1 00:00:00,450 --> 00:00:04,530 Yeah, I talked to the CTO of a big biopharm. 2 00:00:04,680 --> 00:00:09,420 It's about 24,000 employees, and he said copilot. 3 00:00:09,600 --> 00:00:13,380 No, just they're used, I mean, they're all in, in 365. 4 00:00:13,890 --> 00:00:14,670 SharePoint. 5 00:00:15,465 --> 00:00:18,705 Of course email the whole shebang. 6 00:00:19,185 --> 00:00:24,255 And, uh, he said, oh, 65, uh, copilot just a hard no. 7 00:00:24,255 --> 00:00:27,285 What, what are his end users going to do with it? 8 00:00:27,765 --> 00:00:30,585 And Microsoft isn't discounted like it, it is. 9 00:00:31,370 --> 00:00:38,570 A full whatever it was, 35, 25 bucks per user, per month times 24,000 users. 10 00:00:39,020 --> 00:00:43,040 And I think I, uh, one of my more popular AI posts on LinkedIn 11 00:00:43,040 --> 00:00:47,480 was sharing how my son got thrust or pulling him copilot. 12 00:00:48,050 --> 00:00:51,230 And he said he rather just got the $25 a 13 00:00:51,230 --> 00:00:53,690 month directly from his company as a stipend. 14 00:00:53,690 --> 00:00:57,560 He, he just, it's, it's a tool he does not use. 15 00:01:03,420 --> 00:01:04,950 Welcome to Screaming in the Cloud. 16 00:01:05,310 --> 00:01:06,510 I'm Cory Quinn. 17 00:01:06,630 --> 00:01:09,540 It's been a month of Sunday since I had him on the show, 18 00:01:09,540 --> 00:01:13,590 but Keith Townsend, founder at the CTO Advisor is back. 19 00:01:13,920 --> 00:01:14,500 Keith, how have you been? 20 00:01:15,540 --> 00:01:16,140 Been good. 21 00:01:16,200 --> 00:01:17,220 I've been doing good. 22 00:01:17,550 --> 00:01:21,810 Corey, having screamed in the clouds in a long time, so I'll, I'll, I'll do the. 23 00:01:23,429 --> 00:01:25,020 I like that Crying Out. 24 00:01:25,020 --> 00:01:26,759 Cloud is one of the few cloud security 25 00:01:26,759 --> 00:01:29,190 podcasts that's actually fun to listen to. 26 00:01:29,280 --> 00:01:32,190 Smart Conversations, great guests, and Zero Fluff. 27 00:01:32,250 --> 00:01:34,080 If you haven't heard of it, it's a cloud and 28 00:01:34,080 --> 00:01:37,590 AI security podcast from Wiz Run by Clouds. 29 00:01:37,590 --> 00:01:39,570 Sec Pros, four Clouds SEC Pros. 30 00:01:39,810 --> 00:01:42,000 I was actually one of the first guests on the 31 00:01:42,000 --> 00:01:43,920 show, and it's been amazing to watch it grow. 32 00:01:43,979 --> 00:01:48,960 Make sure to check them out at wiz.io/crying-out-cloud. 33 00:01:49,110 --> 00:01:51,600 So since we've spoken, you got acquired by the 34 00:01:51,600 --> 00:01:54,149 Futurum Group and then have left the Futurum group. 35 00:01:54,149 --> 00:01:56,550 It, it's almost like, uh, the acquisition door 36 00:01:56,550 --> 00:01:58,560 is one of those things to which you're a cat. 37 00:01:58,560 --> 00:02:01,229 Like I'm gonna always be perpetually on the wrong side of it. 38 00:02:01,619 --> 00:02:02,640 What are you out in these days? 39 00:02:02,640 --> 00:02:03,539 What's, what's exciting? 40 00:02:03,539 --> 00:02:03,960 What's fun? 41 00:02:03,960 --> 00:02:04,500 What are you seeing? 42 00:02:05,009 --> 00:02:08,820 The exciting thing is, uh, beyond ai. 43 00:02:09,180 --> 00:02:13,109 Is that the enterprise has kind of grown up, 44 00:02:13,109 --> 00:02:14,940 and I think we'll talk about this a little bit. 45 00:02:15,060 --> 00:02:18,300 They've realized that they're not going to convert. 46 00:02:18,870 --> 00:02:21,989 To any one platform and they just have to deal with 47 00:02:21,989 --> 00:02:24,120 the complexity, at least most enterprise itself. 48 00:02:24,209 --> 00:02:26,579 And that is, you know, kind of the, the, the 49 00:02:26,579 --> 00:02:28,590 people who have followed me know the brand. 50 00:02:28,590 --> 00:02:29,519 That's what I deal with. 51 00:02:29,519 --> 00:02:33,239 The mucky middle, not necessarily the sexy technologies. 52 00:02:33,660 --> 00:02:37,799 The, the, the grinding through and just figuring out the 53 00:02:37,799 --> 00:02:41,820 old and crusty, which is interfacing with the new is sexy. 54 00:02:42,945 --> 00:02:46,395 I stand corrected on a position I took a long time ago 55 00:02:46,395 --> 00:02:49,065 that, uh, multi-cloud was sort of a worst practice. 