full stop tile desks can be deployed entirely behind your corporate firewall. This 239 00:12:47,630 --> 00:12:49,580 is often mandatory for sectors 240 00:12:49,580 --> 00:12:53,480 Like finance health care or government entities and this control extends right down 241 00:12:53,480 --> 00:12:55,860 to the most sensitive parts of the AI stack itself 242 00:12:55,860 --> 00:13:00,340 Can you explain that control a little more specifically like what exactly do they 243 00:13:00,340 --> 00:13:02,660 get to control pretty much everything that matters 244 00:13:02,660 --> 00:13:04,660 They get local LLM integration 245 00:13:04,660 --> 00:13:09,250 This means they don't have to send potentially proprietary or sensitive customer 246 00:13:09,250 --> 00:13:12,460 data out to a third-party cloud providers LLM 247 00:13:12,460 --> 00:13:17,150 Like open AI or entropic. Yeah, they can connect to and run open source models like 248 00:13:17,150 --> 00:13:18,060 llama and Mistral 249 00:13:18,060 --> 00:13:22,100 Completely locally within their own infrastructure. Yeah, furthermore. They 250 00:13:22,100 --> 00:13:23,220 maintain vector store control 251 00:13:23,220 --> 00:13:25,980 They can use local storage solutions like Q. Dran 252 00:13:25,980 --> 00:13:30,580 Which for the less technical listener is basically the specialized database that 253 00:13:30,580 --> 00:13:34,060 holds the semantic meaning of your documents for the our age system 254 00:13:34,060 --> 00:13:38,470 Keeping that vector store local and secure behind the firewall is paramount for 255 00:13:38,470 --> 00:13:39,420 compliance and privacy 256 00:13:39,660 --> 00:13:43,260 That kind of flexibility and security isn't just a nice-to-have feature 257 00:13:43,260 --> 00:13:47,820 It feels like a necessary strategic advantage for any large company serious about 258 00:13:47,820 --> 00:13:49,300 digital transformation now 259 00:13:49,300 --> 00:13:54,260 It absolutely is tile desk provides this powerful flexible open source way to 260 00:13:54,260 --> 00:13:56,060 automate customer interactions 261 00:13:56,060 --> 00:13:59,700 But it crucially maintains that human safety net while giving you 262 00:13:59,700 --> 00:14:04,500 unprecedented control and security over the entire data infrastructure and the AI 263 00:14:04,500 --> 00:14:05,060 stack 264 00:14:05,060 --> 00:14:08,580 That's an excellent summary of why this platform is creating such a stir 265 00:14:08,580 --> 00:14:11,650 And once again, we really want to thank our supporter for this deep dive safe 266 00:14:11,650 --> 00:14:12,180 server 267 00:14:12,180 --> 00:14:15,440 They help provide dedicated hosting solutions for digital transformation 268 00:14:15,440 --> 00:14:18,740 initiatives including powerful open source platforms 269 00:14:18,740 --> 00:14:21,160 Just like tile disk you can find out more at 270 00:14:21,160 --> 00:14:26,600 www.safe server dot de and maybe here's one final provocative thought for you to 271 00:14:26,600 --> 00:14:28,840 consider based on everything we've discussed today 272 00:14:28,840 --> 00:14:32,140 We talked about how tail desks self-learning AI 273 00:14:32,140 --> 00:14:35,980 Improves directly from human agent responses and feedback, right? 274 00:14:36,460 --> 00:14:40,910 Especially from those escalated cases the platform also offers these AI coup pilot 275 00:14:40,910 --> 00:14:43,380 tools that augment human agents in real time 276 00:14:43,380 --> 00:14:46,300 Providing suggestions and knowledge during live chats 277 00:14:46,300 --> 00:14:50,780 So think about this if the AI is learning directly from the humans and the humans 278 00:14:50,780 --> 00:14:52,920 are simultaneously be augmented by the AI 279 00:14:52,920 --> 00:14:57,080 How might this deep integration fundamentally change the role of the support staff 280 00:14:57,080 --> 00:14:57,620 over time? 281 00:14:57,620 --> 00:15:02,540 Could it turn them from simple agents focused just on resolution into expert AI 282 00:15:02,540 --> 00:15:06,060 trainers focused on validating and refining the organization's core knowledge? 283 00:15:06,460 --> 00:15:10,280 Something to mull over as you maybe explore these kinds of capabilities yourself.