1 00:00:00,513 --> 00:00:01,976 Corey: Welcome to Screaming in the Cloud. 2 00:00:02,235 --> 00:00:06,705 I'm Corey Quinn, and I, in an atypical move, am in Seattle 3 00:00:06,713 --> 00:00:09,884 once again to have some conversations with folks at AWS. 4 00:00:10,071 --> 00:00:13,049 Um, Pasquale DeMaio is the VP of Connect. 5 00:00:13,071 --> 00:00:14,168 Is, is that where you start? 6 00:00:14,168 --> 00:00:15,139 Is that where you stop? 7 00:00:15,157 --> 00:00:16,661 It, it's gotten very hazy. 8 00:00:16,881 --> 00:00:20,568 Pasquale: Yeah, so I, I'm the vice president of Amazon Connect, and specifically 9 00:00:20,571 --> 00:00:25,093 Amazon Connect Customer is my area of focus, which is a, a software that we 10 00:00:25,093 --> 00:00:30,094 started 16 years ago designed to help customers, um, internally here reach out 11 00:00:30,095 --> 00:00:33,050 to our end customers and really delight them with better customer engagement. 12 00:00:33,338 --> 00:00:35,766 About eight years into that journey, I discovered a team was 13 00:00:35,768 --> 00:00:37,743 building that, and I was … couldn't believe what they had built. 14 00:00:37,743 --> 00:00:40,350 I was so excited, and I said to them, "Hey, do you guys think 15 00:00:40,352 --> 00:00:44,063 we could give this to customers of AWS who'd want to, who'd want 16 00:00:44,065 --> 00:00:46,632 to use it?" And they said, "Yeah, we'd love that." And literally 17 00:00:46,632 --> 00:00:49,424 we started down that path, um, we talked Andy Jassy into it. 18 00:00:49,432 --> 00:00:52,297 He was the, uh, CEO of AWS at that point, 19 00:00:52,313 --> 00:00:54,378 eventually the CEO of all of Amazon now. 20 00:00:54,699 --> 00:00:56,559 And, uh, he approved it, and we haven't looked 21 00:00:56,566 --> 00:00:58,382 back, and it's been a great, a great journey. 22 00:00:58,743 --> 00:01:00,634 Corey: So the history of it is sort of interesting because 23 00:01:00,637 --> 00:01:03,682 it started off, my understanding of Connect was it's a call 24 00:01:03,682 --> 00:01:08,055 center software to basically manage customer support reps, uh, 25 00:01:08,103 --> 00:01:11,301 inbound primarily, then it started supporting outbound as well. 26 00:01:11,585 --> 00:01:14,737 And then there was sort of a revolution that happened of we're all doing AI now. 27 00:01:14,737 --> 00:01:17,841 And suddenly the, the AI term started applying to it, 28 00:01:18,197 --> 00:01:21,225 and I confess, I thought that a lot of Connect was a 29 00:01:21,229 --> 00:01:24,202 little, okay, that seems like a weird distraction thing. 30 00:01:24,207 --> 00:01:26,220 I didn't take it particularly seriously. 31 00:01:26,980 --> 00:01:28,713 But then I had some customers who did, and 32 00:01:28,723 --> 00:01:31,216 okay, w- m- I have opinions, that's great. 33 00:01:31,237 --> 00:01:33,487 But customers are actually using this to solve business 34 00:01:33,502 --> 00:01:36,698 problems, and it's not all of them by a landslide. 35 00:01:36,698 --> 00:01:39,179 I mean, most of us hate the phone, let's not kid ourselves. 36 00:01:39,323 --> 00:01:42,503 But there are enough of them out there that their entire business 37 00:01:42,508 --> 00:01:45,775 is predicated on this, banks, airlines, insurance companies, 38 00:01:45,776 --> 00:01:49,731 that shape of customer, where this becomes a significant problem. 39 00:01:49,813 --> 00:01:53,900 Uh, how do you get here from its somewhat humble beginnings as this 40 00:01:53,914 --> 00:01:56,529 is basically a phone system for people that don't wanna build one? 41 00:01:56,658 --> 00:01:58,500 Pasquale: You made some points that I think are really interesting. 42 00:01:58,500 --> 00:02:00,649 One of them you implied, and I think kinda maybe stated, 43 00:02:00,651 --> 00:02:03,601 was we are a little different than other AWS services. 44 00:02:03,604 --> 00:02:03,814 Right. 45 00:02:03,814 --> 00:02:04,803 We are a little up the stack. 46 00:02:04,835 --> 00:02:09,456 Um, we, we… Our first reason for existing, um, as a product external 47 00:02:09,502 --> 00:02:14,020 to Amazon was really about the fact that the scale and security we 48 00:02:14,020 --> 00:02:17,176 needed ourselves, and then a native and AI integration was there from 49 00:02:17,176 --> 00:02:21,228 day one, although obviously AI has transformed since, a little bit. 50 00:02:21,241 --> 00:02:23,029 But it, and which has been incredibly exciting. 51 00:02:23,035 --> 00:02:25,512 Um, and then we had a couple really interesting inflection points. 52 00:02:25,522 --> 00:02:29,228 The first one was the scale and security, and it was something that just wasn't 53 00:02:29,228 --> 00:02:32,454 available at all at that point in the cloud from a contact center product, 54 00:02:32,456 --> 00:02:35,223 and we really were just a really a voice-based contact center from day one. 55 00:02:35,478 --> 00:02:37,934 And by the way, everyone said we were nuts, which I kind of enjoyed at the time. 56 00:02:37,935 --> 00:02:40,045 You know, we were like, you said, it was like we're a little bit different. 57 00:02:40,340 --> 00:02:41,705 I think peculiar is the Amazonian term. 58 00:02:41,709 --> 00:02:42,417 Yeah, exactly. 59 00:02:42,418 --> 00:02:43,201 That's exactly right. 60 00:02:43,220 --> 00:02:46,777 And so that, that opportunity to go do that allowed us to be a 61 00:02:46,778 --> 00:02:49,198 little bit unfettered and, and just really listen to customers. 62 00:02:49,204 --> 00:02:51,725 Not that we don't do that everywhere in AWS, but, but, like, 63 00:02:51,744 --> 00:02:55,268 we, we chartered that course, and then some things happened. 64 00:02:55,269 --> 00:02:58,101 Uh, you know, COVID happened, and again, the scalability, 65 00:02:58,147 --> 00:03:00,463 work from home, all of the fact that we're really cloud-based. 66 00:03:00,493 --> 00:03:03,406 Corey: Suddenly it wasn't a, a concern of, oh, everyone on the calling floor. 67 00:03:03,427 --> 00:03:05,910 When I started, I wouldn't even say my career, but back when 68 00:03:05,912 --> 00:03:08,817 I was in college, I did a stint as a telemarketer for a credit 69 00:03:08,819 --> 00:03:11,807 card company that no longer exists, and that was a lot of fun. 70 00:03:11,819 --> 00:03:15,263 Lot of calls every hour, outbound to interrupt people at dinner to pitch 71 00:03:15,263 --> 00:03:18,398 'em credit cards, which probably explains a lot about my personality. 72 00:03:18,691 --> 00:03:20,365 But it was always come on site. 73 00:03:20,505 --> 00:03:22,911 I- in the fullness of time, and then with COVID, well, 74 00:03:22,916 --> 00:03:25,353 suddenly people are not going to all be in the same room. 75 00:03:25,355 --> 00:03:26,312 How do you do this? 76 00:03:26,320 --> 00:03:28,829 We're not gonna install a PBX in everyone's house. 77 00:03:29,143 --> 00:03:31,074 How do we solve this problem without giving out, 78 00:03:31,086 --> 00:03:33,270 you know, people's cell phone numbers to customers? 79 00:03:33,790 --> 00:03:36,665 And suddenly, you were the center of attention in 80 00:03:36,666 --> 00:03:38,254 a way that I don't think you had been previously. 81 00:03:38,508 --> 00:03:39,935 Pasquale: The interesting thing is we were pretty 82 00:03:39,936 --> 00:03:42,035 passionate about the cloud-based nature and being able 83 00:03:42,037 --> 00:03:44,907 to work from home, um, before any of that happened. 84 00:03:44,907 --> 00:03:47,797 Um, but that put such a fine point on it and put such a pressure 85 00:03:47,798 --> 00:03:50,493 point on it for so many businesses that they had to modernize quickly. 86 00:03:50,494 --> 00:03:53,493 And another aspect of Connect was you can set it up in minutes. 87 00:03:53,493 --> 00:03:56,031 You can be taking your first phone call literally five minutes into it. 88 00:03:56,311 --> 00:03:58,893 Um, and so that just was very differentiated. 89 00:03:58,908 --> 00:04:01,554 And then again, the scaling, the cloud-based nature, you 90 00:04:01,555 --> 00:04:03,939 know, come for the five-minute setup, stay for the cloud- 91 00:04:03,978 --> 00:04:06,268 Yeah … was a pretty compelling message for folks back then. 92 00:04:06,569 --> 00:04:09,868 Uh, we weren't a compelling product in, in terms of being multimodal. 93 00:04:09,872 --> 00:04:11,602 We didn't have even chat at that point. 