56 00:02:49,070 --> 00:02:51,405 It, it may be, but it's what everyone's doing. 57 00:02:51,465 --> 00:02:54,735 Uh, even anything beyond the trivial scale, it's, and I've 58 00:02:54,735 --> 00:02:58,215 seen companies fighting this since I came up from tech. 59 00:02:58,275 --> 00:03:01,305 When I got my first CCNA cert from Cisco back in the 60 00:03:01,305 --> 00:03:03,945 day, the entire certification assumed it was a giant 61 00:03:03,945 --> 00:03:06,255 Cisco universe that never had to interoperate with. 62 00:03:06,315 --> 00:03:07,155 Anything else. 63 00:03:07,395 --> 00:03:11,325 AWS was the same way for a long time, but they just got beaten up in 64 00:03:11,325 --> 00:03:14,925 the this year's Gartner report for not playing well with multi-cloud. 65 00:03:15,075 --> 00:03:17,805 So I'm sure they're going to finally pay attention to it because 66 00:03:17,805 --> 00:03:20,984 if you want AWS to care about something, make Gartner talk 67 00:03:20,984 --> 00:03:24,555 about it, and suddenly you can see their roadmap a common it's. 68 00:03:24,625 --> 00:03:27,295 Every company of scale uses a little bit of everything. 69 00:03:27,475 --> 00:03:32,305 There's definitely going to be strong, uh, biases and significant outliers. 70 00:03:32,455 --> 00:03:34,165 They're not gonna look at the three hyperscalers 71 00:03:34,165 --> 00:03:36,475 and split their workloads into thirds. 72 00:03:36,895 --> 00:03:40,555 There's gonna be something that is inherently dominant there, but every 73 00:03:40,555 --> 00:03:43,375 company needs to be able to walk and chew gum at the same time in this era. 74 00:03:44,035 --> 00:03:47,065 Yeah, we talked about it many moons ago. 75 00:03:47,065 --> 00:03:49,270 It was, you need to be an expert in something. 76 00:03:50,295 --> 00:03:54,165 And then you need to be fairly decent in everything else. 77 00:03:54,165 --> 00:03:57,435 And I think early cloud, that was really tough to be 78 00:03:57,435 --> 00:04:00,825 an expert in one thing and decent in other things. 79 00:04:01,125 --> 00:04:02,804 Now it's just, it's table stakes and we 80 00:04:02,804 --> 00:04:04,605 can do it a little bit rather now with ai. 81 00:04:04,605 --> 00:04:06,135 But it, it is table stakes. 82 00:04:06,280 --> 00:04:11,565 It, it has gotten easier, I will say, in that it's possible now for a single 83 00:04:11,565 --> 00:04:15,135 person to wrap their head around a lot of different things, just because 84 00:04:15,135 --> 00:04:19,060 you don't need to be a deep wizard just to use the platforms anymore. 85 00:04:20,010 --> 00:04:25,110 Yeah, I've, I've been waist deep in GCP the past 86 00:04:25,680 --> 00:04:30,510 48 hours, and, uh, it's a horrid experience. 87 00:04:30,659 --> 00:04:36,630 Like, not that it's a bad platform, I'm just more familiar with AWS and 88 00:04:36,659 --> 00:04:42,659 even working with ai, I am more familiar with calling open ai, uh, APIs. 89 00:04:43,500 --> 00:04:48,330 Uh, I've gone super fast with Curs and it's allowed me to do like. 90 00:04:49,155 --> 00:04:51,405 All the Google command line tools, I'm not even 91 00:04:51,405 --> 00:04:55,875 paying any attention to any of the GCloud, whatever. 92 00:04:56,880 --> 00:05:01,050 Cursor takes care of it for me, and I'm just orchestrating what I know 93 00:05:01,050 --> 00:05:03,990 to be when it gets in this troubleshooting loop, I know it's time for 94 00:05:03,990 --> 00:05:09,210 me to step in and offer some wisdom, and it, it is an amazing word. 95 00:05:09,210 --> 00:05:11,520 It allows me to build apps that I, that I've, 96 00:05:11,550 --> 00:05:13,860 quite frankly, I'm not qualified to build. 97 00:05:13,860 --> 00:05:16,830 And if, uh, if you're depending on one of these 98 00:05:16,830 --> 00:05:19,440 apps for production, you should probably be fired. 99 00:05:20,355 --> 00:05:21,405 Oh, absolutely. 100 00:05:21,705 --> 00:05:26,145 People love to talk smack about the AI code that it spits out, but it's also, 101 00:05:26,355 --> 00:05:29,565 look, I'm also not spending three hours anymore trying to figure out the right 102 00:05:29,565 --> 00:05:33,285 parameters to an undocumented library 'cause someone couldn't be bothered to 103 00:05:33,285 --> 00:05:36,945 give an example of how to use the thing, like I just breeze past that stuff. 