1 00:00:00,180 --> 00:00:02,220 We started to feel we were onto something and we 2 00:00:02,220 --> 00:00:04,950 felt like we were onto something in in two parts. 3 00:00:05,565 --> 00:00:10,275 One of the parts was when we were pitching with Rewind, it felt differentiated 4 00:00:10,275 --> 00:00:14,655 and like it gave people like the safety blanket essentially, or that they didn't 5 00:00:14,655 --> 00:00:18,615 really have with, um, with their AI systems, in fact, pretty consistently. 6 00:00:18,615 --> 00:00:20,384 One of the things that I heard was, oh, I 7 00:00:20,384 --> 00:00:22,034 didn't even know that this would be possible. 8 00:00:22,424 --> 00:00:25,485 Just to give you like the 10 seconds on what AI Agent Rewind is. 9 00:00:25,784 --> 00:00:28,185 Rubrik maintains backups of the most important production 10 00:00:28,185 --> 00:00:30,525 systems that an organization or enterprise has. 11 00:00:30,975 --> 00:00:34,815 Agent Rewind says if an agent operates on those systems and makes a mistake. 12 00:00:35,045 --> 00:00:38,015 Delete something, it shouldn't have edited a field in the wrong way. 13 00:00:38,345 --> 00:00:40,925 We can allow you to correlate those agents' actions with 14 00:00:40,925 --> 00:00:42,965 the actual change to production system and allow you to 15 00:00:42,965 --> 00:00:45,995 recover in one click, uh, from that healthy snapshot. 16 00:00:46,055 --> 00:00:48,695 So that's like the agent re want pitch and it gave people a lot of safety. 17 00:00:49,025 --> 00:00:50,135 That was one thing we learned. 18 00:00:50,615 --> 00:00:51,765 But the second thing we learned was. 19 00:00:52,425 --> 00:00:54,765 This ability to go ahead and recover was still 20 00:00:54,765 --> 00:00:57,555 a little bit of a a future problem, right? 21 00:00:57,555 --> 00:00:59,415 It gives 'em comfort that allows them to go 22 00:00:59,415 --> 00:01:00,885 ahead and do something that's coming out. 23 00:01:01,365 --> 00:01:04,395 But in order to build rewind, we also had to build 24 00:01:04,395 --> 00:01:07,245 a deeper understanding of the agent's systems. 25 00:01:07,245 --> 00:01:09,585 We call it the agent map and the agent registry. 26 00:01:10,310 --> 00:01:12,320 Like we had to know, well, what agents are running inside of your 27 00:01:12,320 --> 00:01:14,870 ecosystem so we can figure out which ones we might need to rewind. 28 00:01:15,259 --> 00:01:18,410 And we need to know what are the types of things that they have access to? 29 00:01:18,410 --> 00:01:20,899 What actions can they take so we can rewind a 30 00:01:20,899 --> 00:01:22,910 given action that it actually ended up taking. 31 00:01:27,920 --> 00:01:29,660 Welcome to Screaming in the Cloud. 32 00:01:29,810 --> 00:01:30,800 I'm Corey Quinn. 33 00:01:31,125 --> 00:01:35,535 This promoted guest episode is brought to us by our friends at Rubrik. 34 00:01:35,715 --> 00:01:41,205 Also brought to us by our friends at Rubrik is Dev Richi, their GM of AI Dev. 35 00:01:41,415 --> 00:01:42,315 Thanks for joining me. 36 00:01:42,495 --> 00:01:42,975 Hey, Corey. 37 00:01:43,035 --> 00:01:43,365 Thanks. 38 00:01:43,365 --> 00:01:44,385 I'm looking forward to being here. 39 00:01:44,895 --> 00:01:47,355 Let's be honest about disaster recovery. 40 00:01:47,685 --> 00:01:52,185 Your DR plan assumes everything fails gracefully and your backups are pristine. 41 00:01:52,545 --> 00:01:55,995 But what happens when ransomware's already in your backups or 42 00:01:56,145 --> 00:01:59,805 when the credentials your Dr. Runbook depends on are compromised? 43 00:02:00,105 --> 00:02:02,805 Most DR strategies are built for accidents. 44 00:02:02,860 --> 00:02:04,240 Not adversaries. 45 00:02:04,510 --> 00:02:05,920 Rubrik gets this. 46 00:02:06,130 --> 00:02:08,830 They isolate your backups from production credentials, 47 00:02:09,039 --> 00:02:11,920 make them immutable and can scan them for threats so you're 48 00:02:11,920 --> 00:02:15,295 not restoring malware along with your data when your Dr. 49 00:02:15,295 --> 00:02:17,950 Day comes, and it inevitably will. 50 00:02:18,190 --> 00:02:20,470 You need more than a runbook and hope you 51 00:02:20,470 --> 00:02:23,020 need backups that are actually recoverable. 52 00:02:23,290 --> 00:02:27,190 Learn more@rubrik.com slash SI tc. 53 00:02:28,155 --> 00:02:33,345 You are a somewhat recent newcomer over to Rubrik. 54 00:02:33,345 --> 00:02:35,565 Before this, you figured, you know what I'm 55 00:02:35,565 --> 00:02:37,995 gonna do, I'm gonna start an AI company. 56 00:02:37,995 --> 00:02:40,125 But you, you did it the hipster style while 57 00:02:40,125 --> 00:02:42,045 it was still underground, before it was cool. 58 00:02:42,195 --> 00:02:43,155 What's the backstory there? 59 00:02:43,665 --> 00:02:43,935 Yeah. 60 00:02:43,935 --> 00:02:44,835 Honestly, I'm not sure. 61 00:02:44,835 --> 00:02:47,115 AI was never, uh, it was ever not cool. 62 00:02:47,175 --> 00:02:48,765 I remember even when we started the company 63 00:02:48,765 --> 00:02:52,035 in 2021, I thought it was one of the most. 64 00:02:52,750 --> 00:02:56,049 I actually thought at the time most of AI was a little bit overvalued. 65 00:02:56,140 --> 00:02:57,910 Um, I thought it was overhyped. 66 00:02:58,209 --> 00:03:00,489 And one of the reasons I thought that was I had 67 00:03:00,489 --> 00:03:02,739 spent a long time as a product manager at Google. 68 00:03:02,739 --> 00:03:04,540 I was the first PM for Kaggle, which is the 69 00:03:04,540 --> 00:03:06,160 data science machine learning community. 70 00:03:06,850 --> 00:03:10,149 And I saw this massive influx of, I think like citizen 71 00:03:10,149 --> 00:03:12,130 data scientists, people that were learning a lot about 72 00:03:12,130 --> 00:03:14,350 machine learning and ai and we're very excited about it. 73 00:03:15,160 --> 00:03:17,799 But, um, I was one of the product managers on the team that eventually 74 00:03:17,799 --> 00:03:21,640 became called Vertex ai, uh, Google Cloud's machine learning platform. 75 00:03:22,245 --> 00:03:26,385 And candidly, very few enterprises were actually getting value, 76 00:03:26,505 --> 00:03:29,265 uh, I think out of enter, uh, ai and very few had actually 77 00:03:29,265 --> 00:03:32,295 even figured out how to go into production with those systems. 78 00:03:32,475 --> 00:03:32,625 Yeah. 79 00:03:32,625 --> 00:03:36,045 My line for a while back in that era was that machine learning can sort 80 00:03:36,045 --> 00:03:40,635 through vast quantities of data and discover anything except a business model. 81 00:03:40,855 --> 00:03:41,125 Yeah, 82 00:03:41,485 --> 00:03:42,895 I think that's about accurate. 83 00:03:42,895 --> 00:03:45,175 You know, when we looked at companies that were actually making money in 84 00:03:45,175 --> 00:03:48,535 this space, it was data labeling companies like Scale ai, I think back 85 00:03:48,535 --> 00:03:52,075 in the day was making a killing, but their core business was not actually 86 00:03:52,075 --> 00:03:54,955 delivering necessarily production ready AI models for the enterprise. 87 00:03:55,015 --> 00:03:56,515 It was just labeling data. 88 00:03:56,575 --> 00:04:00,025 And so I started the company in 2021, or along with a few 89 00:04:00,025 --> 00:04:02,965 of my co-founders, because we just believed in two things. 90 00:04:03,385 --> 00:04:04,285 The first was. 91 00:04:04,525 --> 00:04:06,205 This technology is powerful. 92 00:04:06,295 --> 00:04:09,355 It isn't delivered on that promise yet, but we saw what 93 00:04:09,355 --> 00:04:11,365 it was able to do at leading organizations like Google or 94 00:04:11,365 --> 00:04:13,735 YouTube or you know, Uber, where my co-founders came from. 95 00:04:14,275 --> 00:04:16,584 And then the second was we thought we had an abstraction 96 00:04:16,584 --> 00:04:18,295 that would make it a little bit more accessible. 97 00:04:18,535 --> 00:04:20,815 You know, we'd seen a lot of companies die on this hill of trying 98 00:04:20,815 --> 00:04:23,215 to democratize machine learning and democratize deep learning. 99 00:04:23,515 --> 00:04:25,105 We thought we'd take our stab at it as well. 100 00:04:25,255 --> 00:04:28,915 Yeah, I remember back then the, the use cases that they came up with were 101 00:04:28,915 --> 00:04:33,655 either highly specific to the point of uselessness for anyone else or. 102 00:04:34,325 --> 00:04:34,985 Banal. 103 00:04:35,135 --> 00:04:39,125 Uh, the two examples that stuck in my mind were if you are a credit 104 00:04:39,125 --> 00:04:43,145 card, uh, company and you have a massive transaction volume, you 105 00:04:43,145 --> 00:04:46,055 can start to identify fraud via the power of machine learning. 