94 00:04:11,838 --> 00:04:14,515 Since then, we've radically… The product doesn't look, even from a 95 00:04:14,515 --> 00:04:17,822 year ago, looks radically different than what it looks like, um, today. 96 00:04:18,036 --> 00:04:20,588 But, but that's been one of the blessings, is because we, we 97 00:04:20,588 --> 00:04:23,150 grew really fast as a service, we had a lot of investment, a 98 00:04:23,150 --> 00:04:26,619 lot of support from AWS, and, and Andy as the, as the CEO, uh, 99 00:04:26,816 --> 00:04:28,778 I think sometimes he still has a soft spot in his heart for us. 100 00:04:28,778 --> 00:04:32,416 Um, but, like, we've had, uh, the, a real runway there, and then the 101 00:04:32,416 --> 00:04:35,570 investments in AI that we've made broadly in AWS and with so many of our 102 00:04:35,571 --> 00:04:38,907 partners, you know, we partner with all the big AI players as well, and 103 00:04:38,907 --> 00:04:41,885 that's enabled us to run really fast at some of these more interesting 104 00:04:41,894 --> 00:04:45,131 things later on than just the, than more, the more boring scalability 105 00:04:45,142 --> 00:04:47,832 aspects of it, which are still really important, but not as exciting. 106 00:04:47,968 --> 00:04:48,133 Corey: Yeah. 107 00:04:48,166 --> 00:04:51,323 You've had a couple of recent announcements that are, I guess, 108 00:04:51,336 --> 00:04:53,763 expanding the surface area of what it, of what you're building. 109 00:04:53,955 --> 00:04:55,188 What are you doing now? 110 00:04:55,404 --> 00:04:58,708 Pasquale: Our, our whole mental model around this is that we think 111 00:04:58,708 --> 00:05:01,747 that the future of engaging with customers is gonna radically 112 00:05:01,749 --> 00:05:05,177 change, and our belief is that with Connect Customer, you 113 00:05:05,177 --> 00:05:08,930 should be able to go 100% human all the way to 100% agentic AI. 114 00:05:09,188 --> 00:05:10,769 But almost nobody will wanna do that. 115 00:05:10,793 --> 00:05:13,169 What they'll do is they'll pick a medium across that, and they'll 116 00:05:13,169 --> 00:05:17,089 use what I like to call, um, deterministic AI and workflows to 117 00:05:17,089 --> 00:05:19,935 help support those things, and so you have AI every touchpoint. 118 00:05:20,260 --> 00:05:23,784 So in the last two to three years, we've just changed Connect Customer 119 00:05:23,784 --> 00:05:27,664 to be solely focused on how AI brings to bear on every aspect of 120 00:05:27,665 --> 00:05:30,166 what you do, whether that be predictive outbound, whether that 121 00:05:30,170 --> 00:05:32,973 be predictive, you know, even just helping make a webpage better. 122 00:05:33,244 --> 00:05:35,924 Just understanding your customer and delivering them the service 123 00:05:35,935 --> 00:05:39,186 and the engagement they need either proactively or I like to say 124 00:05:39,483 --> 00:05:42,586 reactive proactivity, which is when someone calls in, and instead of 125 00:05:42,586 --> 00:05:44,842 saying, "Hey, how can I help you?" Which gets them yelling, "Operator. 126 00:05:44,843 --> 00:05:48,409 Operator," you say- "Hey, I see your flight was canceled. 127 00:05:48,596 --> 00:05:49,761 I've booked you on the next one. 128 00:05:49,825 --> 00:05:50,947 Is that what you're calling about?" 129 00:05:51,267 --> 00:05:52,699 Corey: Yeah, that, that works really well. 130 00:05:52,764 --> 00:05:55,560 A- again, you also forgot the most important necessary prerequisite 131 00:05:55,560 --> 00:05:58,451 to this when someone calls in, it's, "Please pay attention as our 132 00:05:58,451 --> 00:06:02,119 menu items have changed." You… And they haven't changed in 50 years. 133 00:06:02,119 --> 00:06:02,139 Yeah. 134 00:06:02,139 --> 00:06:05,446 Let's not kid ourselves about this, but of course, neither has the recording. 135 00:06:05,446 --> 00:06:05,468 Mm-hmm. 136 00:06:05,468 --> 00:06:08,680 So it's, it's always that model of approach. 137 00:06:08,999 --> 00:06:12,674 It, it feels like you have a hard row to hoe in something like Connect 138 00:06:12,703 --> 00:06:16,690 because the human that is calling in to a call center, in almost every 139 00:06:16,690 --> 00:06:20,268 case, is doing it because something has gone wrong for them, where 140 00:06:20,269 --> 00:06:23,353 the, the self-service option on a webpage hasn't gone super well. 141 00:06:23,522 --> 00:06:26,004 Like, the whole challenge of chatbots, where people give demos 142 00:06:26,006 --> 00:06:28,627 of, "And I ask it about my order status." Like, well, you 143 00:06:28,627 --> 00:06:30,871 wouldn't be asking that if your order had shown up as expected. 144 00:06:30,871 --> 00:06:30,885 Yeah. 145 00:06:30,885 --> 00:06:31,632 So what's the deal? 146 00:06:31,633 --> 00:06:34,365 You're already starting with an annoyed customer. 147 00:06:34,674 --> 00:06:34,703 Mm-hmm. 148 00:06:34,703 --> 00:06:38,542 So finding ways to do that, that accelerate the customer journey, that deliver 149 00:06:38,544 --> 00:06:42,200 a better time to resolution without making them feel like they're being pawned 150 00:06:42,201 --> 00:06:47,355 off onto, onto a computer that doesn't care, is always a delicate balance. 151 00:06:47,398 --> 00:06:50,525 Pasquale: Well, and that, that's exactly what we're trying to transform 152 00:06:50,525 --> 00:06:53,162 here in, in, in our releases, and we'll talk about them in a second. 153 00:06:53,222 --> 00:06:56,282 I, I, we can go dive deep into them, um, if you'll, if you allow me, the, uh, 154 00:06:56,301 --> 00:07:00,639 which would be awesome, the, uh- are, are really about a, a, a cumulative thing. 155 00:07:00,640 --> 00:07:02,660 So I can, I'll, I'll talk about the point nature 156 00:07:02,661 --> 00:07:04,197 of them because I think they're really compelling. 157 00:07:04,494 --> 00:07:07,929 But the very first thing is you talked about how 15 years nothing's changed. 158 00:07:08,172 --> 00:07:10,688 The first thing that we launched is this new agentic canvas 159 00:07:10,690 --> 00:07:12,753 that allows you to build these things really quickly. 160 00:07:13,074 --> 00:07:17,117 And what I've seen happen is a lot of the old badness of the industry has 161 00:07:17,118 --> 00:07:20,314 kind of re-come back with agentic solutions that people have been delivering. 162 00:07:20,588 --> 00:07:23,927 You start to see people making them point solutions about getting rid of 163 00:07:23,927 --> 00:07:26,890 customers, deflecting customers, and they'll say stuff like, you know, "Well, 164 00:07:26,890 --> 00:07:29,103 you can even…" They'll even say, "We do outcome-based billing, and the 165 00:07:29,103 --> 00:07:32,280 outcome is we deflected your customer." And I'm like, "Deflected them to who? 166 00:07:32,464 --> 00:07:33,310 Like, ChatGPT? 167 00:07:33,339 --> 00:07:34,031 Do you have better?" 168 00:07:34,031 --> 00:07:35,069 Corey: Oh, yeah, mean time to resolution 169 00:07:35,069 --> 00:07:36,894 when I was working support once upon a time. 170 00:07:36,894 --> 00:07:38,422 You're not allowed to sit through a computer reboot, 171 00:07:38,432 --> 00:07:40,558 so get them off the phone and keep your metrics good. 172 00:07:40,558 --> 00:07:40,588 Yeah. 173 00:07:40,588 --> 00:07:42,643 Whatever you measure, people optimize for, but that 174 00:07:42,643 --> 00:07:44,679 turns out that's not great customer experience. 175 00:07:44,723 --> 00:07:45,003 Pasquale: No. 176 00:07:45,003 --> 00:07:47,221 We've seen, and we've seen all sorts of crazy behaviors. 177 00:07:47,226 --> 00:07:48,250 People hang up on people. 178 00:07:48,254 --> 00:07:50,298 People transfer them back in the pool if they think it's gonna be a 179 00:07:50,299 --> 00:07:53,206 long call because they're being metric'd on the wrong, wrong outcomes. 180 00:07:53,271 --> 00:07:54,946 Corey: The trick is to hang up while you're speaking- Yes … 'cause 181 00:07:54,947 --> 00:07:57,126 only a lunatic hangs up on themselves, so you must have gotten 182 00:07:57,128 --> 00:07:59,786 disconnected prob- Yeah, here this connecting is terrible. 183 00:07:59,798 --> 00:08:00,735 Yeah, great, click. 184 00:08:00,979 --> 00:08:01,304 Yeah. 185 00:08:01,415 --> 00:08:03,212 Pasquale: So that's where I think a lot of these new AI 186 00:08:03,225 --> 00:08:05,732 capabilities are, are interesting is that we can now deliver an 187 00:08:05,732 --> 00:08:08,093 experience that feels good and sounds good, and we were starting 188 00:08:08,093 --> 00:08:10,446 to see a lot of demos of that, um, all across the industry. 