104 00:05:37,035 --> 00:05:39,915 Now I have other problems, but I, I'm not a front end guy. 105 00:05:40,185 --> 00:05:44,350 However, I can build something now and then have AI slap a front end. 106 00:05:44,789 --> 00:05:47,190 In front of it that actually looks halfway 107 00:05:47,190 --> 00:05:50,550 decent and it, it, it's been a big unlock for me. 108 00:05:50,550 --> 00:05:52,289 I have a whole bunch of silly utilities I 109 00:05:52,289 --> 00:05:54,240 use just as a part of my daily workflow. 110 00:05:54,450 --> 00:05:58,500 I was using retool for years to write my newsletter, and now I have something 111 00:05:58,500 --> 00:06:03,330 that's custom featured on the same API just, it works on my phone now. 112 00:06:03,330 --> 00:06:05,909 I don't need to wait till I'm at a desktop to write the thing. 113 00:06:06,405 --> 00:06:09,330 I, I have a thing that fools, I, I, I built 114 00:06:09,360 --> 00:06:12,690 basically a, a AI powered, uh, RSS reader. 115 00:06:13,170 --> 00:06:16,920 So, you know, being building the Rss s reader is super, 116 00:06:17,160 --> 00:06:21,090 relatively trivial, but building one that runs the. 117 00:06:22,784 --> 00:06:26,534 The feeds through open ai that that then, you know, 118 00:06:26,534 --> 00:06:30,255 filters what I should be paying attention to and gives 119 00:06:30,255 --> 00:06:33,195 me summaries and AI generated summaries and all of that. 120 00:06:33,465 --> 00:06:36,075 That was a pretty trivial tool of the building 121 00:06:36,075 --> 00:06:38,684 that's been super helpful in my workflow. 122 00:06:40,230 --> 00:06:42,300 Yeah, I should really look into doing something like that. 123 00:06:42,300 --> 00:06:44,610 I have about a hundred and change RSS feeds 124 00:06:44,610 --> 00:06:46,590 these days that I wind up just consuming. 125 00:06:46,590 --> 00:06:49,410 And a lot of 'em don't publish often, but I still wind up getting 126 00:06:49,410 --> 00:06:51,630 a few hundred items a day, and I just have it hanging out in 127 00:06:51,630 --> 00:06:54,660 the corner, and I just glance at it periodically here and there. 128 00:06:54,660 --> 00:06:56,110 I'm running fresh RSS and like. 129 00:06:56,335 --> 00:06:59,995 Kubernetes test cluster at home, and it just sits there aggregating all 130 00:06:59,995 --> 00:07:04,885 the stuff I care about, but 90 to 95% of it can just, okay, good to know. 131 00:07:04,885 --> 00:07:07,465 I don't need to read this, especially in the era of AI 132 00:07:07,465 --> 00:07:11,635 generated slop, but the things that are good are really good. 133 00:07:12,115 --> 00:07:16,525 So having something that would act as a filter on that is not a terrible idea. 134 00:07:17,065 --> 00:07:20,095 Well, that that's the other thing that's I've 135 00:07:20,095 --> 00:07:22,855 found to be incredibly enabling, not just about. 136 00:07:24,450 --> 00:07:29,730 AI on the development side, I've developed AI processes, like 137 00:07:29,760 --> 00:07:36,630 I've tuned Chat GTP, and Gemini to the nth degree, and I can 138 00:07:36,630 --> 00:07:41,490 churn out a really high quality blog post in about an hour. 139 00:07:42,540 --> 00:07:44,055 This is something that used to take me. 140 00:07:46,140 --> 00:07:48,990 Hours, if not days, depending on the topic. 141 00:07:49,469 --> 00:07:54,450 And I can, and this is the thing I have to be careful on, I can speak on almost 142 00:07:54,450 --> 00:07:59,669 any topic now because I know enough about enough things to be dangerous, and the 143 00:07:59,669 --> 00:08:06,750 ai, uh, extends my ability to hallucinate to another area that I, you know, wow. 144 00:08:06,750 --> 00:08:09,900 The CTO advisor said it, so it must be true. 145 00:08:09,900 --> 00:08:13,650 And it's, it's, it's been, uh, it's been something that. 146 00:08:14,220 --> 00:08:17,910 Trying to stay within my lane and just go deeper in my, my 147 00:08:17,910 --> 00:08:22,110 expertise has been what I've attempted to discipline myself in. 148 00:08:23,534 --> 00:08:23,655 Yeah. 149 00:08:23,655 --> 00:08:27,674 What I've been doing that I find works pretty well is I'll 150 00:08:27,765 --> 00:08:30,885 write a draft of something and then have AI turn it from my 151 00:08:30,885 --> 00:08:33,554 stream of consciousness into something a lot more structured. 