106 00:04:46,295 --> 00:04:46,745 Great. 107 00:04:46,865 --> 00:04:47,735 I'm not that. 108 00:04:47,945 --> 00:04:51,275 The other one was, I think WeWork did, uh, machine learning to analyze 109 00:04:51,275 --> 00:04:55,235 traffic patterns and wound up, up discovering that they could reduce 110 00:04:55,235 --> 00:04:58,715 them if they had a second barista, uh, at certain hours of the day. 111 00:04:58,925 --> 00:04:59,585 In other words. 112 00:05:00,219 --> 00:05:01,750 Humans like to drink coffee in the morning 113 00:05:01,750 --> 00:05:03,909 was their amazing discovery out of this. 114 00:05:04,180 --> 00:05:07,330 And it, it felt like it was a really interesting 115 00:05:07,330 --> 00:05:10,120 space that people struggled to articulate value from. 116 00:05:10,330 --> 00:05:16,150 Then we saw this massive gen AI explosion, uh, over the past few years and. 117 00:05:16,370 --> 00:05:18,620 Everyone is doing some experiment with it. 118 00:05:18,620 --> 00:05:21,200 Some folks are rapidly rebranding as fast as they can to be 119 00:05:21,200 --> 00:05:25,370 AI companies, but the question of value seems to be one that 120 00:05:25,370 --> 00:05:29,960 hangs over the industry as a whole, just because it's terrific. 121 00:05:29,960 --> 00:05:32,210 When it gets it right, it often doesn't. 122 00:05:32,480 --> 00:05:35,870 There's are distinct costs associated with it, and people are 123 00:05:35,870 --> 00:05:39,440 still trying to figure out how exactly this factors into the thing 124 00:05:39,440 --> 00:05:42,770 that they're doing Has been my experience from talking to folks. 125 00:05:42,860 --> 00:05:44,390 Am I missing something key? 126 00:05:44,390 --> 00:05:44,600 What? 127 00:05:44,690 --> 00:05:45,985 How are you seeing the industry evolve? 128 00:05:46,620 --> 00:05:48,990 I want to answer that in just a second, but the first 129 00:05:48,990 --> 00:05:50,520 thing you mentioned actually was really interesting. 130 00:05:50,520 --> 00:05:52,920 You talked about how use cases back when we started the company 131 00:05:52,920 --> 00:05:55,530 in 21 were pretty banal, or you know, relatively consistent. 132 00:05:55,920 --> 00:05:56,790 I had the same thought. 133 00:05:56,790 --> 00:05:59,070 I wanna talk about technology for a a minute 134 00:05:59,070 --> 00:06:00,930 and then go into this generative AI section. 135 00:06:01,590 --> 00:06:03,840 One of the things that I remember, like lamenting that 136 00:06:03,840 --> 00:06:05,940 actually back at that time was if you looked at every 137 00:06:05,940 --> 00:06:07,985 one of our competitors' websites at a use case section. 138 00:06:08,835 --> 00:06:11,835 Every competitor's use case section looked like the exact same thing. 139 00:06:12,164 --> 00:06:13,664 There's a churn prediction model. 140 00:06:13,784 --> 00:06:16,635 There was an LTV prediction model, there was 141 00:06:16,635 --> 00:06:18,795 a fraud for, uh, you know, detection model. 142 00:06:18,795 --> 00:06:20,715 And there's like all these use cases of look at all 143 00:06:20,715 --> 00:06:22,370 the amazing things you can build with the platform. 144 00:06:23,090 --> 00:06:23,330 I think. 145 00:06:23,830 --> 00:06:28,000 Our insight at the time was that there was a newer technology 146 00:06:28,000 --> 00:06:30,640 and that the power of generative AI was gonna be unlocked. 147 00:06:30,730 --> 00:06:32,740 Uh, or sorry, we didn't, it wasn't generative AI at the time. 148 00:06:32,745 --> 00:06:33,610 It was called deep learning. 149 00:06:34,240 --> 00:06:36,880 Uh, and you know, our insight was that, number one, we 150 00:06:36,880 --> 00:06:39,610 think that it'll work really well with unstructured data. 151 00:06:40,030 --> 00:06:42,190 So if you think about like fraud or churn, there were a lot 152 00:06:42,190 --> 00:06:45,010 about structured data problems, and we were really excited 153 00:06:45,010 --> 00:06:47,470 about unstructured data, raw texts and images and video. 154 00:06:48,015 --> 00:06:50,534 Then the second I think, insight was this 155 00:06:50,534 --> 00:06:52,755 rise of what we called pre-trained models. 156 00:06:53,115 --> 00:06:55,784 Um, and so these are models that you didn't have to have a million 157 00:06:55,784 --> 00:06:58,425 records to be able to go ahead and build a data science team and 158 00:06:58,425 --> 00:07:01,185 clean the dataset on, but they were pre-trained so you could just 159 00:07:01,185 --> 00:07:04,335 sort of adapt what generally understood English towards your task. 160 00:07:04,335 --> 00:07:07,305 Those were like two of the things that we decided to invest in with 161 00:07:07,305 --> 00:07:09,945 deep learning and we set out on a mission to democratize deep learning. 162 00:07:09,945 --> 00:07:12,495 And then I like to say open AI and large language models 163 00:07:12,495 --> 00:07:14,715 really democratize deep learning better than any of us. 164 00:07:15,140 --> 00:07:18,320 But then you asked the question of like outside of consumer, which is 165 00:07:18,320 --> 00:07:21,260 I think where generative AI has probably like delivered a lot of value. 166 00:07:21,260 --> 00:07:22,909 I would argue for users today. 167 00:07:23,610 --> 00:07:26,880 What is the value that enterprises are seeing right in G 2K? 168 00:07:27,090 --> 00:07:29,730 And I'd, I partially agree with your observation 169 00:07:29,730 --> 00:07:31,950 like, are people getting a, like ROI or not out of it? 170 00:07:32,550 --> 00:07:35,820 And I actually think that the, what's been surprising to me 171 00:07:35,850 --> 00:07:40,440 coming into Rubrik is the reason why I think organizations 172 00:07:40,470 --> 00:07:43,320 aren't actually getting nearly as much value yet. 173 00:07:43,520 --> 00:07:45,530 As I think has been forecasted, and I believe that 174 00:07:45,530 --> 00:07:48,320 they will get out in a few years at Preta base. 175 00:07:48,320 --> 00:07:50,690 You know, the startup that, uh, we worked on for the last four and 176 00:07:50,690 --> 00:07:53,690 a half years, we worked with a lot of digital native and leading AI 177 00:07:53,690 --> 00:07:56,990 engineering organizations, household brands you would know, uh, you 178 00:07:56,990 --> 00:07:59,390 know, were the ones that were deploying production models with us. 179 00:08:00,240 --> 00:08:02,400 I would say the biggest difference between them and coming to 180 00:08:02,400 --> 00:08:05,700 Rubrik were like the customer base is Global 2000 Enterprise. 181 00:08:05,700 --> 00:08:07,770 Think about the most important regulated 182 00:08:08,010 --> 00:08:09,870 customers and financial services in healthcare. 183 00:08:10,170 --> 00:08:11,700 The biggest difference is actually on risk 184 00:08:11,700 --> 00:08:14,881 posture, and that is converted downstream into ROI. 185 00:08:15,465 --> 00:08:17,685 The reason I think a lot of organizations haven't gotten the 186 00:08:17,685 --> 00:08:20,145 type of value that they want under A ROI yet is they haven't 187 00:08:20,145 --> 00:08:22,845 figured out the framework where they're going to let AI loose 188 00:08:22,905 --> 00:08:26,265 in terms of like the actual work that will produce value 189 00:08:26,895 --> 00:08:29,925 at certain points of scale, companies become less about seizing 190 00:08:29,925 --> 00:08:32,625 opportunities and more about risk mitigation and management. 191 00:08:32,925 --> 00:08:34,455 And when you have something that. 192 00:08:34,549 --> 00:08:37,250 Gets it right 80% of the time, and the other 20% 193 00:08:37,309 --> 00:08:40,520 goes off the rails to greater or lesser degrees. 194 00:08:40,880 --> 00:08:44,900 That becomes almost an unbounded risk vector and that I feel like 195 00:08:44,900 --> 00:08:48,290 that's why people are taking a very, okay, how, how do we put guardrails 196 00:08:48,290 --> 00:08:51,229 around this thing so that it doesn't destroy the company we have built? 197 00:08:51,465 --> 00:08:53,385 Totally like, uh, let me ask you, what do you think is 198 00:08:53,385 --> 00:08:56,565 like the biggest enterprise AI trend of 2025 or 2026? 199 00:08:56,565 --> 00:08:59,175 I think the most hyped one is probably agentic ai, right? 200 00:08:59,175 --> 00:09:02,835 Like agents is the, the thing everybody wants to talk about. 201 00:09:03,315 --> 00:09:06,045 Even getting a definition of what agent is, is becoming, 202 00:09:06,045 --> 00:09:09,055 uh, a. A exercise and tell me what you have to sell me. 203 00:09:09,295 --> 00:09:12,834 Uh, everyone is defining it in ways that align with their view of the world. 204 00:09:12,834 --> 00:09:16,495 We saw similar things in the observability space in recent cycles. 