189 00:08:10,466 --> 00:08:10,788 Mm-hmm. 190 00:08:10,788 --> 00:08:11,820 But they weren't actually doing anything. 191 00:08:11,820 --> 00:08:13,450 They were a lot more like RAG-based demos. 192 00:08:13,451 --> 00:08:15,676 A lot of them were only chat and to, to be able to be compelling 193 00:08:15,683 --> 00:08:17,581 'cause they weren't fast enough and performant enough. 194 00:08:17,881 --> 00:08:20,525 And so a big part of this release is increasing the quality, the, 195 00:08:20,571 --> 00:08:23,402 the latency on voice, the latency on chat, and also making it so 196 00:08:23,402 --> 00:08:26,639 you can update them quickly, and then that deterministic place 197 00:08:26,640 --> 00:08:29,368 means you can d- drive real outcomes in the back end with them. 198 00:08:29,599 --> 00:08:31,919 So instead of like, you know, "Hey, what's the dress code for casual 199 00:08:31,921 --> 00:08:34,404 Friday?" Which RAG will do great at but probably isn't that useful. 200 00:08:34,644 --> 00:08:36,953 I mean, there are questions people call up for, and they are useful 201 00:08:36,953 --> 00:08:38,832 like, "Where are the directions? What are your hours?" Great. 202 00:08:38,859 --> 00:08:39,834 Let RAG answer those. 203 00:08:40,206 --> 00:08:42,956 But if you're saying, "Hey, I wanna do a balance transfer," all of a 204 00:08:42,959 --> 00:08:45,828 sudden RAG won't be able to do that, and you certainly wouldn't want to 205 00:08:45,829 --> 00:08:50,009 agentically create a balance transfer bot on the fly and then have that act 206 00:08:50,011 --> 00:08:53,023 however it wants to and hallucinate its way through that s- that scenario. 207 00:08:53,235 --> 00:08:55,356 You wanna know the balance transfer goes exactly the 208 00:08:55,356 --> 00:08:57,503 same way every time, follows all the rules, does all the 209 00:08:57,503 --> 00:09:00,338 authentication, and that's something that we bring to bear here. 210 00:09:00,574 --> 00:09:03,304 And now with this new designer, you can actually do that using 211 00:09:03,345 --> 00:09:06,367 natural language, and then we actually allow you to create a 212 00:09:06,367 --> 00:09:09,029 drag and drop view of that where you can inject and bring in 213 00:09:09,213 --> 00:09:12,014 deterministic workflows, so you have the best of both worlds. 214 00:09:12,233 --> 00:09:14,818 And you can update it yourself, so you can make changes that day. 215 00:09:14,818 --> 00:09:17,563 You can press Publish and be good immediately, and 216 00:09:17,564 --> 00:09:19,909 testing stuff, A/B test it, and see what the results are. 217 00:09:20,356 --> 00:09:20,478 Corey: Yeah. 218 00:09:20,486 --> 00:09:23,891 One of the challenges I've always had when I'm calling in to a call center is, 219 00:09:23,934 --> 00:09:27,406 like, after they tell me that their, uh, system options have changed, so pay 220 00:09:27,409 --> 00:09:30,689 attention to it, uh, they, they start trying to drive me back to their website. 221 00:09:30,814 --> 00:09:33,221 But if I'm already on the phone, like, I'm a, I'm 222 00:09:33,221 --> 00:09:35,296 an older millennial, I don't wanna make phone calls. 223 00:09:35,306 --> 00:09:36,396 The, the phone turned on me one day. 224 00:09:36,401 --> 00:09:38,016 It went from a convenience for me to something that started 225 00:09:38,016 --> 00:09:40,982 attacking me, and ever since then, I've been a little leery about it. 226 00:09:40,982 --> 00:09:42,743 But it's the… If I'm calling in, it's 227 00:09:42,744 --> 00:09:45,006 because the, the normal flows don't work. 228 00:09:45,061 --> 00:09:46,825 I am clearly hitting a wall somewhere where 229 00:09:46,825 --> 00:09:48,916 I need to find a person empowered to do this. 230 00:09:49,284 --> 00:09:52,814 If AI can get there to the point where it can solve my problem for 231 00:09:52,818 --> 00:09:57,075 me, terrific, but it's maddening from the customer perspective where, 232 00:09:57,094 --> 00:10:00,757 great, I'm having a weird problem with an order, to use an example. 233 00:10:00,955 --> 00:10:01,348 Great. 234 00:10:01,626 --> 00:10:04,012 Oh, I don't know anything about that, so I'm gonna talk you in 235 00:10:04,016 --> 00:10:07,011 circles and make you prove you're worthy to talk to a human first. 236 00:10:07,012 --> 00:10:08,679 It… I come away from that worse and I 237 00:10:08,680 --> 00:10:10,675 wish I hadn't called at all in some cases. 238 00:10:11,039 --> 00:10:13,839 But done right, if you can empower something agentic 239 00:10:13,846 --> 00:10:16,954 to fix the problem for you, that's an unlock. 240 00:10:17,056 --> 00:10:19,785 Pasquale: The question is, what are the right things to be doing with this? 241 00:10:20,093 --> 00:10:21,734 Uh, you know, obviously, there are gonna be 242 00:10:21,736 --> 00:10:23,354 moments where you don't want a person at all. 243 00:10:23,354 --> 00:10:25,567 If I'm checking my balance, I don't want… I, I don't need a 244 00:10:25,567 --> 00:10:27,434 human being reading my, looking at my bank, and I don't really 245 00:10:27,434 --> 00:10:29,779 want them to, and I don't think most people would want that. 246 00:10:29,829 --> 00:10:32,387 And password changes are annoying for everybody involved, 247 00:10:32,387 --> 00:10:34,096 and certainly a human being is not bringing value to that. 248 00:10:34,096 --> 00:10:34,116 Oh, yes. 249 00:10:34,496 --> 00:10:36,621 And so things like that maybe seem fairly obvious. 250 00:10:36,621 --> 00:10:38,463 On the, on the other side of it, there are things when there are 251 00:10:38,463 --> 00:10:41,186 really delicate, sensitive subjects, like, you know, e- even things 252 00:10:41,189 --> 00:10:43,369 when you're working with, you know, benefits and, and bereavement 253 00:10:43,369 --> 00:10:45,755 and stuff like that, where a human being brings so much value. 254 00:10:46,176 --> 00:10:49,454 In these scenarios, what we really care about is let's make agentic 255 00:10:49,473 --> 00:10:52,539 experience awesome for the ones where a human doesn't bring value. 256 00:10:52,896 --> 00:10:55,882 Let's help the human beings be superhuman in the places where 257 00:10:55,882 --> 00:10:58,811 they do, so they can do things that humans are great at, which 258 00:10:58,813 --> 00:11:00,756 there's a ton of things humans are really, really good at. 259 00:11:01,068 --> 00:11:03,871 We have a lot of customers who don't come to me and tell me, and this is a 260 00:11:03,871 --> 00:11:07,587 change to some degree, but we're trying to get rid of human beings They're 261 00:11:07,588 --> 00:11:11,242 saying, "We want to empower human beings to do great work. In fact, we might 262 00:11:11,248 --> 00:11:14,175 have more people interactions there, but I only want them to be high-level 263 00:11:14,184 --> 00:11:16,561 powerful ones for my customers. I don't want any more of the ones that 264 00:11:16,561 --> 00:11:19,930 aren't helpful and don't build that relationship with my end customer." 265 00:11:20,156 --> 00:11:22,875 Corey: It, it, it comes down to trying to understand where you're 266 00:11:22,875 --> 00:11:25,536 delegating your voice to, uh, uh, where you're delegating judgment, 267 00:11:25,536 --> 00:11:28,049 which is always a dangerous thing to try to industrialize. 268 00:11:28,227 --> 00:11:28,321 Mm-hmm. 269 00:11:28,321 --> 00:11:29,257 You wanna have some guardrails. 270 00:11:29,282 --> 00:11:32,050 "Okay, I wanna do a balance transfer," is a reasonable request. 271 00:11:32,227 --> 00:11:34,766 From someone else's account, maybe less reasonable. 272 00:11:34,776 --> 00:11:35,231 Yeah. 273 00:11:35,231 --> 00:11:36,921 Eh, maybe have a guardrail or two in there. 274 00:11:36,997 --> 00:11:38,603 Pasquale: And the, obviously, the surface areas 275 00:11:38,603 --> 00:11:41,204 are, are becoming more and more, um, interesting. 276 00:11:41,436 --> 00:11:43,680 You've got- You're gonna have agents coming in that are 277 00:11:43,696 --> 00:11:46,416 gonna be trying to abuse the system and, and commit fraud. 278 00:11:46,651 --> 00:11:50,104 The fraud vectors used to be, if I could find one human being who wasn't, who 279 00:11:50,104 --> 00:11:53,304 was willing to break the rules, then I could maybe get some money from them. 280 00:11:53,551 --> 00:11:56,039 But when an agent is working, when agentic AI 281 00:11:56,040 --> 00:11:58,032 is working, that surface area becomes infinite. 