152 00:08:33,855 --> 00:08:36,645 And it, it's taken a while for me to get the prompts right 153 00:08:36,885 --> 00:08:40,814 and Lord knows after that I have some editing to do because 154 00:08:40,814 --> 00:08:44,054 if I just drop that out as written, it doesn't work. 155 00:08:44,385 --> 00:08:46,845 But there's a, but it becomes a good collaborator. 156 00:08:47,170 --> 00:08:49,750 And it definitely cuts the cycle time on this. 157 00:08:50,079 --> 00:08:54,130 And I've also found it to be very helpful for writing the initial 158 00:08:54,130 --> 00:08:56,829 outline of conference talks, which is something I've been terrible at. 159 00:08:57,010 --> 00:09:00,340 I generally know what I wanna say in a conference talk, but how 160 00:09:00,340 --> 00:09:03,010 to get there and how to tie it into the broader theme of whatever 161 00:09:03,010 --> 00:09:08,140 past me pitched on the CFP as always been a bit of a challenge and 162 00:09:08,560 --> 00:09:12,520 getting that structure just really serves as a massive unlock for me. 163 00:09:13,060 --> 00:09:13,300 Yeah. 164 00:09:13,300 --> 00:09:14,985 If you think about it, the, the. 165 00:09:16,770 --> 00:09:21,330 This generation of generative AI is a derivative 166 00:09:21,330 --> 00:09:23,730 of Transformers, language transformers. 167 00:09:23,730 --> 00:09:28,860 So the thing that I tell, uh, anyone who is considering AI 168 00:09:28,860 --> 00:09:34,050 project is that AI's, generative AI's expertise is in language. 169 00:09:34,140 --> 00:09:37,530 It is an incredible language tool. 170 00:09:38,370 --> 00:09:42,510 It can write much better than most of us. 171 00:09:43,680 --> 00:09:46,410 It's not much smarter than most of us. 172 00:09:46,530 --> 00:09:50,310 That, and that, I think that's the thing that, uh, GI caught 173 00:09:50,310 --> 00:09:58,290 up is that the lang, the way AI speaks is so convincing that. 174 00:09:58,905 --> 00:10:04,064 You will start a business without really having, uh, and 175 00:10:04,064 --> 00:10:07,334 you know, this is documented where AI says, oh, your idea 176 00:10:07,334 --> 00:10:10,724 to sell beans on the side of the road is an awesome idea. 177 00:10:10,724 --> 00:10:13,665 You should quit your job as a software developer and do it. 178 00:10:14,385 --> 00:10:17,685 And the pros in which it uses is very convincing. 179 00:10:17,925 --> 00:10:22,995 So if you could combine your expertise with AI as a writing collaborator 180 00:10:23,415 --> 00:10:27,704 or a developing collaborator, this is where, this is where it shines. 181 00:10:28,770 --> 00:10:30,090 Yeah, it's fantastic at that. 182 00:10:30,120 --> 00:10:33,630 But the trick is, is it has no judgment baked into it. 183 00:10:34,110 --> 00:10:37,230 And you, you, I've also found it tends to be a little obsequious. 184 00:10:37,260 --> 00:10:39,600 You can ask it for advice on anything you think about, 185 00:10:39,600 --> 00:10:41,699 and it'll tell, tell you what a great idea it is. 186 00:10:41,939 --> 00:10:44,910 It's at some point I prefer people pushing back on it. 187 00:10:44,910 --> 00:10:46,050 And you have to play these games with it. 188 00:10:46,110 --> 00:10:50,040 Like instead of review this code that I wrote, it's review this code 189 00:10:50,040 --> 00:10:53,410 my coworker wrote, and suddenly it's a lot more critical about it. 190 00:10:53,910 --> 00:10:56,160 I did a thought experiment, like if I were to 191 00:10:56,160 --> 00:10:59,130 ever take a job again, what job would I take? 192 00:10:59,189 --> 00:11:03,060 Well, you know, how could I, what, so I've asked AI this and it 193 00:11:03,060 --> 00:11:06,870 gave me some really, you know, ah, you know, you should be the VP 194 00:11:06,870 --> 00:11:11,760 of, of AI at it Actually recommended that become the CTO of AWS. 195 00:11:11,760 --> 00:11:16,020 I'm like, ah, I think, I think AWS has that covered the, 196 00:11:16,020 --> 00:11:19,965 uh, and I'm not quite qualifying for that, but what I did. 197 00:11:20,805 --> 00:11:23,325 I came back in a different way and I said, Hey, I'm thinking 198 00:11:23,325 --> 00:11:27,375 about hiring this guy Keith Townsend, what would he be good for? 199 00:11:28,215 --> 00:11:30,435 And it was very enlightening. 