205 00:09:16,765 --> 00:09:19,525 Let me offer you a definition without something we sell 206 00:09:19,525 --> 00:09:21,385 you, because we don't sell an agent builder platform. 207 00:09:21,385 --> 00:09:23,605 We don't like make it, you know, something you build agents with. 208 00:09:24,115 --> 00:09:28,584 I always think about agents as just LMS or models with access to tools. 209 00:09:28,615 --> 00:09:30,175 And so if you think about that, that really just 210 00:09:30,175 --> 00:09:32,365 means a model that can do work or take an action. 211 00:09:32,625 --> 00:09:33,855 Um, you know, on your behalf. 212 00:09:34,425 --> 00:09:37,485 And, uh, I think this comes back towards this idea of like, at a 213 00:09:37,485 --> 00:09:40,215 certain point organizations have to be about like risk mitigation. 214 00:09:40,725 --> 00:09:43,125 If you think about like a lot of different software trends, the shift to 215 00:09:43,125 --> 00:09:47,655 cloud as an example, there is a huge amount of, um, I think nervousness at the 216 00:09:47,655 --> 00:09:50,655 beginning of the shift to cloud, predominantly around security and guardrails 217 00:09:50,655 --> 00:09:53,655 and like, you know, this idea of we're gonna go to a multi-tenant architecture 218 00:09:54,165 --> 00:09:57,255 in some way or the other, rather than like on my on-prem data center. 219 00:09:58,005 --> 00:10:00,015 Uh, and now with ai, what we're talking about with 220 00:10:00,135 --> 00:10:01,665 agentic AI is like, look, it's gonna be great. 221 00:10:01,665 --> 00:10:02,595 It's gonna be able to do work. 222 00:10:02,595 --> 00:10:03,765 You're gonna be able to take action. 223 00:10:04,095 --> 00:10:06,855 And I know every it, and like security person I'm talking to 224 00:10:06,855 --> 00:10:09,194 is like, wait, so you wanna give a non-deterministic model? 225 00:10:09,405 --> 00:10:14,295 Meaning model that will, you know, uh, not necessarily operate within a certain 226 00:10:14,295 --> 00:10:17,295 defined framework, something that can come up with random answers at times. 227 00:10:17,295 --> 00:10:20,415 You wanna give a non-deterministic model access to my production 228 00:10:20,415 --> 00:10:23,355 and enterprise systems, and you want me to be on the hook with it? 229 00:10:23,900 --> 00:10:24,410 No way. 230 00:10:24,680 --> 00:10:27,350 Uh, and so while I think it has a lot of different promises, 231 00:10:27,710 --> 00:10:31,100 a lot of the organizations that I think I speak to are kind 232 00:10:31,100 --> 00:10:34,850 of still stuck on like square one of how do I get comfortable 233 00:10:35,060 --> 00:10:37,280 with the idea of this running around inside of my ecosystem? 234 00:10:37,755 --> 00:10:40,245 Yeah, it feels like every vibe coding project 235 00:10:40,245 --> 00:10:42,375 you start begins with, okay, roll for initiative. 236 00:10:42,705 --> 00:10:45,795 And it you, it always goes slightly differently depending 237 00:10:45,795 --> 00:10:49,455 upon the fates, for lack of a better term, which is 238 00:10:49,725 --> 00:10:52,845 great for some use cases and terrifying for others. 239 00:10:53,490 --> 00:10:55,830 But I think like, coming to your point about ROI, I 240 00:10:55,830 --> 00:10:58,200 actually, by the way, I didn't know this three months ago. 241 00:10:58,410 --> 00:11:01,290 Um, you know, when I was working, uh, and we were acquired 242 00:11:01,290 --> 00:11:04,380 into Rubrik about three to four months ago when I was working 243 00:11:04,380 --> 00:11:07,770 within, um, production AI systems and a lot of leading tech 244 00:11:07,770 --> 00:11:10,590 companies, we thought that there was a series of challenges that. 245 00:11:10,610 --> 00:11:12,560 Typically had to do with latency or throughput 246 00:11:12,560 --> 00:11:13,910 and, you know, the efficiency of models. 247 00:11:14,510 --> 00:11:17,540 But over the last two to three months, um, I've had a 248 00:11:17,540 --> 00:11:20,990 chance to speak with about 180 to 200 different customers 249 00:11:20,990 --> 00:11:23,540 representing, you know, all sorts of swats from the Global 2K. 250 00:11:24,320 --> 00:11:27,080 And the thing that was really interesting to me is that 251 00:11:27,140 --> 00:11:29,360 we speak a lot about the promise of what AI could unlock. 252 00:11:29,840 --> 00:11:31,250 It can go ahead and do work on our behalf. 253 00:11:31,250 --> 00:11:33,470 Imagine, imagine that wrote task you had to do to be able to 254 00:11:33,470 --> 00:11:35,630 prepare for an interview or to be able to send out an email. 255 00:11:35,840 --> 00:11:36,980 You can actually go ahead and read through 256 00:11:36,980 --> 00:11:38,450 the systems and write that out for you. 257 00:11:39,330 --> 00:11:42,630 But then in practice, if you think about what type of agent or what type of 258 00:11:42,630 --> 00:11:45,840 AI every single organization is actually rolling out today, for the ones that 259 00:11:45,840 --> 00:11:50,250 actually even are, um, first of all, they're all in read-only mode, right? 260 00:11:50,250 --> 00:11:54,360 Like very few people are giving agents what we call like write or delete access. 261 00:11:54,360 --> 00:11:56,160 Very few agents, uh, people are giving agents 262 00:11:56,160 --> 00:11:58,200 the ability to like actually edit a system. 263 00:11:59,070 --> 00:12:00,990 And it's not because they can't think of the use case, and 264 00:12:00,990 --> 00:12:03,270 it's not because the business value argument isn't there. 265 00:12:03,990 --> 00:12:07,110 But it's because it feels like the risk is almost uncapped. 266 00:12:07,440 --> 00:12:09,630 There's unlimited downside really in it for them. 267 00:12:09,750 --> 00:12:13,470 I mean, I do myself in the test lab, but worst case, something blows up it. 268 00:12:13,530 --> 00:12:15,900 It's not that hard to restore from backups. 269 00:12:15,900 --> 00:12:17,880 Turns out data resilience is a thing. 270 00:12:18,210 --> 00:12:20,190 There's a Yeah, but the idea of doing this with. 271 00:12:20,285 --> 00:12:22,055 Production customer side data. 272 00:12:22,265 --> 00:12:25,204 I, because my laptop has theoretical access to customer 273 00:12:25,204 --> 00:12:28,535 environments, I can't let Claude code run loose on this thing. 274 00:12:28,745 --> 00:12:33,125 I give it a bounded EC2 instance in a dedicated AWS account. 275 00:12:33,125 --> 00:12:35,675 Worst case can blow up my budget, but that's the end of it. 276 00:12:36,005 --> 00:12:38,915 It's not, it has no access to data that could cause me a n. 277 00:12:39,570 --> 00:12:40,410 Yeah, exactly. 278 00:12:40,410 --> 00:12:42,900 We, um, I spoke with our head of InfoSec yesterday, right? 279 00:12:42,900 --> 00:12:46,920 And he was talking about how, look, there are people inside of our 280 00:12:46,920 --> 00:12:51,420 company that have quote unquote, super, super user admin privileges. 281 00:12:51,420 --> 00:12:54,990 They can do incredible amounts of damage, but those people, 282 00:12:54,990 --> 00:12:57,570 number one, have all been background checked by the company. 283 00:12:57,840 --> 00:13:00,480 Number two, there's like all of these guardrails that 284 00:13:00,480 --> 00:13:02,820 are put, put in place, you know, sock that observes them. 285 00:13:03,120 --> 00:13:04,950 And number three, and maybe the most important. 286 00:13:05,330 --> 00:13:08,120 Like from his perspective, they operate at human pace. 287 00:13:08,510 --> 00:13:10,700 So you know, the amount of actions that you can go ahead and take, you 288 00:13:10,700 --> 00:13:13,760 could probably blow up your AWS cosman, you know, uh, reasonably easily. 289 00:13:13,820 --> 00:13:14,060 Yeah. 290 00:13:14,060 --> 00:13:15,080 Or compensating controls. 291 00:13:15,080 --> 00:13:18,020 The way we handle this in polite society, the, the bank 292 00:13:18,020 --> 00:13:21,710 teller theoretically can enrich themselves at your expense. 293 00:13:21,860 --> 00:13:25,130 The reason this doesn't happen is because there are audit controls 294 00:13:25,430 --> 00:13:28,280 and security flags and alarms that will go off everywhere. 295 00:13:28,280 --> 00:13:29,900 The second, something like that happens. 296 00:13:29,990 --> 00:13:30,260 Yeah. 297 00:13:30,320 --> 00:13:32,660 And there's probably like 30 to 40 years of software that's 298 00:13:32,660 --> 00:13:34,655 been developed around validating the employee and the human. 