282 00:11:58,242 --> 00:12:00,185 It's as many c- phone calls as you can place with 283 00:12:00,188 --> 00:12:01,964 as many agents on the other side talking to it. 284 00:12:02,276 --> 00:12:04,681 So the whole point of this thing is to say, let's not just throw 285 00:12:04,683 --> 00:12:07,258 out the baby and the bathwater with governance and things like 286 00:12:07,258 --> 00:12:10,383 that because these are super important to all these businesses 287 00:12:10,387 --> 00:12:12,959 from both a business standpoint and a fraud standpoint, but also 288 00:12:13,218 --> 00:12:15,494 they have a lot of legal ramifications and things like that. 289 00:12:15,496 --> 00:12:18,084 If you start giving away someone else's money, that's a pretty big deal. 290 00:12:18,092 --> 00:12:18,223 Yeah. 291 00:12:18,477 --> 00:12:22,508 Um, and so these are the things where we say, we, we wanna go, we'll help you 292 00:12:22,510 --> 00:12:26,225 walk through the door into agentic- Um, at the exact speed you want to, knowing 293 00:12:26,226 --> 00:12:29,636 you have not just guardrails in the, in the agentic AI sense, but guardrails, 294 00:12:29,662 --> 00:12:33,169 but also deterministic flows that allow you, and they're gonna feel seamless. 295 00:12:33,274 --> 00:12:34,822 You're not gonna be able to tell when you're switching back 296 00:12:34,823 --> 00:12:36,621 and forth from them because the deterministic things are going 297 00:12:36,621 --> 00:12:39,171 to lead you down a path that's really quite deterministic, 298 00:12:39,475 --> 00:12:41,604 but we're gonna get you there through an agentic experience. 299 00:12:41,834 --> 00:12:43,611 So that ha- you know, the part where it says, "Is that what you 300 00:12:43,612 --> 00:12:45,354 called?" And you say, "Well, no, actually, I wanted to book a 301 00:12:45,354 --> 00:12:47,529 flight next week," or, "No, I wanted to apply a companion pass 302 00:12:47,534 --> 00:12:50,281 to my wife's ticket," those things can take you through that. 303 00:12:50,281 --> 00:12:51,379 And then when you get to the companion 304 00:12:51,379 --> 00:12:53,411 pass, it can still sound just as wonderful. 305 00:12:53,412 --> 00:12:57,045 In fact, we just launched 50 new voices internationally, um, and 306 00:12:57,045 --> 00:13:00,207 these things, they sound great, and that experience is just terrific 307 00:13:00,257 --> 00:13:02,991 while I'm still honoring the things I really care about in the 308 00:13:02,991 --> 00:13:06,264 business that need to be honored, including don't let me apply a 309 00:13:06,294 --> 00:13:09,107 companion pass to somebody else or take away one from somebody else. 310 00:13:09,200 --> 00:13:10,801 Corey: Do you start seeing it the other way, too, where 311 00:13:10,802 --> 00:13:13,381 folks will have agents go and call in to a system? 312 00:13:13,424 --> 00:13:17,277 So you wind up with an agent inbounding to an agent on the recipient 313 00:13:17,278 --> 00:13:21,026 side, and basically, this is just like, um, two APIs talking to 314 00:13:21,026 --> 00:13:23,680 each other with a bunch of tokens as an extra step in the middle. 315 00:13:23,943 --> 00:13:26,902 Pasquale: I think that is the risk, and the thing you always wanna do is make 316 00:13:26,907 --> 00:13:30,318 sure you're building the systems to prevent the vectors being available to them. 317 00:13:30,318 --> 00:13:30,617 Mm-hmm. 318 00:13:30,617 --> 00:13:32,686 You always wanna be the least attractive fraud target. 319 00:13:32,704 --> 00:13:32,890 Corey: Yeah. 320 00:13:32,934 --> 00:13:35,056 Pasquale: And so because once they- 321 00:13:35,065 --> 00:13:36,273 Corey: Well, I'm not even necessarily talking fraud. 322 00:13:36,279 --> 00:13:36,290 Yeah. 323 00:13:36,290 --> 00:13:38,284 I'm talking about, oh, I, I don't wanna make a phone call, and 324 00:13:38,285 --> 00:13:41,024 I'm going to have, I don't know, ChatGPT of the future is gonna go 325 00:13:41,026 --> 00:13:44,112 and call in for me and move my flight from this time to this time. 326 00:13:44,120 --> 00:13:44,499 Go. 327 00:13:44,518 --> 00:13:45,325 Make no mistakes. 328 00:13:45,342 --> 00:13:46,409 Otherwise, it'll make mistakes. 329 00:13:46,414 --> 00:13:46,428 Yeah. 330 00:13:46,629 --> 00:13:47,789 And hope for the best. 331 00:13:47,792 --> 00:13:49,681 But then you… But I'm not talking about necessarily 332 00:13:49,681 --> 00:13:51,658 bad actors, though that's always a concern. 333 00:13:52,111 --> 00:13:54,427 But it does start to be interesting as far as you almost wanna 334 00:13:54,427 --> 00:13:57,889 start baking shibboleths, uh, where, uh, I'm a bot, are you a bot? 335 00:13:57,901 --> 00:13:58,198 Great. 336 00:13:58,198 --> 00:14:01,450 Why don't we just drop down to a highly, uh, optimized 337 00:14:01,474 --> 00:14:03,978 interface format that doesn't require the human niceties 338 00:14:03,978 --> 00:14:05,873 and you have to wonder what the future of that looks like. 339 00:14:05,921 --> 00:14:07,375 Pasquale: I would always joke and say, if people 340 00:14:07,377 --> 00:14:09,055 could just fix their webpages, I'd be out of a job. 341 00:14:09,090 --> 00:14:11,129 And the reality is there's some truth to that. 342 00:14:11,131 --> 00:14:13,083 There's some things where you always wanna talk to a person, but, 343 00:14:13,084 --> 00:14:16,598 like, and, and it's better, but the reality of, of the situation is 344 00:14:16,899 --> 00:14:19,869 the scenarios you're describing I think will exist, but more it is 345 00:14:19,875 --> 00:14:22,112 gonna be really people calling in because they're gonna be calling in 346 00:14:22,112 --> 00:14:24,083 because they have something that's really important at that moment. 347 00:14:24,086 --> 00:14:25,275 And for… I'll give you the example. 348 00:14:25,277 --> 00:14:28,321 Now, maybe there's a world where I've missed that flight. 349 00:14:28,657 --> 00:14:30,745 You can imagine that that's a high-stress time. 350 00:14:30,791 --> 00:14:32,946 People tend to wanna really be hands-on in that scenario 351 00:14:32,946 --> 00:14:34,928 and really make sure what's going on is happening. 352 00:14:35,280 --> 00:14:38,922 Another feature we just launched, actually, United is using ig- in, in real 353 00:14:38,923 --> 00:14:42,725 world right now to, to do this, is this live sync model where it actually 354 00:14:42,725 --> 00:14:46,454 does true synchronization of the app experience and the voice experience. 355 00:14:46,500 --> 00:14:46,830 Mm-hmm. 356 00:14:46,907 --> 00:14:48,982 And so this is something that's just way easier 357 00:14:48,982 --> 00:14:51,001 to do with this scenario for a human being. 358 00:14:51,284 --> 00:14:54,191 If you call in and your flight's been canceled, if the next one is 359 00:14:54,313 --> 00:14:56,627 already booked and you're super happy, then you won't bother to call. 360 00:14:56,641 --> 00:14:56,918 It's great. 361 00:14:56,945 --> 00:14:57,794 I can see it in the app. 362 00:14:57,805 --> 00:14:59,398 But the real problem is like, oh, actually I 363 00:14:59,398 --> 00:15:00,875 need to do a diff- something different now. 364 00:15:01,392 --> 00:15:04,191 Um, they call it irregular operations in, in most airlines, and 365 00:15:04,420 --> 00:15:06,801 it's one of the hardest scenarios because the flights are changing. 366 00:15:06,801 --> 00:15:09,299 The, the ca- there's a bunch of people who just got canceled potentially who 367 00:15:09,299 --> 00:15:12,091 wanna move to someplace else, and in real time they have to deal with this. 368 00:15:12,423 --> 00:15:14,884 And so you can imagine calling in and saying, "Hey, my flight just got 369 00:15:14,884 --> 00:15:18,304 canceled. I've gotta fix this." And then getting a list of flights with 370 00:15:18,314 --> 00:15:22,127 stopovers, it's impossible for your, uh, my brain anyways can't process that. 371 00:15:22,187 --> 00:15:23,891 Corey: Oh, I, I'm a visual, but I need to see some of this. 372 00:15:23,891 --> 00:15:23,914 Yeah. 373 00:15:23,914 --> 00:15:26,579 I can't, like if someone just reads off a list of times, I don't know. 374 00:15:26,781 --> 00:15:30,104 Like, it, it also, getting that synchronization helps the other 375 00:15:30,106 --> 00:15:32,680 problem too, where historically I would get notifications from 376 00:15:32,680 --> 00:15:35,680 Flighty or other services a few minutes before I would get the 377 00:15:35,680 --> 00:15:38,594 United native notification that my flight had been canceled. 