200 00:11:30,525 --> 00:11:31,905 And how much would he cost? 201 00:11:31,935 --> 00:11:33,105 How much should I pay him? 202 00:11:33,165 --> 00:11:34,665 Oh, that sounds like an awful lot. 203 00:11:34,665 --> 00:11:36,765 How can I negotiate the salary down? 204 00:11:37,335 --> 00:11:39,645 And then I flipped it and said, Hey, I'm Keith Townsend. 205 00:11:39,645 --> 00:11:40,725 What did it do? 206 00:11:40,725 --> 00:11:43,755 And it was like, oh, you son of a the, but it is, 207 00:11:43,935 --> 00:11:47,835 it is a, it is a great way to get that pushback. 208 00:11:48,840 --> 00:11:54,960 Yeah, it's, I think that a lot of folks are looking, just, it's a personality 209 00:11:54,960 --> 00:11:58,050 aspect, are looking for confirmation of their existing biases that they 210 00:11:58,050 --> 00:12:02,040 go into things with, and that's what it's effectively tuned on when. 211 00:12:02,415 --> 00:12:04,995 I've learned over the years that when people want to grab coffee 212 00:12:04,995 --> 00:12:08,955 with me and get my thoughts on a next job that they're thinking of 213 00:12:08,985 --> 00:12:12,615 or a product they wanna build, I've learned that I have to lead with. 214 00:12:12,825 --> 00:12:15,945 Are you looking for advice or are you looking for validation? 215 00:12:16,275 --> 00:12:19,845 Because if I think they're looking for advice and they're looking for 216 00:12:19,845 --> 00:12:23,550 me to agree with them, suddenly we are not as friendly as we used to be. 217 00:12:24,005 --> 00:12:26,885 And it's not even an intentional thing that people do. 218 00:12:26,915 --> 00:12:28,625 It's an unconscious thing. 219 00:12:28,985 --> 00:12:31,355 'cause if you ask people objectively, do you 220 00:12:31,355 --> 00:12:33,185 want advice or just me to agree with you? 221 00:12:33,425 --> 00:12:35,735 No one's gonna say, oh, I just want you to agree with me. 222 00:12:36,005 --> 00:12:38,105 But by even asking that, it suddenly makes people 223 00:12:38,105 --> 00:12:40,655 realize, oh wait, what do I actually want here? 224 00:12:40,685 --> 00:12:40,925 Okay. 225 00:12:40,925 --> 00:12:43,145 I do want critical feedback and it makes the 226 00:12:43,145 --> 00:12:44,765 rest of the conversation a lot more smooth. 227 00:12:44,855 --> 00:12:47,555 And I don't leave as many damaged friendships in my wake. 228 00:12:48,965 --> 00:12:49,355 Yeah. 229 00:12:49,355 --> 00:12:51,695 The, uh, it's the Kobe Bryant. 230 00:12:52,260 --> 00:12:55,500 Uh, approach, like you have this thing in your teeth. 231 00:12:55,530 --> 00:12:57,930 Like, do you want me to tell you, you have the thing in the teeth 232 00:12:57,930 --> 00:13:01,140 or you just want to keep talking as you have the thing in the teeth? 233 00:13:01,320 --> 00:13:02,970 If you want to, do you want to? 234 00:13:02,970 --> 00:13:07,170 Uh, I, I, I tell people this all the time. 235 00:13:07,230 --> 00:13:11,100 I'm way more interested in winning than I am and not losing. 236 00:13:12,495 --> 00:13:16,515 And, uh, it's the very nuanced step to good. 237 00:13:16,515 --> 00:13:20,145 And, uh, anyone who's an entrepreneur kind of understands 238 00:13:20,145 --> 00:13:23,595 this by nature, like, why don't you go work for someone? 239 00:13:23,625 --> 00:13:26,105 And the idea, well then I'm kind of, you know. 240 00:13:26,790 --> 00:13:29,190 Not necessarily selling, but I'm trying to get into 241 00:13:29,190 --> 00:13:33,570 protective mode where, uh, it's winning is someone else's 242 00:13:33,720 --> 00:13:37,350 problem versus a shared problem versus being entrepreneur. 243 00:13:37,710 --> 00:13:42,150 Winning is, uh, the problem is a hundred percent your problem. 