299 00:13:35,385 --> 00:13:38,324 And there's a certain pace that a human can go ahead and make a damage on. 300 00:13:38,745 --> 00:13:40,905 I would say with agents, you know, the line that we use 301 00:13:40,905 --> 00:13:43,185 internally is, well, they could do 10 x or a hundred x 302 00:13:43,185 --> 00:13:45,824 to damage in a hundred, like yeah, a 10th of the time. 303 00:13:46,185 --> 00:13:49,605 The pace at which I think the operation is changing is something 304 00:13:49,605 --> 00:13:53,715 that there isn't really a resilience infrastructure for today. 305 00:13:53,985 --> 00:13:55,910 Um, and I think that's what's stopping a lot of the RI. 306 00:13:56,625 --> 00:13:56,845 It. 307 00:13:56,850 --> 00:13:58,500 It feels like a lot of the value that AI 308 00:13:58,500 --> 00:14:00,030 has generated comes at the personal level. 309 00:14:00,030 --> 00:14:01,860 You alluded to some of it recently where we're 310 00:14:01,860 --> 00:14:04,230 talking about, oh, respond to this email. 311 00:14:04,230 --> 00:14:08,160 For me, like I have what I like to politely call the asshole and email problem. 312 00:14:08,310 --> 00:14:10,050 I tend to write relatively tersely. 313 00:14:10,050 --> 00:14:11,430 It's a bad habit I get from Twitter. 314 00:14:11,430 --> 00:14:11,490 It. 315 00:14:11,870 --> 00:14:16,460 And it turns out that short emails look like I'm being imperious, so make 316 00:14:16,460 --> 00:14:19,580 this polite, but I still have to iterate through it a couple of times. 317 00:14:19,580 --> 00:14:23,570 It starts off with, I hope this email finds you well, which is not my style. 318 00:14:23,570 --> 00:14:24,800 I'm likelier to begin with. 319 00:14:24,800 --> 00:14:28,340 I hope this email finds you before I do, because if that's at least my brand. 320 00:14:28,740 --> 00:14:31,800 And you have to tweak it and all, and like there's a bit of a human 321 00:14:31,800 --> 00:14:34,830 in the loop story and it's helpful, but it's not, you know, I'm 322 00:14:34,830 --> 00:14:37,920 not gonna pay $5,000 a month of personal money for these things. 323 00:14:37,920 --> 00:14:40,530 It's, it's helpful, but it doesn't necessarily 324 00:14:40,530 --> 00:14:43,260 justify a lot of the investment on the Yeah. 325 00:14:43,260 --> 00:14:47,490 On the enterprise side, that story may look radically different. 326 00:14:47,790 --> 00:14:49,800 I have a number of customers. 327 00:14:49,885 --> 00:14:53,395 In effectively all of them who are doing some experiments with ai, the 328 00:14:53,395 --> 00:14:57,175 ones that are spending meaningfully on it, those use cases start to 329 00:14:57,175 --> 00:15:01,645 look a lot more like, uh, the B2C aspects of what it is that they do. 330 00:15:01,975 --> 00:15:06,985 The B2B companies that I work with are using it, uh, obviously for 331 00:15:06,985 --> 00:15:10,495 a number of things, but their spend is nowhere within orders of 332 00:15:10,495 --> 00:15:13,405 magnitude of what we're seeing when it starts getting mass deployed. 333 00:15:13,825 --> 00:15:17,455 How are you seeing the global Fortune 2000 largely emerge? 334 00:15:17,455 --> 00:15:19,525 As far as trends go when it comes to ai? 335 00:15:19,850 --> 00:15:23,150 I'm seeing a little bit of a bifurcation, and so I'm seeing a leading edge 336 00:15:23,180 --> 00:15:27,080 of companies and I'm talking about Global 2000 Enterprise still right now. 337 00:15:27,080 --> 00:15:27,320 Right? 338 00:15:27,535 --> 00:15:29,115 But I'm seeing a leading edge of companies. 339 00:15:29,850 --> 00:15:32,460 That are fully leaning in to ai. 340 00:15:32,520 --> 00:15:36,540 And I, I, you know, I spoke with a fortune hundred company, uh, that walked 341 00:15:36,540 --> 00:15:40,290 me through how they were thinking about, uh, things at a board level. 342 00:15:40,770 --> 00:15:43,800 And they were going to brand their company, which is, you know, uh, 343 00:15:44,310 --> 00:15:48,000 household name decades and decades old into an AI native company. 344 00:15:48,510 --> 00:15:50,819 Uh, and I sat there and I was like, this is incredible. 345 00:15:50,850 --> 00:15:52,770 Like this company which has. 346 00:15:53,505 --> 00:15:55,935 More employees than you know, any other company I can think of and 347 00:15:55,935 --> 00:15:59,535 like has such a large distribution network is now thinking about 348 00:15:59,535 --> 00:16:02,295 how to be able to go AI native, thinking about how to be able to 349 00:16:02,295 --> 00:16:05,444 do agents and other workflows for every single type of use case. 350 00:16:06,240 --> 00:16:09,570 And they're really looking like they're taking lead a thousand 351 00:16:09,570 --> 00:16:12,480 flowers bloom, but actually investing behind every single one 352 00:16:12,480 --> 00:16:14,970 of those flowers to be able to see where am I gonna get value? 353 00:16:15,360 --> 00:16:18,330 Simple reason why they think that, you know, in five to 10 354 00:16:18,330 --> 00:16:21,450 years, their workforce is gonna look massively different, right? 355 00:16:21,450 --> 00:16:23,160 Like they think the entire way that the work happens. 356 00:16:23,580 --> 00:16:27,480 I would say like five to 10% of companies that I speak to are in that bucket. 357 00:16:28,260 --> 00:16:31,710 And I think 90% of organizations that I speak to. 358 00:16:32,084 --> 00:16:36,135 Are in the bucket of kind of like a let's wait and see posture. 359 00:16:36,495 --> 00:16:38,204 Let's go ahead and run a few different experiments. 360 00:16:38,204 --> 00:16:39,735 Let's have a center of excellence. 361 00:16:39,824 --> 00:16:43,964 Um, and like, let's start to in, uh, experiment with across both 362 00:16:43,964 --> 00:16:46,245 of them though, the one thing that I see that's really interesting 363 00:16:46,605 --> 00:16:50,295 is that typically your first handful of use cases are the ones 364 00:16:50,295 --> 00:16:54,285 that take the longest use cases are agents one through five. 365 00:16:54,735 --> 00:16:55,785 Take a lot of doing. 366 00:16:55,995 --> 00:16:59,235 There's a lot that you have to be able to establish from a framework standpoint. 367 00:16:59,235 --> 00:17:01,125 How do you measure ROI, whose approvals are in there? 368 00:17:01,485 --> 00:17:02,685 What tasks are they good at? 369 00:17:02,685 --> 00:17:04,065 Which do they struggle with? 370 00:17:04,244 --> 00:17:06,675 But the fascinating thing, Corey, is how quickly 371 00:17:06,675 --> 00:17:09,704 organizations go from five to hundreds or thousands. 372 00:17:10,155 --> 00:17:13,694 When we were thinking about like, what would rubrics play with an AI be, 373 00:17:13,694 --> 00:17:16,454 we had a huge question of like, when is this moment gonna come when people 374 00:17:16,454 --> 00:17:19,665 are going to have, you know, hundreds or thousands of agents deployed? 375 00:17:19,665 --> 00:17:22,095 And our thought process, well, okay, well is this gonna be. 376 00:17:22,454 --> 00:17:26,535 12, 24, 36, uh, months away, we think it's going 377 00:17:26,535 --> 00:17:28,274 to happen, but what's the timeframe over it? 378 00:17:28,754 --> 00:17:30,915 The thing that's really been interesting is like a set of 379 00:17:30,915 --> 00:17:34,245 conversations we had, we think like maybe, you know, 1, 4, 1 and 380 00:17:34,245 --> 00:17:37,365 five or so where people tell us, oh, that's already happening today. 381 00:17:37,605 --> 00:17:41,355 I'm actually seeing it kind of scale out, and again, it's only the, I'd say 382 00:17:41,355 --> 00:17:46,245 leading quartile at best that are in that bucket, but you can kind of see those 383 00:17:46,250 --> 00:17:49,425 early adopter motions that we think the rest of the market's going to follow. 384 00:17:50,205 --> 00:17:53,085 So I think, you know, the very brief like summary for 385 00:17:53,085 --> 00:17:55,034 what are we seeing in the enterprise, what we're seeing 386 00:17:55,155 --> 00:17:58,665 most people in that one to five experimentation route. 387 00:17:59,115 --> 00:18:00,885 But what we see is the people that graduate 388 00:18:00,885 --> 00:18:02,570 from it start to scale up very, very quickly. 389 00:18:03,930 --> 00:18:06,180 And I think that there's a misunderstanding around a lot of it 390 00:18:06,180 --> 00:18:09,240 too, where if I'm reaching out to support, for example, from a 391 00:18:09,240 --> 00:18:11,700 company I do business with, I don't wanna talk to a chat bot. 392 00:18:11,730 --> 00:18:14,820 However, it would be convenient if the human that I talked 393 00:18:14,820 --> 00:18:19,080 to has, uh, an AI assisted context on their side of, 394 00:18:19,110 --> 00:18:21,090 oh, here are the previous support tickets I've opened. 395 00:18:21,120 --> 00:18:24,300 Oh, it looks like I might actually know how networks work. 396 00:18:24,300 --> 00:18:25,530 So is it plugged in? 397 00:18:25,530 --> 00:18:27,450 Might not be the first thing you lead with. 398 00:18:27,450 --> 00:18:30,180 For me, it starts to tailor the responses there. 