378 00:15:38,668 --> 00:15:41,344 Great, I wanna jump on that, and I wanna make that phone call 379 00:15:41,344 --> 00:15:43,743 in because there's sort of an information arbitrage of getting- 380 00:15:43,743 --> 00:15:46,539 Yeah … the push notification first so I beat the rush. 381 00:15:46,925 --> 00:15:49,245 It would be great not to have to worry about that or do it, 382 00:15:49,252 --> 00:15:51,723 where suddenly all the data everyone has is synchronous. 383 00:15:51,877 --> 00:15:54,240 Pasquale: I think the synchronization of this in these multimodal 384 00:15:54,240 --> 00:15:57,540 experiences are gonna be a huge win because as human beings we are 385 00:15:57,606 --> 00:16:00,859 naturally visual, but we have incredible language skills as well, right? 386 00:16:00,862 --> 00:16:03,510 And so bringing those two things to bear I think is super compelling. 387 00:16:03,774 --> 00:16:06,668 And the scenario you had, like I love the fact that, like how long 388 00:16:06,668 --> 00:16:09,245 you take to pick your next flight could impact whether you get it. 389 00:16:09,304 --> 00:16:11,429 When you're talking to this thing and you say, it starts 390 00:16:11,430 --> 00:16:13,329 reading you the flights, it can show you a list of the flights. 391 00:16:13,329 --> 00:16:15,603 Instead you say, "Actually, I just want the third one." Boom, got it. 392 00:16:15,631 --> 00:16:18,197 Okay, confirming you want the third one, and literally 393 00:16:18,246 --> 00:16:20,075 it can be selected on the screen automatically for you. 394 00:16:20,075 --> 00:16:21,794 You don't even have to click it, and then you're good. 395 00:16:21,882 --> 00:16:22,334 You're good. 396 00:16:22,680 --> 00:16:25,061 That, that's a game changer in terms of the stress 397 00:16:25,163 --> 00:16:27,017 experience you were having of trying to get through this 398 00:16:27,017 --> 00:16:29,503 thing as fast as possible and trying to fix your situation. 399 00:16:30,012 --> 00:16:32,151 And, and, like you don't get better at remembering 400 00:16:32,182 --> 00:16:34,159 10 flight options when you're under high stress. 401 00:16:34,367 --> 00:16:35,324 Corey: No, no, generally not. 402 00:16:35,326 --> 00:16:35,483 Yeah. 403 00:16:35,795 --> 00:16:38,483 Uh, you mentioned 50 voices coming, a lot of international. 404 00:16:38,483 --> 00:16:38,637 Yes. 405 00:16:38,650 --> 00:16:40,887 Is that one of those, like genuinely international? 406 00:16:40,892 --> 00:16:42,838 Or is it like you start looking at them and they're all- No … different 407 00:16:42,848 --> 00:16:45,519 accents so that every- everything you call doesn't sound the same. 408 00:16:45,522 --> 00:16:45,555 Yeah, yeah. 409 00:16:45,555 --> 00:16:48,385 Like, "Ooh, this one says drunk and belligerent from New Jersey. 410 00:16:48,395 --> 00:16:50,781 Let's try that." It, yeah, I mean, you could have some great 411 00:16:50,782 --> 00:16:53,611 conversations, but maybe not the one business wants to have. 412 00:16:53,685 --> 00:16:55,392 Pasquale: There was always a newscaster one that 413 00:16:55,396 --> 00:16:56,727 we offered that I thought was kind of funny. 414 00:16:56,727 --> 00:16:58,920 It made it always sound like the s- like, uh, some 415 00:16:58,923 --> 00:17:00,737 familiar voices you might hear at the 6:00 news. 416 00:17:00,738 --> 00:17:02,447 But other than that, they're all really designed for real 417 00:17:02,448 --> 00:17:05,492 world use cases that are mostly in international languages. 418 00:17:05,493 --> 00:17:06,962 I mean, we might have a, we have a few, um, 419 00:17:06,963 --> 00:17:08,899 in, in US English and usually male, female. 420 00:17:08,899 --> 00:17:10,310 We try to have a- Mm … some of those. 421 00:17:10,313 --> 00:17:11,668 And, and obviously the more popular languages are 422 00:17:11,668 --> 00:17:13,470 more likely to have a couple, uh, different options. 423 00:17:13,774 --> 00:17:16,115 But we're definitely moving into more and more interesting 424 00:17:16,115 --> 00:17:18,604 languages, and we have, like we have big customers in Singapore, 425 00:17:18,604 --> 00:17:21,059 and they love the fact that it handles that, that accent 426 00:17:21,070 --> 00:17:23,242 really well and works really well in that space, for example. 427 00:17:23,435 --> 00:17:25,826 I love the fact that we're, be able to expand much, 428 00:17:25,831 --> 00:17:28,244 much faster internationally now with the generative AI. 429 00:17:28,524 --> 00:17:29,966 Um, it, it makes creating these voices 430 00:17:29,967 --> 00:17:31,678 faster, and then we can keep adding to them. 431 00:17:31,858 --> 00:17:33,514 We have to make sure the quality is good too. 432 00:17:33,514 --> 00:17:35,004 So it's not like you just set it and forget it. 433 00:17:35,048 --> 00:17:36,481 We are always constantly evolving and 434 00:17:36,482 --> 00:17:37,804 understanding how our customers are using them. 435 00:17:37,804 --> 00:17:40,900 And, and it's interesting because one of the things you see is in these 436 00:17:40,900 --> 00:17:43,793 different cultures, you know, we have a zip code in the United States. 437 00:17:43,793 --> 00:17:46,229 You're very, you say what the zip code, you know exactly what to expect. 438 00:17:46,259 --> 00:17:47,820 That's not the way zip codes work in UK. 439 00:17:47,821 --> 00:17:51,842 It's not the way they work in, in Singapore and in Japan, in 440 00:17:51,856 --> 00:17:55,572 instead of having houses which are places on a street, streets 441 00:17:55,811 --> 00:17:58,230 are based on houses that happen to have space between them. 442 00:17:58,734 --> 00:17:59,474 utilisez sure more pasrup de ,дляürlich や か に 悉 ,ionate 443 00:17:59,474 --> 00:18:00,164 spoiler am 啦 pertin handczas 芸 術 ándolo allé rind 琼 444 00:18:00,164 --> 00:18:01,033 ods G 要 做 Turk snare incicлишком in니까 workbookжалуйста 445 00:18:01,033 --> 00:18:01,904 Corey: Ceres 捜 времяะนั้น, Ag lowercase/ 泉 佐 遂 lept K 阳 be utilisez 遂 446 00:18:01,904 --> 00:18:02,894 abreast convivial 刊, cib 偏, økgj renੱ sicherlich exporterischem Öğ tros 447 00:18:02,894 --> 00:18:03,934 fyra- .apjauk hoantsiest ぶ ん, ullenARA 据 悉 寸 佈 Art canned tonelle распростран 448 00:18:03,954 --> 00:18:04,774 extracurricular 格 外 объсок 撸 畢 ният 升 の 発 zkod złoż 気 に 入 り. 449 00:18:04,774 --> 00:18:05,533 近 期, 하시면 raindrops sensation obsolete sociableers gole 450 00:18:05,553 --> 00:18:06,524 ବା on Enldigt inanfeet while gera 人 民 政 府 :l đáng 同 志 2 mooticsrotzorin 白 451 00:18:06,524 --> 00:18:07,593 天 whopping 얘는nd タ ー ン airspeedm ningarna precauлишком, bec 澤 ピ ッ ク fruition 452 00:18:07,593 --> 00:18:08,326 [ plump…лишкомпециа…сні in наоборотantin----- incym of 被 打 сні in 捜: 453 00:18:08,326 --> 00:18:09,792 ะนั้น… экземпляandre o 是 一 家 2 mek- sn headway 关 节, き っ か け n 卓 udy… 中 454 00:18:09,792 --> 00:18:11,039 华 人 民 共 和 国 лишком, brochureзей 秘 in Scheinーーiest 放 射 姿 iales sub Saлость 455 00:18:11,039 --> 00:18:11,845 клубаischem, am gymnastein шт./ desolate byeltenобраз, 盲--- lept on 幻 룩 456 00:18:11,845 --> 00:18:12,504 slu or exportermiddels thatлость 出 兴 ющеéisaunque ( Gospelsmiddels 않을,. 457 00:18:12,504 --> 00:18:13,311 [ fakiz Iz Don and [pd [ постанов ( nal b thaterness irloup be. 458 00:18:13,311 --> 00:18:13,824 麒, izde 嫌 lje лице i2రూ to. 459 00:18:13,824 --> 00:18:14,337 out of 祥 in iners - p pre. 460 00:18:14,337 --> 00:18:14,997 , Napole such 恩 стый in hot in nutshell ' and,лишком 461 00:18:14,997 --> 00:18:15,656 fruition lh 可 惜, な く て も whatnot. 462 00:18:15,656 --> 00:18:16,756 unsuitable consum 的 天 献 ilis in Birdsize фанdaukقاتidung in 只 想, lira o1, 甚 463 00:18:16,756 --> 00:18:18,002 vier to nutshellmemeersøk positiest to o p íвающей and/ 情 人 き ゃ in прошло 偷 偷. 464 00:18:18,002 --> 00:18:18,222 lowercase utilisez, lig. 465 00:18:18,222 --> 00:18:19,028 [纵 preschoolers glances dicas と こ ろ に nad 放, subsided,entlicheлишком 466 00:18:19,028 --> 00:18:20,274 白 天 诉 igste 屌 amort 血 管 去 做 Jésusccc deщим よ い よ, 4 이야væ 467 00:18:20,274 --> 00:18:20,861 Podcastssteller alg NASDAQобразнойaukancia 途 T to e delasа 468 00:18:20,861 --> 00:18:21,520 footing DDoSfters XXI dien 記 録 cell/zo [ так abreastiestлишкомånd 469 00:18:21,520 --> 00:18:24,562 l Um, so when you have 50 voices, how do you present that? 470 00:18:24,581 --> 00:18:26,581 Because the o- the old way that I've seen a lot of 471 00:18:26,581 --> 00:18:29,263 interfaces do is they give all of them a first name. 472 00:18:29,263 --> 00:18:29,536 Mm-hmm. 473 00:18:29,553 --> 00:18:31,931 And you're… Like, I'm supposed to know what that means. 