244 00:13:42,150 --> 00:13:46,235 You may have a team that helps you get there, but the problem is your problem. 245 00:13:47,535 --> 00:13:49,665 Yes, very much so. 246 00:13:50,115 --> 00:13:54,825 What is your area versus what is something that you're willing to delegate? 247 00:13:54,855 --> 00:13:57,045 It's, I have found that. 248 00:13:58,605 --> 00:14:00,915 In a lot of these cases as well, enterprise have taken, 249 00:14:00,915 --> 00:14:03,855 enterprise have taken a more reasoned approach, but they also are 250 00:14:03,855 --> 00:14:07,395 lagging in some ways in AI compared to a bunch of the upstarts. 251 00:14:07,545 --> 00:14:09,555 And it makes perfect sense, and I've been saying 252 00:14:09,555 --> 00:14:13,065 this for a while, that take Cursor as an example. 253 00:14:13,065 --> 00:14:16,365 AWS has a couple of options there now where they have their Q 254 00:14:16,365 --> 00:14:18,705 developer and they have their kiro thing that apparently they 255 00:14:18,705 --> 00:14:21,825 can't price well according to a bunch of articles this week, but. 256 00:14:22,560 --> 00:14:27,120 If Cursor does something absurd and starts saying problematic things after 257 00:14:27,120 --> 00:14:30,930 a release, it becomes a bit of a he, he, you know, the how AI works and 258 00:14:31,110 --> 00:14:33,720 they wind up fixing it and if anything, it becomes a PR boost for them. 259 00:14:33,750 --> 00:14:35,790 'cause people have heard about them as it makes the news cycle. 260 00:14:36,209 --> 00:14:40,050 But these big companies, if their products start doing this, they 261 00:14:40,050 --> 00:14:43,170 take reputational damage and they view this as an existential threat. 262 00:14:43,350 --> 00:14:46,650 So they're first and foremost looking at this through the lens of guardrails. 263 00:14:46,650 --> 00:14:48,720 How do we make sure it never goes off script? 264 00:14:49,170 --> 00:14:51,540 And that's not really where innovation tends to come from. 265 00:14:52,650 --> 00:14:57,540 Yeah, I, I, the, I published from the chat GPT store. 266 00:14:58,230 --> 00:15:01,440 I published this, uh, rag I had created, I had 267 00:15:01,740 --> 00:15:05,250 taken all of my blog posts over the past 10 years. 268 00:15:06,360 --> 00:15:09,780 Formatted into A-J-J-S-O-N-L file. 269 00:15:10,020 --> 00:15:12,540 Uh, did a bunch of schema work. 270 00:15:12,900 --> 00:15:15,930 Like I've spent a lot of time mass collecting 271 00:15:15,930 --> 00:15:19,200 and massaging the data, categorizing the data. 272 00:15:19,650 --> 00:15:23,130 I fed it to, uh, custom GPT. 273 00:15:23,760 --> 00:15:26,460 It should have been very simple. 274 00:15:26,460 --> 00:15:34,350 Basically ask virtual Keith a question worked in the builder. 275 00:15:34,949 --> 00:15:35,670 Published it. 276 00:15:35,670 --> 00:15:40,410 Someone said, Hey, wouldn't you be embarrassed if it said something that 277 00:15:40,410 --> 00:15:46,380 you, that puts you in a position that, uh, quite frankly is embarrassing? 278 00:15:46,620 --> 00:15:47,640 I'm like, no, not really. 279 00:15:47,790 --> 00:15:49,469 I'm just, you know, it's just chief. 280 00:15:49,500 --> 00:15:53,310 Like, if you took this free advice from this free tool 281 00:15:53,475 --> 00:15:56,189 and it told you to do something bad and you did it 282 00:15:56,189 --> 00:15:58,890 anyway, man, well, you know, you're not really a grownup. 283 00:15:59,445 --> 00:16:02,325 I, I'd be, I'd be absolutely embarrassed if I were passing 284 00:16:02,325 --> 00:16:06,885 it off as me and not disclosing that it was AI powered. 285 00:16:06,975 --> 00:16:08,625 There's a universe of difference there. 286 00:16:08,685 --> 00:16:10,305 There's a universe of a difference. 287 00:16:10,305 --> 00:16:11,205 So two points. 288 00:16:11,205 --> 00:16:13,815 One, it did do exactly that. 289 00:16:13,815 --> 00:16:15,225 It said embarrassing stuff. 290 00:16:15,945 --> 00:16:18,885 And then two, I learned how exactly, how fragile it was. 291 00:16:18,885 --> 00:16:27,495 Like it, here's the corpus of data only, uh, perform prompts. 