399 00:18:30,490 --> 00:18:34,450 And that's, that is a neat, transformative customer experience. 400 00:18:34,690 --> 00:18:37,720 But so often it shortens to, oh, we can lay off our 401 00:18:37,720 --> 00:18:40,660 entire support staff, which generally is not happening. 402 00:18:40,900 --> 00:18:43,690 Every time companies have tried this, it seems to have gone disastrously. 403 00:18:44,070 --> 00:18:46,050 Yeah, I think that, um, you know, you said 404 00:18:46,290 --> 00:18:47,580 you don't wanna shout to the shop bot. 405 00:18:47,580 --> 00:18:48,720 How many times have you been on a phone and 406 00:18:48,720 --> 00:18:50,160 been like, speak to an agent, speak to an agent. 407 00:18:50,160 --> 00:18:50,790 Speak to an agent. 408 00:18:50,850 --> 00:18:52,110 I'm already pissed off 'cause I don't want, 409 00:18:52,110 --> 00:18:53,730 I'm, I'm a millennial, I'm an older millennial. 410 00:18:53,730 --> 00:18:55,680 I don't want to talk to people on the phone. 411 00:18:55,680 --> 00:18:58,620 If I did, I would take out a personal's ad. I great if I'm 412 00:18:58,680 --> 00:19:00,870 calling in and something has already gone off the rails. 413 00:19:01,105 --> 00:19:01,314 Yeah. 414 00:19:01,314 --> 00:19:02,274 Corey, I'm a millennial. 415 00:19:02,274 --> 00:19:03,955 I'll say like, I don't wanna speak to somebody 416 00:19:03,955 --> 00:19:05,935 when I have a support issue almost, period. 417 00:19:05,935 --> 00:19:08,514 I just want it solved as fast as possible, right? 418 00:19:08,725 --> 00:19:10,435 Like, I don't wanna, I don't wanna speak to a chatbot. 419 00:19:10,465 --> 00:19:11,845 I also don't really wanna spend a long time 420 00:19:11,845 --> 00:19:14,304 explaining to a human what exactly is happening. 421 00:19:14,365 --> 00:19:17,905 I will use a chatbot on my side because I wanna explain, here's the, here's the. 422 00:19:18,160 --> 00:19:20,260 Two hours of logging data that I have for this. 423 00:19:20,560 --> 00:19:23,470 Can you skim this down to a concise ticket 424 00:19:23,470 --> 00:19:25,750 that shows a skeleton reproduction case? 425 00:19:26,020 --> 00:19:27,580 I used to have to do that by hand. 426 00:19:27,580 --> 00:19:29,650 That was how I fixed things and invariably the chat 427 00:19:29,650 --> 00:19:32,770 bot will often fix it for me while doing that behavior. 428 00:19:32,920 --> 00:19:34,150 It's incredibly helpful. 429 00:19:34,480 --> 00:19:36,940 It's, but the support tickets of it broke. 430 00:19:37,300 --> 00:19:38,950 Not the most helpful thing in the universe. 431 00:19:39,540 --> 00:19:42,420 Let me connect it to what I think we need to be able to actually tie it 432 00:19:42,420 --> 00:19:45,720 together, which is, um, a lot of chat bots are, you know, they're interactive. 433 00:19:45,720 --> 00:19:47,550 The conversations, they're almost, they almost 434 00:19:47,550 --> 00:19:49,140 feel like they're incentivized to keep you talking. 435 00:19:49,530 --> 00:19:53,130 What I want to be able to do is get to the fastest action possible. 436 00:19:53,400 --> 00:19:56,190 I want that system, which today, the reason I wanna speak to 437 00:19:56,190 --> 00:20:00,240 human is to today the person who can issue me my refund, or who 438 00:20:00,240 --> 00:20:03,990 can, you know, cancel the order, take that action is a human. 439 00:20:05,265 --> 00:20:08,595 I believe, I fundamentally believe that organizations are gonna get 440 00:20:08,595 --> 00:20:11,025 to the point where those actions are gonna be done autonomously. 441 00:20:11,115 --> 00:20:13,965 They are gonna be done by like an agent or AI system. 442 00:20:14,265 --> 00:20:15,435 In order for that agent to be able to do 443 00:20:15,435 --> 00:20:16,665 it, it needs to be connected with that tool. 444 00:20:16,665 --> 00:20:18,315 It needs to be connected to the database that 445 00:20:18,315 --> 00:20:20,685 manages order cancellations as an example. 446 00:20:21,405 --> 00:20:23,160 Um, and I think where you're gonna see that. 447 00:20:23,430 --> 00:20:27,030 Flipping value happen from like, Hey, the AI driven chatbot is something 448 00:20:27,030 --> 00:20:30,270 that's driving me nuts to, oh, this was a way better experience 449 00:20:30,270 --> 00:20:32,820 than others, is when those chatbots start to get access to tools. 450 00:20:32,820 --> 00:20:36,390 There's a lot of organizations they already have, but the type of 451 00:20:36,390 --> 00:20:39,570 organizations that have have a kind of risk forward posture still today. 452 00:20:40,530 --> 00:20:44,250 Rubrik is sponsoring this segment because they figured out something important. 453 00:20:44,460 --> 00:20:47,850 Protecting your AI isn't just about backing up databases. 454 00:20:48,000 --> 00:20:51,420 It's about protecting the entire supply chain, the training 455 00:20:51,420 --> 00:20:54,030 data that teaches your models, the source code that 456 00:20:54,030 --> 00:20:56,790 defines them, and the live data that they're acting on. 457 00:20:57,150 --> 00:20:59,040 Miss any piece of that, and you're basically 458 00:20:59,040 --> 00:21:00,835 rebuilding from scratch when things break. 459 00:21:01,324 --> 00:21:05,284 Rubrik secures all of it in one platform across your multi-cloud circus. 460 00:21:05,554 --> 00:21:08,435 So when your AI inevitably does something creative, you 461 00:21:08,435 --> 00:21:12,095 didn't expect you can recover without losing months of work. 462 00:21:12,304 --> 00:21:15,574 That's the difference between resilience and resume updating. 463 00:21:15,995 --> 00:21:18,814 Learn more at rubrik.com/si. 464 00:21:18,814 --> 00:21:19,835 Tc. 465 00:21:20,675 --> 00:21:22,445 There's a lot of. 466 00:21:23,760 --> 00:21:25,140 I ought say optimism in the space. 467 00:21:25,140 --> 00:21:28,440 There's a lot of hype as well where companies are suddenly 468 00:21:28,440 --> 00:21:30,900 trying to wrap the exact same thing that they've been doing for 469 00:21:30,900 --> 00:21:36,090 15 years in the AI story, despite the fact that they did not 470 00:21:36,090 --> 00:21:40,860 revamp their bestselling product to completely be AI native. 471 00:21:41,220 --> 00:21:43,950 Uh, and the departure from what it previously was over an 472 00:21:43,950 --> 00:21:47,040 18 month span like that, that would be in some cases lunacy. 473 00:21:47,250 --> 00:21:49,860 Um, in some cases it's, oh, great, we're an AI company. 474 00:21:49,860 --> 00:21:50,700 Like that's great. 475 00:21:50,895 --> 00:21:55,574 I thought you were a bus company, but there's this idea of being 476 00:21:55,574 --> 00:22:00,405 able to spill, to spill this out in, in ways that are, that are 477 00:22:00,405 --> 00:22:03,284 iterative and transformative and do lead to better outcomes. 478 00:22:03,435 --> 00:22:06,375 Which I guess brings us to Rubrik on some level here. 479 00:22:06,375 --> 00:22:10,665 I, why did they acquire you and what is, what are you 480 00:22:10,665 --> 00:22:13,425 doing these days now that you're the GM of AI over there? 481 00:22:13,665 --> 00:22:14,534 Yeah, it's a great question. 482 00:22:14,534 --> 00:22:16,455 I'll answer them in two parts, right? 483 00:22:16,905 --> 00:22:17,955 I think the first is. 484 00:22:18,500 --> 00:22:20,180 Well, why make the acquisition? 485 00:22:20,330 --> 00:22:24,470 Um, and I, Rubrik is a company, um, co-founder and CEO BIP Bull, 486 00:22:24,470 --> 00:22:27,290 I think has defined, Rubrik is a company that's been perpetually. 487 00:22:27,840 --> 00:22:30,570 That the market's been perfectly confused how to bucket it, right? 488 00:22:30,570 --> 00:22:33,990 Like it started off in like a core on data backup, and then 489 00:22:33,990 --> 00:22:36,810 data protection and cyber resilience really became a larger 490 00:22:36,810 --> 00:22:39,899 market as we saw the rise of ransomware and other attacks. 491 00:22:40,320 --> 00:22:42,030 So it went away from like natural disaster, 492 00:22:42,030 --> 00:22:43,800 fire, flood for why you need the technology. 493 00:22:44,630 --> 00:22:47,270 I think that the, um, kind of executive and 494 00:22:47,270 --> 00:22:51,320 founding team here have a long-term ambition in ai. 495 00:22:51,800 --> 00:22:54,980 The reason that they have a long-term ambition in AI is because the view is that 496 00:22:54,980 --> 00:22:58,640 one of the most fundamental kind of like substrates for feeding into AI models 497 00:22:58,640 --> 00:23:02,450 or others, is like, what is the actual customer data that is backing it up? 498 00:23:02,690 --> 00:23:06,050 And Rubrik is one of the largest pre-populated data lakes for 499 00:23:06,050 --> 00:23:08,480 every single enterprise customer that it backs up and protects. 