474 00:18:32,158 --> 00:18:32,187 Yeah. 475 00:18:32,187 --> 00:18:32,866 Great. 476 00:18:32,877 --> 00:18:36,023 At some point it becomes better to have a descriptor for it. 477 00:18:36,220 --> 00:18:38,210 Other things go too far in the other direction, where 478 00:18:38,211 --> 00:18:40,245 it's voice one, voice two, voice three, voice four. 479 00:18:40,379 --> 00:18:43,360 That doesn't mean anything to me until I listen to all of them. 480 00:18:43,466 --> 00:18:44,267 How do you present that? 481 00:18:44,317 --> 00:18:46,118 Pasquale: We try to do the best of both worlds, but the thing 482 00:18:46,118 --> 00:18:49,210 that's changing radically now is that because you can actually 483 00:18:49,210 --> 00:18:52,783 interact with the system using, using natural language, using 484 00:18:52,787 --> 00:18:56,043 chat, um, that allows you to create these things without having to 485 00:18:56,053 --> 00:18:58,706 be so much of click and drop-down and all, in all of these cases. 486 00:18:58,726 --> 00:19:01,617 Now, you can go in and click and take the drop-downs, but you also just 487 00:19:01,618 --> 00:19:04,469 can create the experience and then immediately experience it yourself. 488 00:19:04,714 --> 00:19:07,430 And so one of the things I love about this is, again, with Connect 489 00:19:07,436 --> 00:19:11,039 being something you can hit go and have it be live immediately, you can 490 00:19:11,041 --> 00:19:13,414 change the voice and actually see what the whole experience is, which 491 00:19:13,414 --> 00:19:15,908 is part of what I think you wanna do in that scenario you described. 492 00:19:16,319 --> 00:19:19,219 You're see- you're experiencing in the real world environment, not just on 493 00:19:19,220 --> 00:19:22,221 a click a button and hear it play back on your computer kind of environment. 494 00:19:22,226 --> 00:19:22,242 Yeah. 495 00:19:22,528 --> 00:19:25,308 And so we, we encourage folks to experiment with it. 496 00:19:25,543 --> 00:19:28,211 Um, most of the folks are pretty happy with the voice we 497 00:19:28,211 --> 00:19:30,083 have, but occasionally we have customers that actually wanna 498 00:19:30,083 --> 00:19:32,047 create their own too, and we can, we support that as well. 499 00:19:32,126 --> 00:19:37,801 Corey: Yeah, we wanna do a Midwestern American with a very bad French accent. 500 00:19:37,839 --> 00:19:38,194 Mm-hmm. 501 00:19:38,194 --> 00:19:38,870 Go. 502 00:19:38,924 --> 00:19:41,206 Uh, it, it, at some point it almost becomes amusing. 503 00:19:41,206 --> 00:19:43,750 It's like, let's turn that off before it accidentally gets into production. 504 00:19:43,750 --> 00:19:46,339 But I can see a story where the horrible voice is the one you use in 505 00:19:46,341 --> 00:19:49,550 the staging environment, so you know that it's the staging environment. 506 00:19:49,550 --> 00:19:49,560 Right. 507 00:19:49,562 --> 00:19:51,609 Like, the reason production is bright red at the prompt. 508 00:19:51,647 --> 00:19:51,671 Yes. 509 00:19:51,671 --> 00:19:53,342 Mm, let's make sure we know what we're touching. 510 00:19:53,415 --> 00:19:55,193 Pasquale: The, the flexibility and the continued 511 00:19:55,195 --> 00:19:57,625 evolution of the, of these voice capabilities is awesome. 512 00:19:57,629 --> 00:20:00,947 Like, I, I feel like with the re- re- release we're doing right now, people 513 00:20:00,949 --> 00:20:03,493 are seeing, like, results in the real world where you're actually able 514 00:20:03,493 --> 00:20:06,588 to make changes in real time that feel good to be interacting with this 515 00:20:06,596 --> 00:20:09,311 thing, and then allow you to move quickly and get through the hard parts. 516 00:20:09,311 --> 00:20:11,480 And when you need to get to a real human, you can get to them too. 517 00:20:11,739 --> 00:20:14,933 And that experience is transformative, but we're not anywhere near done. 518 00:20:15,096 --> 00:20:17,499 We're gonna be… You know, we have a ton of releases still coming through the 519 00:20:17,501 --> 00:20:21,267 rest of the year, and we have, we'll be well into 2027 continuing to build. 520 00:20:21,591 --> 00:20:24,577 I think this is a watershed moment for us where it's a sea shift 521 00:20:24,578 --> 00:20:27,484 in terms of the capabilities 'cause the latencies and the quality 522 00:20:27,488 --> 00:20:29,941 of the voices has gotten so much better with this, this launch. 523 00:20:30,237 --> 00:20:33,192 But it's gonna just keep getting better, and that's the nice thing about 524 00:20:33,210 --> 00:20:36,196 something like Connect, is that Connect customer gets better every day. 525 00:20:36,196 --> 00:20:37,478 We're always shipping new features. 526 00:20:37,674 --> 00:20:39,628 And of course, we respect the fact that people have implemented 527 00:20:39,629 --> 00:20:41,824 stuff on top of it and don't pull the rug out from underneath them. 528 00:20:41,824 --> 00:20:43,925 But- Ah … but we are very careful to make 529 00:20:43,926 --> 00:20:45,462 sure these new features are, are additive. 530 00:20:45,472 --> 00:20:48,013 But, but the additive features are, are compelling. 531 00:20:48,178 --> 00:20:48,399 Corey: Right. 532 00:20:48,399 --> 00:20:49,783 And, and that's part of the challenge, I think. 533 00:20:49,816 --> 00:20:49,826 Yeah. 534 00:20:49,830 --> 00:20:53,069 That, especially as you move up the stack, um- Mm … there are some 535 00:20:53,070 --> 00:20:57,746 things that I know AWS is never going to break without more or less 536 00:20:57,746 --> 00:21:00,511 a hand-delivered engraved educa- in- invitation to every customer. 537 00:21:00,511 --> 00:21:03,041 Like, EC2's APIs will, will never be additive. 538 00:21:03,041 --> 00:21:04,793 And they're like, "Yeah, it turns out that Linux isn't 539 00:21:04,793 --> 00:21:06,938 a thing any…" Or no, that's not going to happen. 540 00:21:07,303 --> 00:21:11,406 But, uh, moving up the stack for things like this, especially at scale for stuff 541 00:21:11,407 --> 00:21:14,432 like this- Mm … when I talk to customers, there's often a, a deep concern 542 00:21:14,473 --> 00:21:18,252 that- Look, this has replaced an awful lot of legacy stuff, and we love it, 543 00:21:18,255 --> 00:21:22,239 but if they change their minds or change the way that we interact with this, 544 00:21:22,257 --> 00:21:25,796 like, we're a serious company, and th- there's a lot of business riding on this. 545 00:21:25,989 --> 00:21:29,740 We need change logs that are months ahead for any breaking changes that hit. 546 00:21:29,773 --> 00:21:33,205 Ad-libbed is fine, but it, it scares them on some level. 547 00:21:33,307 --> 00:21:35,493 Pasquale: The interesting thing about this is the 548 00:21:35,494 --> 00:21:37,840 enterprise is where we started and worked down. 549 00:21:37,852 --> 00:21:40,497 Almost everyone works from small to medium business up. 550 00:21:40,809 --> 00:21:44,024 Um, but in, we, because we had started with such large customers internally, 551 00:21:44,029 --> 00:21:46,996 AWS support, you don't even think about necessarily the size and scale of that. 552 00:21:46,996 --> 00:21:47,653 We've got Audible. 553 00:21:47,653 --> 00:21:49,239 We've got tons of customers that are these 554 00:21:49,242 --> 00:21:52,358 enterprise-grade businesses, Ring, inside of Amazon. 555 00:21:52,648 --> 00:21:55,817 Uh, and those businesses demand these things, too. 556 00:21:56,010 --> 00:21:59,544 So we were well into a birds of a feather with compliance, and we obviously 557 00:21:59,545 --> 00:22:02,318 want to make sure we're always doing the right things with compliance at Amazon. 558 00:22:02,319 --> 00:22:06,491 So certainly for our customers, we are, we're way ahead of the curve on that. 559 00:22:06,507 --> 00:22:09,096 It's one of the reasons why I think so many industries trust us so much. 560 00:22:09,096 --> 00:22:10,948 Like the airlines, the top five airlines by 561 00:22:10,949 --> 00:22:13,314 passenger count all use Amazon Connect at this point. 