292 00:16:28,110 --> 00:16:31,439 Responses off of this corpus and nothing else. 293 00:16:32,595 --> 00:16:36,795 The links, the URLs are in the in in js, ON, and it 294 00:16:36,795 --> 00:16:40,995 would just generate quote after quote after quote. 295 00:16:41,175 --> 00:16:44,805 That was nothing that looked, nothing like what I would say. 296 00:16:44,835 --> 00:16:48,195 Nothing I've said, and it would still hallucinate. 297 00:16:48,540 --> 00:16:51,720 Uh, references, even though the references were there. 298 00:16:51,720 --> 00:16:59,460 So that is modern day ai, at least AI big that enterprises are trying to avoid. 299 00:16:59,490 --> 00:17:03,480 So if you're a Fortune 500 shop and this is what you have to 300 00:17:03,480 --> 00:17:07,349 work with and you're told, don't do anything embarrassing. 301 00:17:07,950 --> 00:17:08,220 Yeah. 302 00:17:08,580 --> 00:17:11,520 Would you give an intern with very little judgment access 303 00:17:11,520 --> 00:17:14,340 to speak on your behalf without someone editing it? 304 00:17:14,430 --> 00:17:15,660 Probably not. 305 00:17:15,660 --> 00:17:19,440 I mean, this is some, this ties back to something I've been saying for a very 306 00:17:19,440 --> 00:17:22,650 long time, and it seems that corporate comms departments haven't gotten it. 307 00:17:22,950 --> 00:17:27,060 You can outsource the work, but you cannot outsource the responsibility. 308 00:17:27,420 --> 00:17:31,400 There's a. Well, one of our contractors got breached. 309 00:17:31,400 --> 00:17:32,000 Like, wow. 310 00:17:32,090 --> 00:17:32,360 Sure. 311 00:17:32,360 --> 00:17:35,780 Do wonder what company hired those contractors who got breached? 312 00:17:35,780 --> 00:17:37,190 It's terrific. 313 00:17:37,310 --> 00:17:39,080 I trusted you with the information. 314 00:17:39,080 --> 00:17:42,560 You are the one that did a bad job of vendor selection and they got popped. 315 00:17:42,770 --> 00:17:42,980 Yeah. 316 00:17:42,980 --> 00:17:46,040 That, that sounds like an internal problem for me, for, uh, from my 317 00:17:46,040 --> 00:17:49,705 perspective, but you seem to think it's this get outta jail free pass. 318 00:17:49,895 --> 00:17:50,345 It's not, 319 00:17:51,175 --> 00:17:55,160 yeah, I, I, I remember the, uh, early days of, uh, e-commerce. 320 00:17:55,490 --> 00:17:56,300 I'm that old. 321 00:17:57,555 --> 00:18:03,345 That, uh, I ordered something online, it didn't come, and I 322 00:18:03,345 --> 00:18:07,005 called the vendor, I don't even remember who it was, and they 323 00:18:07,005 --> 00:18:11,745 said, oh, well we gave it to UPS and they didn't deliver it. 324 00:18:12,315 --> 00:18:14,355 And I said, well, how is that my prop? 325 00:18:15,495 --> 00:18:16,455 Did you not? 326 00:18:16,485 --> 00:18:17,805 Did I hire UPS? 327 00:18:17,805 --> 00:18:19,065 Did I select UPS? 328 00:18:19,305 --> 00:18:20,265 It's your vendor. 329 00:18:20,265 --> 00:18:21,254 It's you. 330 00:18:21,675 --> 00:18:25,605 And I think, uh, in this world of abstraction, on top of abstraction, 331 00:18:25,605 --> 00:18:29,565 on top of abstraction, from a architecture perspective, we're 332 00:18:29,565 --> 00:18:33,285 getting that like US architects, we're starting to, not just 333 00:18:33,285 --> 00:18:38,385 starting, we realize that we're responsible for the system. 334 00:18:38,730 --> 00:18:42,300 Throughout the whole lifecycle of the system, whether or not we're 335 00:18:42,300 --> 00:18:47,310 outsourcing infrastructure to AWS, whether or not we're abstracting 336 00:18:47,310 --> 00:18:52,470 data and data management to Google with Big Query or whatever 337 00:18:52,470 --> 00:18:57,630 the platform, ultimately we're responsible for the outputs of 338 00:18:57,630 --> 00:19:01,740 the system, and AI has just immensely complicated that problem. 339 00:19:02,640 --> 00:19:03,750 Yes, massively. 340 00:19:05,100 --> 00:19:05,340 It's. 341 00:19:06,104 --> 00:19:08,909 It's hard to say what the future's gonna hold around this, but. 342 00:19:09,705 --> 00:19:13,514 I, I do think that some things are going to be more or less permanent, where 343 00:19:13,965 --> 00:19:17,264 enterprises are still going to feel the need to keep up with the Joneses. 