500 00:23:08,930 --> 00:23:12,290 All of the most important data that is gonna be important for our customers, 501 00:23:12,290 --> 00:23:16,310 like, you know, business resilience applications and day-to-day operations that 502 00:23:16,310 --> 00:23:20,330 our, that Arik customer is backed up really by like our underlying systems. 503 00:23:20,810 --> 00:23:24,020 So the first observation that I think the rubric, um, exec team had 504 00:23:24,020 --> 00:23:26,870 motivating the acquisition was like, Hey, we think data is gonna be like 505 00:23:26,870 --> 00:23:29,630 a fundamental asset in AI and we wanna have a unique right to play here. 506 00:23:29,930 --> 00:23:31,880 And then the second piece was like, I think 507 00:23:31,880 --> 00:23:33,560 how that fit into the Preta base platform. 508 00:23:33,620 --> 00:23:34,915 And you know, what we did as a technology. 509 00:23:35,580 --> 00:23:39,330 What we did, just as like a, you know, uh, I'd say the 32nd overview 510 00:23:39,330 --> 00:23:42,629 was our favorite customer quote is, generalized intelligence might be 511 00:23:42,629 --> 00:23:45,959 great, but I don't need my point of sale system to recite French poetry. 512 00:23:46,440 --> 00:23:49,500 We were targeting enterprise applications where you needed something narrow and 513 00:23:49,500 --> 00:23:52,440 specific done, and we wanted to help you be able to build out that application. 514 00:23:52,980 --> 00:23:56,580 So the thesis was what Rubrik had in terms of the data, uh, uh, 515 00:23:56,610 --> 00:23:59,940 substrate, and then more recently the identity component as well. 516 00:24:00,120 --> 00:24:01,530 That helped you understand who has access 517 00:24:01,530 --> 00:24:03,990 to what could be paired with our platform. 518 00:24:04,140 --> 00:24:07,470 That gave you, you know, the ability to think about more tailored applications 519 00:24:07,470 --> 00:24:11,340 with models, uh, and we could build a more resilient enterprise story. 520 00:24:11,820 --> 00:24:13,710 And so that was really, I think, you know, what brought 521 00:24:13,710 --> 00:24:16,650 us in at, um, at like, I would say 80% of the story. 522 00:24:16,710 --> 00:24:19,500 And I think 20% of the story is just the great kind of, I 523 00:24:19,500 --> 00:24:23,220 think, cultural fit that also existed across, um, the two teams. 524 00:24:23,790 --> 00:24:26,790 And then, uh, fast forwarding to today, what 525 00:24:26,790 --> 00:24:28,710 is it exactly that, you know, I'm doing here? 526 00:24:29,385 --> 00:24:32,625 I like to kind of say that I don't have any, 527 00:24:32,775 --> 00:24:33,645 I don't have any idea. 528 00:24:33,645 --> 00:24:34,575 I don't know either. 529 00:24:34,575 --> 00:24:35,535 Don't tell anyone. 530 00:24:35,595 --> 00:24:36,345 Kidding, kidding. 531 00:24:36,690 --> 00:24:37,245 No, no, no. 532 00:24:37,245 --> 00:24:38,385 I, yeah, actually, very true. 533 00:24:38,535 --> 00:24:41,445 I would say when I come into like a newer market, we had our customer base 534 00:24:41,445 --> 00:24:44,775 and now I'm coming into newer market, the rubric customer base, which is 535 00:24:44,775 --> 00:24:49,095 tends to be IT, and security in large parts of global 2000 enterprises. 536 00:24:49,440 --> 00:24:52,170 I think I walked in with the assumption, I don't know, like I actually 537 00:24:52,170 --> 00:24:54,270 don't know what the right way is to take the product that we have 538 00:24:54,270 --> 00:24:57,210 and the product that Rubrik has and be able to start to, you know, 539 00:24:57,210 --> 00:25:01,050 retrofit it towards what would the future of this specific market needs. 540 00:25:01,889 --> 00:25:04,770 So the only way I know how to answer that question is to get 541 00:25:04,770 --> 00:25:07,590 out and talk to as many people and customers that are in the 542 00:25:07,590 --> 00:25:10,080 field and hear from them what are they actually struggling with. 543 00:25:10,649 --> 00:25:13,350 So in the last two and a half months, um, I've had a chance 544 00:25:13,350 --> 00:25:16,950 to speak with a little over like 180, um, organizations. 545 00:25:17,400 --> 00:25:18,960 And, you know, many, many different people 546 00:25:18,960 --> 00:25:20,670 inside of each of those organizations. 547 00:25:20,670 --> 00:25:24,420 Typically IT security, but also you can think of everywhere 548 00:25:24,420 --> 00:25:26,940 from the backup admin, all the way to the person who's heading 549 00:25:26,940 --> 00:25:29,880 AI inside of those organizations and understanding what is 550 00:25:29,880 --> 00:25:31,950 it that they're actually, you know, uh, struggling with. 551 00:25:32,760 --> 00:25:35,310 And, uh, we, you know, it's hard to go ahead 552 00:25:35,310 --> 00:25:37,200 and have a super open-ended conversation. 553 00:25:37,440 --> 00:25:40,590 So we came in with an initial pitch that we called Agent Rewind. 554 00:25:41,295 --> 00:25:43,185 Agent Rewind connected what's happening 555 00:25:43,185 --> 00:25:46,004 with AI to rubric's core around resilience. 556 00:25:46,035 --> 00:25:48,705 It was this idea that if agents are gonna do 10 x damage in one 10th a 557 00:25:48,705 --> 00:25:52,665 time, what if we allowed you to revert back a destructive agent action? 558 00:25:53,175 --> 00:25:56,295 We started to kind of have a, you know, we had these conversations that were. 559 00:25:56,699 --> 00:25:59,580 Very genuinely not sales oriented, in part because we, 560 00:25:59,669 --> 00:26:01,919 you know, the, the product was still too early to sell 561 00:26:01,919 --> 00:26:03,689 at the time that I was having these conversations. 562 00:26:04,020 --> 00:26:05,129 I love those conversations. 563 00:26:05,129 --> 00:26:06,899 I just wish I could trust it a little bit more. 564 00:26:06,899 --> 00:26:09,689 When people say, I'm not trying to sell you anything, and you have 565 00:26:09,689 --> 00:26:13,110 those conversations and it quickly transitions into a sales pitch, 566 00:26:13,110 --> 00:26:16,469 it's, yeah, but if I say it's a sales pitch, you won't take my call. 567 00:26:16,469 --> 00:26:19,379 It's, and you think lying to me is gonna lead you to a better outcome. 568 00:26:19,679 --> 00:26:20,520 I digress. 569 00:26:20,570 --> 00:26:21,379 Please continue. 570 00:26:22,040 --> 00:26:25,010 I had the opposite experience because I think people were like, well, when can I 571 00:26:25,010 --> 00:26:27,260 go ahead and try this and when can I, like, you know, how much is this gonna be? 572 00:26:27,260 --> 00:26:30,379 And I was like, we gotta, we gotta slow you down a little bit there. 573 00:26:30,560 --> 00:26:31,669 That's, you know, you're onto something. 574 00:26:32,120 --> 00:26:32,929 Yeah, exactly. 575 00:26:32,929 --> 00:26:33,919 That's that. 576 00:26:34,250 --> 00:26:37,399 But what you said is actually what we all started to feel, we started to feel 577 00:26:37,399 --> 00:26:41,570 we were onto something and we felt like we were onto something in, in two parts. 578 00:26:42,210 --> 00:26:46,890 One of the parts was when we were pitching with Rewind, it felt differentiated 579 00:26:46,890 --> 00:26:51,270 and like it gave people like the safety blanket essentially, or that they didn't 580 00:26:51,270 --> 00:26:55,260 really have with, um, with their AI systems, in fact, pretty consistently. 581 00:26:55,260 --> 00:26:57,000 One of the things that I heard was, oh, I 582 00:26:57,000 --> 00:26:58,620 didn't even know that this would be possible. 583 00:26:59,040 --> 00:27:02,100 Just to give you like the 10 seconds on what AI Agent Rewind is. 584 00:27:02,400 --> 00:27:04,800 Rubrik maintains backups of the most important production 585 00:27:04,800 --> 00:27:07,140 systems that an organization or enterprise has. 586 00:27:07,590 --> 00:27:11,430 Agent Rewind says if an agent operates on those systems and makes a mistake. 587 00:27:11,675 --> 00:27:14,675 Delete something, it shouldn't have edited a field in the wrong way. 588 00:27:14,975 --> 00:27:17,554 We can allow you to correlate those agents' actions with 589 00:27:17,554 --> 00:27:19,564 the actual change to production system and allow you to 590 00:27:19,564 --> 00:27:22,625 recover in one click, uh, from that healthy snapshot. 591 00:27:22,685 --> 00:27:25,324 So that's like the agent re want pitch and it gave people a lot of safety. 592 00:27:25,655 --> 00:27:26,764 That was one thing we learned. 593 00:27:27,215 --> 00:27:28,395 But the second thing we learned was. 594 00:27:29,080 --> 00:27:31,389 This ability to go ahead and recover was still 595 00:27:31,389 --> 00:27:34,210 a little bit of a a future problem, right? 