562 00:22:13,707 --> 00:22:17,906 Um, we've had 23 million interactions a day on Connect. 563 00:22:17,948 --> 00:22:21,521 W- That scale is, I don't think anyone compares to, and certainly when I look 564 00:22:21,522 --> 00:22:24,715 at the customers that I'm aware of out there, uh, from what I can tell, we 565 00:22:24,716 --> 00:22:28,926 have, we have the biggest customers are on Connect of any, any provider because 566 00:22:28,926 --> 00:22:31,697 of that understanding of the enterprise and the need for people to be able to 567 00:22:31,697 --> 00:22:35,275 control their destinies and, and make sure they're really shaping their brand. 568 00:22:35,577 --> 00:22:38,394 And when I think about this, you know, we're reeling back to the 569 00:22:38,395 --> 00:22:40,575 beginning of the conversation, but so much of what we're doing 570 00:22:40,575 --> 00:22:44,338 here is about the brand If you're deflecting your customers away to 571 00:22:44,339 --> 00:22:47,888 something else, that is a bad brand decision, in my opinion, in 2026. 572 00:22:48,169 --> 00:22:50,379 ChatGPT is happy to take your customer and, by the 573 00:22:50,380 --> 00:22:52,266 way, recommend someone else to them potentially, right? 574 00:22:52,309 --> 00:22:53,755 And you don't wanna be in that situation. 575 00:22:53,756 --> 00:22:55,285 You want your customers to love your brand. 576 00:22:55,528 --> 00:22:57,605 And so having a delightful experience, people would 577 00:22:57,606 --> 00:22:59,908 talk about delight and, you know, all these things, but 578 00:22:59,931 --> 00:23:02,135 they were really still concerned a lot with deflection. 579 00:23:02,420 --> 00:23:04,805 I think people still obviously wanna solve problems, that's good. 580 00:23:05,127 --> 00:23:09,628 But let's not deflect, let's solve, let's engage, and then let's follow 581 00:23:09,628 --> 00:23:12,072 up, and let's think about getting ahead of the problem next time. 582 00:23:12,073 --> 00:23:14,546 And next time, instead of telling you when you call up, "Hey, 583 00:23:14,546 --> 00:23:16,935 your flight been booked," I'll send you a text to tell you that. 584 00:23:16,966 --> 00:23:20,000 And by the way, I'll engage with you then if you wanna make a change. 585 00:23:20,090 --> 00:23:22,672 That will cause human beings to feel very different about the 586 00:23:22,672 --> 00:23:25,673 experience, and feeling heard is a huge part of the human existence. 587 00:23:25,673 --> 00:23:28,570 It's just so important to people, and it's gotten lost so many times. 588 00:23:28,877 --> 00:23:31,690 Um, with these high-quality voices, we can make it feel natural. 589 00:23:31,691 --> 00:23:33,885 I don't wanna make it sound like it's a… I don't wanna fake people out. 590 00:23:33,886 --> 00:23:35,132 The goal is to be clear. 591 00:23:35,132 --> 00:23:35,143 Yeah. 592 00:23:35,143 --> 00:23:37,925 You're interacting with something that is designed to help them. 593 00:23:38,101 --> 00:23:39,413 It- but also one, something that really- 594 00:23:39,413 --> 00:23:40,510 Corey: We're sorry your flight's delayed. 595 00:23:40,510 --> 00:23:40,522 Yeah. 596 00:23:40,533 --> 00:23:41,486 The wing fell off. 597 00:23:41,498 --> 00:23:42,500 We're putting it back on. 598 00:23:42,738 --> 00:23:43,088 Yeah, yeah. 599 00:23:43,088 --> 00:23:45,069 Like, um, yeah, and you take all the time you need on that. 600 00:23:45,069 --> 00:23:45,234 Yeah. 601 00:23:45,271 --> 00:23:45,542 Yeah. 602 00:23:45,736 --> 00:23:49,005 So I, I heard something about A2A coming out, where… Which is always fun. 603 00:23:49,006 --> 00:23:51,113 How do we make computers talk to other different computers, 604 00:23:51,134 --> 00:23:53,469 preferably ones in… respond to someone else's cost center? 605 00:23:53,698 --> 00:23:54,338 What is it? 606 00:23:54,788 --> 00:23:56,207 Pasquale: Well, it's super interesting. 607 00:23:56,273 --> 00:24:01,485 Um, A2A is obviously a, a, a, a, a mental model and a, and then a standard 608 00:24:01,488 --> 00:24:04,430 for how you interact, how agents interact within an agentic system. 609 00:24:04,805 --> 00:24:08,940 And it's been obviously proving value to, um, people building agents 610 00:24:08,940 --> 00:24:11,665 outside of contact centers and outside of customer engagement. 611 00:24:12,208 --> 00:24:14,151 We took a look at it and said most of our 612 00:24:14,154 --> 00:24:16,005 competitors seem to be building walled gardens. 613 00:24:16,140 --> 00:24:18,281 Again, like a lot what we saw sort of the battle days, I 614 00:24:18,310 --> 00:24:19,925 think, where you'd have two voices, and people just- That's 615 00:24:19,925 --> 00:24:20,404 Corey: my data. 616 00:24:20,404 --> 00:24:21,598 You're not allowed to touch that. 617 00:24:21,606 --> 00:24:22,079 Great. 618 00:24:22,079 --> 00:24:22,255 Yeah. 619 00:24:22,255 --> 00:24:23,614 Yeah, silos, that'll go well in AI. 620 00:24:23,678 --> 00:24:25,137 Pasquale: Yeah, or even, like, the two voices. 621 00:24:25,137 --> 00:24:27,302 You'd call up, and it'd be like you'd hear Bob saying, "Hey, how 622 00:24:27,302 --> 00:24:29,125 can I help you?" And then you'd say, "Oh, I need to do a balance 623 00:24:29,125 --> 00:24:31,099 transfer." And all of a sudden it'd be like, "Hi, I'm Jill, I'll 624 00:24:31,100 --> 00:24:33,022 help you do a balance transfer." And it was this sort of insa- 625 00:24:33,075 --> 00:24:35,549 insanity, and I felt like we were headed back in that direction. 626 00:24:35,549 --> 00:24:38,213 So I, I challenged my development team, and I said, "Hey, what would 627 00:24:38,213 --> 00:24:41,468 it be to be at the most open solution for people to bring different 628 00:24:41,468 --> 00:24:44,621 agents in?" We're gonna deliver an incredible tool for building 629 00:24:44,862 --> 00:24:47,651 agentic capabilities and experiences in, in Connect Customer. 630 00:24:47,984 --> 00:24:49,843 But what if people wanna bring other ones, either they 631 00:24:49,843 --> 00:24:51,567 build their own or they partner with somebody else? 632 00:24:51,763 --> 00:24:53,999 I said, "I… We wanna be the best of that." And so we actually 633 00:24:53,999 --> 00:24:56,365 took a look at that A2A standard, and we are actually working with 634 00:24:56,367 --> 00:24:59,397 them to evolve that to work better with these voice solutions. 635 00:24:59,658 --> 00:25:02,164 And it's really exciting because it means, like, if you're a 636 00:25:02,164 --> 00:25:04,930 large company, you may have a bunch of agents that do some things 637 00:25:04,931 --> 00:25:07,656 really specific and really well, but the general purpose stuff, 638 00:25:07,699 --> 00:25:09,632 you know, we're gonna do a great job with and help you with. 639 00:25:09,905 --> 00:25:12,683 Bringing those two things together seamlessly, same voice, same 640 00:25:12,686 --> 00:25:16,011 feel, same brand, it, it change… is a game-changer for them. 641 00:25:16,199 --> 00:25:18,281 Um, and so many of my really big customers wanna do that. 642 00:25:18,295 --> 00:25:19,689 The smaller ones, maybe they don't need it, 643 00:25:19,689 --> 00:25:20,848 but the big ones, they're gonna love it. 644 00:25:21,148 --> 00:25:23,987 Corey: When I kicked the tires on Connect a few years ago, it- I 645 00:25:23,988 --> 00:25:25,952 found a couple of things that were interesting, but also at odds. 646 00:25:25,989 --> 00:25:30,077 Uh, one was that I… The pricing is just like any other AWS service. 647 00:25:30,081 --> 00:25:31,640 It's not one of those, you must be at least 648 00:25:31,640 --> 00:25:33,739 this big to actually begin using this. 649 00:25:33,741 --> 00:25:37,263 Like, kicking the tires on it cost me, uh, singles of dollars, 650 00:25:37,390 --> 00:25:40,314 not tens or hundreds of thousands of dollars, which was great. 651 00:25:40,647 --> 00:25:42,634 But so much of the onboarding, the interface 652 00:25:42,635 --> 00:25:44,995 experience was, it felt very enterprise-coded. 653 00:25:44,995 --> 00:25:48,503 It felt like it was very much targeted at large scale-out things, 654 00:25:48,685 --> 00:25:52,122 and at the time I was just trying to build a relatively amusing, 655 00:25:52,123 --> 00:25:55,120 some might say too amusing voice line that we… 'Cause every 656 00:25:55,120 --> 00:25:57,367 company needs to have one, even though most folks don't call it. 657 00:25:57,629 --> 00:25:59,521 Pasquale: You know, I think it's a great point, and 658 00:25:59,533 --> 00:26:01,870 because we were built in that initial thing, and we 659 00:26:01,871 --> 00:26:04,419 also come from, our DNA is we'll always have great APIs. 660 00:26:04,436 --> 00:26:05,638 That's, you know, we're AWS. 661 00:26:06,055 --> 00:26:10,880 But we started in that space, as you said, and we said we want to 662 00:26:10,880 --> 00:26:15,053 make Amazon Connect Customer radically easier to use this year. 663 00:26:15,077 --> 00:26:16,433 That was a goal we set out for. 664 00:26:16,697 --> 00:26:19,533 And so, and the thing is we now have the capabilities, if you think 665 00:26:19,534 --> 00:26:22,826 about the way people use ChatGPT, the way they use Kira, our product, 666 00:26:22,828 --> 00:26:26,698 or Claude, all these different great tools, the idea of having to battle 667 00:26:26,698 --> 00:26:31,187 against some hardened DUI that drag and drop everything is sort of insane. 668 00:26:31,475 --> 00:26:34,019 And so we are doing a ton of launches around 669 00:26:34,210 --> 00:26:36,047 how we just make it so much easier to use this. 670 00:26:36,047 --> 00:26:39,386 So you really think of Amazon Connect Customer as a teammate that's 671 00:26:39,386 --> 00:26:42,757 helping you versus just being a tool that you're trying to poke at. 672 00:26:43,198 --> 00:26:46,311 And that'll come in, like everything else at AWS, we 673 00:26:46,312 --> 00:26:49,057 tend to build things and, and release them over time. 674 00:26:49,093 --> 00:26:52,323 We're continually evolving, but we've got big launches that we're doing. 675 00:26:52,325 --> 00:26:54,794 This agentic voice one I just talked about is a huge one for us. 676 00:26:55,069 --> 00:26:57,105 Um, we are also doing… done a ton of stuff. 677 00:26:57,105 --> 00:27:01,018 We have a, a launch where we just previewed the ability to do configuration 678 00:27:01,033 --> 00:27:04,415 to just ask Connect, "What's, C- Connect customer, what's happening inside 679 00:27:04,417 --> 00:27:07,774 my contact center?" And it can tell you without having to drag and drop 680 00:27:07,774 --> 00:27:11,095 something or do some, you know, command line style arbitrage as you- 681 00:27:11,440 --> 00:27:13,307 Corey: Or even go sit there and monitor calls on the call 682 00:27:13,307 --> 00:27:15,005 center- Yeah … see what the actual ground truth looks like. 683 00:27:15,047 --> 00:27:15,281 Pasquale: Yeah. 684 00:27:15,281 --> 00:27:18,849 And, and that, that's another great example where AI should 685 00:27:18,857 --> 00:27:22,099 be understanding every single contact, every call, every chat 686 00:27:22,101 --> 00:27:25,614 you have, every single second, every time anyone contacts you. 687 00:27:25,878 --> 00:27:28,262 Corey: That's the win that I wanna see more support orgs take. 688 00:27:28,266 --> 00:27:28,293 Yeah. 689 00:27:28,293 --> 00:27:31,863 Where when I call in, they… it should ideally give the, 690 00:27:31,886 --> 00:27:34,495 whoever I'm talking to, agent or human, context on me. 691 00:27:34,511 --> 00:27:36,599 What is the support case history on this account? 692 00:27:36,624 --> 00:27:38,949 Am I a time waster who doesn't seem to understand this stuff? 693 00:27:39,111 --> 00:27:41,319 Do I have a baseline understanding and I just wanna get 694 00:27:41,328 --> 00:27:43,937 one thing unblocked, and I mostly know what's going on? 695 00:27:44,117 --> 00:27:45,799 Am I belligerent as a first language? 696 00:27:45,811 --> 00:27:46,165 Sure. 697 00:27:46,174 --> 00:27:46,821 Probably. 698 00:27:47,074 --> 00:27:49,447 But it's the, like, understanding the case history- 699 00:27:49,551 --> 00:27:51,385 Mm … solves for a lot of it, and it, it means you don't 700 00:27:51,386 --> 00:27:53,276 have to keep asking the same question over and over. 701 00:27:53,498 --> 00:27:56,351 Pasquale: The human data aspect of this cannot be underestimated. 702 00:27:56,352 --> 00:28:01,180 And so last year we pulled into my organization a team called Personalize. 703 00:28:01,187 --> 00:28:04,490 It was building just AI to understand and predict what people would want. 704 00:28:04,754 --> 00:28:06,174 And there's lots of uses for this. 705 00:28:06,182 --> 00:28:07,301 It can improve web pages. 706 00:28:07,301 --> 00:28:08,489 It can improve all sorts of things. 707 00:28:08,489 --> 00:28:09,553 One of the things it can definitely improve 708 00:28:09,553 --> 00:28:12,156 is engagements on, through chat and voice. 709 00:28:12,411 --> 00:28:14,599 It also is great, still great for web pages, but we 710 00:28:14,601 --> 00:28:16,813 built this into a thing we call customer profiles. 711 00:28:17,134 --> 00:28:19,822 And customer profiles allows you to understand and update 712 00:28:19,849 --> 00:28:22,699 every interaction with that customer from all over e- 'Cause 713 00:28:22,699 --> 00:28:24,540 you've got silos all over your enterprise most likely. 714 00:28:24,540 --> 00:28:25,766 Most- Yeah … companies just do. 715 00:28:26,144 --> 00:28:28,345 And so pulling all that data into one place and then having it be 716 00:28:28,501 --> 00:28:31,246 ultra-performant, you know, we're seeing early data now that shows 717 00:28:31,250 --> 00:28:34,081 it's kind of revolutionary for people to just bring that data into the 718 00:28:34,081 --> 00:28:36,658 experience and say, "Hey, how should I think about this customer when 719 00:28:36,659 --> 00:28:39,858 they call in?" It, it radically simplifies, again, trying to get that 720 00:28:39,859 --> 00:28:43,433 radical simplification, it radically simplifies the, your ability to 721 00:28:43,433 --> 00:28:46,553 understand your customer, and as you said, that changes everything. 722 00:28:46,851 --> 00:28:49,323 I'm delighted to see this, like, come to fruition. 723 00:28:49,352 --> 00:28:52,490 This is, like, I will say eight years ago, when we started this 724 00:28:52,491 --> 00:28:55,003 journey, when I started, the team had been around for longer than that. 725 00:28:55,005 --> 00:28:57,143 They, uh, they were a great… The incredible engineers 726 00:28:57,145 --> 00:28:59,444 are all still pretty much with the team now even. 727 00:28:59,682 --> 00:29:03,044 Um- I was like, I had this vision we could do this, and the 728 00:29:03,045 --> 00:29:05,243 changes that have happened in AI in the last couple years have been 729 00:29:05,244 --> 00:29:08,778 like, "Hey, we really can do it now." And so it's fundamentally 730 00:29:08,782 --> 00:29:11,897 changing the way people interact with Amazon Connect Customer. 731 00:29:11,897 --> 00:29:14,510 It's fundamentally changing the way end customers interact with businesses. 732 00:29:14,879 --> 00:29:18,120 This is that sea shed, watershed moment that I think we've been waiting for. 733 00:29:18,318 --> 00:29:20,237 Uh, I don't know if you can tell, but I'm pretty excited about it. 734 00:29:20,815 --> 00:29:23,259 Corey: It, it feels like you are, if not there yet, you're nearing 735 00:29:23,260 --> 00:29:25,803 a tipping point, and I'm very curious to see what happens next. 736 00:29:26,231 --> 00:29:28,386 Thank you so much for taking the time to speak with me. 737 00:29:28,415 --> 00:29:31,475 If people wanna learn more, where can they go to figure out how 738 00:29:31,475 --> 00:29:34,583 Connect might solve some problems and make others just plain funnier? 739 00:29:34,874 --> 00:29:37,367 Pasquale: Yeah, I mean, you can go to the AWS website, but I would say 740 00:29:37,371 --> 00:29:40,292 just do a search for Amazon Connect Customer, and you will find us there. 741 00:29:40,319 --> 00:29:42,945 Uh, w- we are releasing new things every day. 742 00:29:42,945 --> 00:29:45,033 We've got, we've got ton more releases coming this 743 00:29:45,035 --> 00:29:47,204 year, so, uh, I'm, I'm happy to come back and talk more, 744 00:29:47,204 --> 00:29:49,285 but, but it's been a great time talking to you today. 745 00:29:49,354 --> 00:29:50,105 Corey: An absolute pleasure. 746 00:29:50,108 --> 00:29:51,215 Thank you so much for your time. 747 00:29:51,218 --> 00:29:51,866 I appreciate it. 748 00:29:51,931 --> 00:29:52,304 Pasquale: Thank you. 749 00:29:52,645 --> 00:29:54,528 Corey: I'm Corey Quinn, this is Pasquale, and 750 00:29:54,538 --> 00:29:56,952 this of course remains Screaming in the Cloud. 751 00:29:57,103 --> 00:29:57,570 Stick around.