596 00:27:34,210 --> 00:27:36,040 It gives 'em comfort that allows them to go 597 00:27:36,040 --> 00:27:37,510 ahead and do something that's coming out. 598 00:27:37,990 --> 00:27:41,020 But in order to build rewind, we also had to build 599 00:27:41,020 --> 00:27:43,870 a deeper understanding of the agent's systems. 600 00:27:43,870 --> 00:27:45,855 We call it the agent map and the agent registry. 601 00:27:46,580 --> 00:27:48,979 All the, like, we had to know, well, what agents are running inside of your 602 00:27:48,979 --> 00:27:52,340 ecosystem so we can figure out which ones we might need to rewind, and we need 603 00:27:52,340 --> 00:27:55,699 to know what are the types of things that they have access to, what actions 604 00:27:55,699 --> 00:27:59,540 can they take so we can rewind a given action that it actually ended up taking. 605 00:27:59,600 --> 00:28:04,610 Yeah, it turns out natively there's no, uh, undo for an RMRF or a drop table. 606 00:28:05,129 --> 00:28:09,000 Natively, there's no undo for a drop table, especially when it's like an 607 00:28:09,000 --> 00:28:12,810 MCP tool call made from some agent, you know, that sits upstream towards it. 608 00:28:13,260 --> 00:28:13,470 Oh God. 609 00:28:13,470 --> 00:28:15,000 We've all had the thing where it's just, oh, 610 00:28:15,000 --> 00:28:16,830 this file looks, I'm just gonna overwrite it. 611 00:28:16,830 --> 00:28:18,510 Like that was important. 612 00:28:18,510 --> 00:28:20,639 What are you doing important to the project? 613 00:28:20,639 --> 00:28:20,939 Not. 614 00:28:20,990 --> 00:28:24,440 I rev not, not important to the grand scheme of things, but yeah, 615 00:28:24,830 --> 00:28:27,560 that's why we have guardrails and test this in constrained environments. 616 00:28:27,710 --> 00:28:28,760 You'd hope, right? 617 00:28:28,760 --> 00:28:31,730 Like that's why before I think people go through it, but it's, 618 00:28:31,730 --> 00:28:34,820 it's actually hard to even test these because software testing, you 619 00:28:34,820 --> 00:28:37,850 had unit tests and you'd essentially say like, okay, I wrote it. 620 00:28:37,970 --> 00:28:40,310 I wrote the sub routine to be able to do X. Like I kind of know 621 00:28:40,310 --> 00:28:43,370 that it's not going to come out and start reciting French poetry. 622 00:28:43,760 --> 00:28:46,160 But this testing a non-deterministic, a 623 00:28:46,160 --> 00:28:48,440 non-deterministic system is meaningfully. 624 00:28:48,500 --> 00:28:52,010 More challenging and I think Rewind gave people a safety blanket. 625 00:28:52,040 --> 00:28:53,450 But as we were talking to people, the other 626 00:28:53,450 --> 00:28:55,640 thing that really kinda lit up with folks was. 627 00:28:56,475 --> 00:28:58,635 A lot of folks told us, I actually don't even know what are 628 00:28:58,635 --> 00:29:00,554 all the different agents that are running in my ecosystem. 629 00:29:00,584 --> 00:29:03,945 I don't have that registry right now, and I don't have the ability 630 00:29:03,945 --> 00:29:06,465 to go ahead and map out what types of tools and actions can it do. 631 00:29:06,614 --> 00:29:11,594 And so we took that as well as the first party challenges that Rubrik was 632 00:29:11,594 --> 00:29:15,675 actually dealing with ourselves as we roll out agents as a company, and we 633 00:29:15,675 --> 00:29:19,455 formulated into like what our product vision and thesis ultimately has become. 634 00:29:19,965 --> 00:29:23,085 Which to your point, Corey, I think you said earlier, uh, a lot of these 635 00:29:23,085 --> 00:29:26,775 companies are, uh, a paperclip company and they're now like, Hey, we're an AI 636 00:29:26,775 --> 00:29:29,835 company and you know, have you changed the core product of the paperclip or not? 637 00:29:30,629 --> 00:29:35,100 We kind of took a view that we needed to launch a a net new product really. 638 00:29:35,100 --> 00:29:38,010 So there's the Rubrik Security Cloud, which is the thing that 639 00:29:38,010 --> 00:29:40,290 Rubrik has really kind of built its core business around. 640 00:29:40,830 --> 00:29:43,410 And we just announced, uh, most recently our new 641 00:29:43,410 --> 00:29:45,389 product, which is called the Rubrik Agent Cloud. 642 00:29:46,020 --> 00:29:48,240 Uh, and the rubric Agent Cloud comes up with three key 643 00:29:48,240 --> 00:29:50,490 pillars that are based on these conversations we had. 644 00:29:50,879 --> 00:29:53,490 The first pillar is monitoring and observability. 645 00:29:54,070 --> 00:29:55,480 I often think about monitoring observability 646 00:29:55,480 --> 00:29:57,580 is like that base layer you need to have. 647 00:29:57,670 --> 00:29:59,290 You need to know what kind of agents are running. 648 00:29:59,290 --> 00:30:02,020 You need to be able to understand a little bit about what is the blast radius. 649 00:30:02,770 --> 00:30:05,860 The second pillar that I haven't seen a lot of people, I see a lot of people 650 00:30:05,860 --> 00:30:08,950 talk about, but I haven't seen a lot of really great software solutions towards. 651 00:30:09,510 --> 00:30:12,630 Our governance and policy enforcement with guardrails. 652 00:30:12,930 --> 00:30:14,460 So we talk about guardrails, but what does that 653 00:30:14,460 --> 00:30:16,980 actually mean to be able to do at a systems level? 654 00:30:17,310 --> 00:30:20,370 That's what we solved, and one of the reasons we decided to solve it 655 00:30:20,370 --> 00:30:23,370 is I saw what it looked like to do AI governance inside of rubric. 656 00:30:23,370 --> 00:30:27,150 First party, I sat on our AI governance committee and things were done with via. 657 00:30:27,445 --> 00:30:30,504 Documents that legal wrote, Google Sheets, you know, sort of the 658 00:30:30,504 --> 00:30:33,385 best of intentions, but the hardest things to enforce in practice. 659 00:30:33,774 --> 00:30:36,054 So we wanted to platformized that, and that's our second pillar. 660 00:30:36,205 --> 00:30:37,945 And then the third pillar is remediation. 661 00:30:38,274 --> 00:30:40,344 Rubrics always had a mentality assumed breach. 662 00:30:40,375 --> 00:30:42,895 I think everyone listening should probably assume something is going to 663 00:30:42,895 --> 00:30:46,945 happen to, if you deploy agents at the level of promise that they achieve 664 00:30:46,945 --> 00:30:48,865 defense in depth, especially when the uh, 665 00:30:48,865 --> 00:30:50,365 attack is coming from inside the house. 666 00:30:50,725 --> 00:30:53,905 Exactly defense and depth across the hyperscaler capabilities, but also 667 00:30:53,905 --> 00:30:57,504 the single control plane that gives you access across the different agents. 668 00:30:57,804 --> 00:30:59,095 So that's what I'm up to. 669 00:30:59,125 --> 00:31:01,885 Uh, now these days, uh, we're, we're building, working on 670 00:31:01,885 --> 00:31:05,455 rubric, agent cloud, and, uh, we started with this idea around 671 00:31:05,455 --> 00:31:07,915 we should be able to rewind these actions and then realized 672 00:31:08,185 --> 00:31:11,605 there's actually really a need for a broader control system here. 673 00:31:11,935 --> 00:31:13,100 Uh, and that's what we've architected. 674 00:31:13,710 --> 00:31:15,990 One last question I have because there are a few 675 00:31:15,990 --> 00:31:20,490 people that can answer this question honestly without. 676 00:31:21,660 --> 00:31:24,090 Sounding I, I guess like they're just imagining 677 00:31:24,090 --> 00:31:25,680 and prognosticating here, but you've done it. 678 00:31:26,129 --> 00:31:30,420 What lessons did you learn by starting a machine learning 679 00:31:30,420 --> 00:31:34,740 company in an era before Gen AI was all the rage. 680 00:31:34,860 --> 00:31:36,629 On some level, I feel like it has to be frustrating 681 00:31:36,629 --> 00:31:38,700 'cause you've done the work and lived in this space, and 682 00:31:38,700 --> 00:31:41,245 now every grifter out there claims to be an AI expert. 683 00:31:42,160 --> 00:31:45,580 I think, um, it, it doesn't bother me so much because 684 00:31:45,580 --> 00:31:48,129 I've always had, like, you know, since I started my 685 00:31:48,160 --> 00:31:50,500 career in machine learning started well over a decade ago. 686 00:31:50,530 --> 00:31:54,960 Uh, and since I started I always felt like AI was a. Candidly, like 687 00:31:54,960 --> 00:31:58,680 overhyped technology over the last two years is the first time I 688 00:31:58,680 --> 00:32:02,430 felt like it's under hyped, which is insane to say because I think 689 00:32:02,430 --> 00:32:06,060 that AI has such a hype cycle that's come in over the last two years. 690 00:32:06,360 --> 00:32:08,400 The reason I say that I think it's actually, you know, potentially 691 00:32:08,400 --> 00:32:11,340 under hyped is because I think that most organizations haven't 692 00:32:11,340 --> 00:32:13,800 gotten to that value discovery point yet because we haven't figured 693 00:32:13,800 --> 00:32:16,680 out, hopefully the easier part than figuring out how to make models 694 00:32:16,680 --> 00:32:19,530 super intelligent, which is the risk components around those models. 695 00:32:19,890 --> 00:32:22,140 But like in terms of advice or how I think about it. 696 00:32:22,290 --> 00:32:25,800 Um, the thing that we've often struggled towards, I think in 697 00:32:25,800 --> 00:32:28,260 machine learning and AI conventionally, is really two things. 698 00:32:28,260 --> 00:32:31,020 The first is tying ML directly with value. 699 00:32:31,590 --> 00:32:34,020 Um, and so I worked with healthcare systems early on 700 00:32:34,020 --> 00:32:36,930 that were looking to be able to, um, build models that 701 00:32:36,930 --> 00:32:40,350 would, could help, um, expedite radiology processes. 702 00:32:40,825 --> 00:32:42,835 Um, and this seems like a very straightforward use case 703 00:32:42,835 --> 00:32:45,145 that would be tied towards value, but it was sort of a 704 00:32:45,145 --> 00:32:47,665 data science and machine learning team doing it in a silo. 705 00:32:47,905 --> 00:32:50,935 There's a big question of like, okay, can the actual physician or 706 00:32:50,935 --> 00:32:54,535 operator, or can the insurance pair whoever needs to be able to use this? 707 00:32:54,835 --> 00:32:56,515 Can they actually use the outputs of that? 708 00:32:56,995 --> 00:32:59,605 So I think the very first thing, it's like a startup lesson 709 00:32:59,605 --> 00:33:02,455 that also extends to everyone building towards AI is. 710 00:33:03,075 --> 00:33:06,495 Get to the dollars or the cents like as quickly as possible. 711 00:33:06,945 --> 00:33:08,115 We'll just sell ads later. 712 00:33:08,115 --> 00:33:08,805 It'll be fine. 713 00:33:09,105 --> 00:33:11,355 Yeah, well, or, or like, or get to value. 714 00:33:11,355 --> 00:33:13,905 If you're a consumer business, get to engagement, right? 715 00:33:13,905 --> 00:33:15,345 Like I would say go for retention user. 716 00:33:15,345 --> 00:33:18,105 If you're a B2B business, like if you're a consumer, the ultimate 717 00:33:18,105 --> 00:33:20,325 source of truth is like our people spending time in your property. 718 00:33:20,325 --> 00:33:23,085 You will figure that piece out later if you're a B2B business. 719 00:33:23,085 --> 00:33:25,725 The ultimate source of truth is, does somebody trust me enough? 720 00:33:25,760 --> 00:33:28,520 To legitimately go to their boss and sign a purchase order. 721 00:33:28,550 --> 00:33:30,110 It doesn't have to be for a lot of money. 722 00:33:30,110 --> 00:33:31,070 Just validate. 723 00:33:31,070 --> 00:33:33,710 Will they transfer a dollar from their bank account to ours? 724 00:33:33,770 --> 00:33:35,660 My hot take is like in a consumer business, 725 00:33:35,660 --> 00:33:38,000 your job is to make users feel delighted. 726 00:33:38,150 --> 00:33:40,790 'cause your source of truth is, do they refer you to a um. 727 00:33:40,855 --> 00:33:41,575 See you friend. 728 00:33:41,905 --> 00:33:43,795 My hot take is for an enterprise business, your 729 00:33:43,795 --> 00:33:46,345 job is to make your champion uncomfortable. 730 00:33:46,705 --> 00:33:49,345 'cause you want them to be able to go like, feel the 731 00:33:49,345 --> 00:33:51,475 discomfort of going to their management of their boss. 732 00:33:51,655 --> 00:33:53,575 No one wants to ask for money, and honestly, people don't 733 00:33:53,575 --> 00:33:55,735 wanna go through procurement and security and vendor audits. 734 00:33:56,035 --> 00:33:57,115 You want them to feel. 735 00:33:57,350 --> 00:33:59,209 Convicted enough where they're like, it's worth 736 00:33:59,209 --> 00:34:01,040 the discomfort of going through those processes. 737 00:34:01,070 --> 00:34:02,629 'cause that's when you know that you're onto something. 738 00:34:03,290 --> 00:34:05,899 Um, and so yeah, the very first piece of advice I think was just, 739 00:34:05,899 --> 00:34:08,960 um, really, and you know, the longer chunk of it is tie it directly 740 00:34:08,960 --> 00:34:11,779 towards value and know what truth looks like in your space. 741 00:34:12,259 --> 00:34:16,339 Um, and then the second is, uh, just recognize that you need 742 00:34:16,339 --> 00:34:20,420 to be able, recognize the importance of small and quick wins. 743 00:34:20,719 --> 00:34:22,819 Conventional machine learning and data science looked like. 744 00:34:22,850 --> 00:34:26,360 Um, there was a conve, there was a joke, which was like ML was. 745 00:34:27,159 --> 00:34:30,190 90% about cleaning data and 10% about 746 00:34:30,190 --> 00:34:32,379 complaining about clinically uh, cleaning data. 747 00:34:32,830 --> 00:34:35,379 Uh, and you know, there was a long process you 748 00:34:35,379 --> 00:34:37,060 would take before we started to realize value. 749 00:34:37,419 --> 00:34:39,790 And I think the other piece is like, recognize time to value. 750 00:34:39,790 --> 00:34:42,790 Gen AI has never made it faster to be able to build a prototype or an agent. 751 00:34:42,819 --> 00:34:44,770 I actually think building agents now is the easy part. 752 00:34:44,799 --> 00:34:44,980 No. 753 00:34:44,980 --> 00:34:47,259 Now the problem is, well, the prototype agents I've built 754 00:34:47,259 --> 00:34:49,810 is, oh, there's a new Claude Sonnet model out there. 755 00:34:50,404 --> 00:34:51,214 Oh crap. 756 00:34:51,214 --> 00:34:54,334 Where do I have to go and update all the model strings and I get to play 757 00:34:54,334 --> 00:34:58,325 whack-a-mole in my, uh, project directory of where did this all live 758 00:34:58,384 --> 00:35:01,205 and hopefully nothing I built to scaffolding or observability breaks around it. 759 00:35:01,265 --> 00:35:04,415 And the right answer, frankly, is, oh, you just instantiate Claude 760 00:35:04,415 --> 00:35:07,055 code and have it do it and just, you know, keep hitting at it. 761 00:35:07,055 --> 00:35:07,535 It'll be fine. 762 00:35:07,834 --> 00:35:10,504 And my hope Corey, is the right answer will become you 763 00:35:10,504 --> 00:35:12,935 substantiate Rubrik Agent Cloud, all of the scaffold, 764 00:35:13,145 --> 00:35:15,455 great models out there, great agent builders out there. 765 00:35:15,455 --> 00:35:17,254 But if you wanna make sure that they all kind of work across 766 00:35:17,254 --> 00:35:20,345 your organization, Rubrik Agent Cloud will like, monitor, 767 00:35:20,404 --> 00:35:22,805 govern, and help you remediate anytime something goes wrong. 768 00:35:23,075 --> 00:35:24,424 That's really what I think I'm excited about. 769 00:35:25,365 --> 00:35:27,765 If people wanna learn more, where's the best place for 'em to find you? 770 00:35:28,065 --> 00:35:28,965 Yeah, it's a great question. 771 00:35:28,965 --> 00:35:31,305 You can actually go towards our, um, website and 772 00:35:31,305 --> 00:35:33,915 there's an agent Subpage that'll tell you about it. 773 00:35:33,915 --> 00:35:38,475 We've now started to post the webinars online and, uh, also demo videos. 774 00:35:38,865 --> 00:35:40,305 Just go ahead and reach out to us. 775 00:35:40,335 --> 00:35:43,875 Right now we're in an early access program, uh, and so we're starting to 776 00:35:43,875 --> 00:35:46,800 selectively onboard customers and we'd love to be able to chat further. 777 00:35:47,985 --> 00:35:50,445 And we'll of course put links to that in the show notes. 778 00:35:50,655 --> 00:35:52,815 Thank you so much for taking the time to speak with me. 779 00:35:52,875 --> 00:35:53,595 I appreciate it. 780 00:35:53,745 --> 00:35:54,645 Co. Really enjoyed it. 781 00:35:54,645 --> 00:35:55,160 Thanks for having me on. 782 00:35:56,265 --> 00:35:59,535 Dev Rishi GM of AI at Rubrik. 783 00:35:59,715 --> 00:36:02,955 I'm cloud economist Corey Quinn, and this is Screaming at the Cloud. 784 00:36:03,194 --> 00:36:05,955 If you've enjoyed this podcast, please leave a five star review on 785 00:36:05,955 --> 00:36:09,285 your podcast platform of choice, whereas if you've hated this episode, 786 00:36:09,285 --> 00:36:12,375 please, we have a five star review on your podcast platform of 787 00:36:12,375 --> 00:36:16,060 choice, along with an angry comment written by an agent that has gone 788 00:36:16,060 --> 00:36:19,335 completely outside the bounds of what guardrails you thought were there.