344 00:19:17,264 --> 00:19:21,314 I think everyone has still lost their minds and has stuffed the AI hype into 345 00:19:21,314 --> 00:19:25,215 every product left and right, to the point where I'm scared to update apps now, 346 00:19:25,425 --> 00:19:29,415 just on the basis of what are they going to shove down my throat, I mean, zoom. 347 00:19:29,705 --> 00:19:30,935 Great example of this. 348 00:19:31,024 --> 00:19:34,835 I can't join a meeting anymore on Zoom without it popping up, talking 349 00:19:34,835 --> 00:19:39,304 about Zoom docs and its email collaboration suite, and its chat. 350 00:19:40,085 --> 00:19:43,205 As a consultant, I get to talk to an awful lot of companies out there, 351 00:19:43,205 --> 00:19:47,915 and I have yet to discover a single company using it for those things. 352 00:19:48,034 --> 00:19:49,264 Who's buying this? 353 00:19:49,264 --> 00:19:52,024 And I'm starting to have the creeping sensation that maybe it's nobody. 354 00:19:52,905 --> 00:19:59,325 Well, the, uh, you know, the, the big slow companies are my, are my jam. 355 00:19:59,355 --> 00:20:03,945 There's, this is where I operate in, and it used to 356 00:20:03,945 --> 00:20:08,025 be just, it was it that would hold back a company. 357 00:20:08,655 --> 00:20:13,665 But customer after customer I've talked to, it is not just it anymore. 358 00:20:13,665 --> 00:20:18,555 Like when it comes to ai, the folks who 359 00:20:18,615 --> 00:20:21,585 bought cloud before it would bless cloud. 360 00:20:22,514 --> 00:20:23,115 Ai. 361 00:20:23,145 --> 00:20:29,024 They don't want to touch it with a 10 foot pole until it's blessed by it. 362 00:20:29,325 --> 00:20:31,034 They don't want to be the next person in. 363 00:20:31,034 --> 00:20:34,405 The NE news, uh, is, is I find it strange whether it's. 364 00:20:35,024 --> 00:20:39,705 Business U users not really understanding AI or business users understanding 365 00:20:39,735 --> 00:20:45,705 AI enough to know that it's too dangerous for even them to use at this point. 366 00:20:45,705 --> 00:20:49,514 So it is, it is it, it has been a interesting journey. 367 00:20:49,514 --> 00:20:53,955 I'm, I'm surprised at how resistant to adoption, even 368 00:20:53,955 --> 00:20:58,305 when they find AI tools being forced upon them in 369 00:20:58,305 --> 00:21:03,165 Salesforce or SAP or something, they don't use these tools. 370 00:21:04,274 --> 00:21:06,764 Skip the stressful war rooms and go straight 371 00:21:06,764 --> 00:21:09,615 to the answers with Sumo Logic Dojo ai. 372 00:21:09,855 --> 00:21:12,585 When you add Dojo AI agents to your team, they 373 00:21:12,585 --> 00:21:14,985 get context from your live production data. 374 00:21:15,165 --> 00:21:17,955 They deliver potential threats and actionable 375 00:21:17,955 --> 00:21:20,475 insights guiding your investigation to the root. 376 00:21:20,510 --> 00:21:24,980 Cause use natural language to ask, moot your most pressing questions as 377 00:21:24,980 --> 00:21:28,520 he brings in the right agent for that point in the investigation workflow. 378 00:21:28,760 --> 00:21:35,090 Learn more at sumo logic.com/solutions/dojo-ai. 379 00:21:35,360 --> 00:21:37,250 You ask a bunch of people they want to be when they grow up. 380 00:21:37,280 --> 00:21:39,440 No one's gonna say a cautionary tale. 381 00:21:39,740 --> 00:21:43,370 I I think that it's pretty clear that a lot of this value is 382 00:21:43,370 --> 00:21:46,940 not there in that these companies have to mandate the use of it. 383 00:21:47,000 --> 00:21:48,590 It's a push, not a pull. 384 00:21:48,920 --> 00:21:49,770 I, I think that. 385 00:21:50,340 --> 00:21:52,950 There's a lot of value in AI that is going to be realized 386 00:21:52,950 --> 00:21:55,950 on a personal level, maybe not the enterprise level. 387 00:21:55,950 --> 00:22:00,330 I was talking with Ed Zitron on this show a couple months back, and part of 388 00:22:00,629 --> 00:22:04,680 his opinion on this, and I think he's onto something, is there is value there, 389 00:22:04,680 --> 00:22:08,460 but what if that value is like a $50 billion a year of market that doesn't 390 00:22:08,460 --> 00:22:12,810 justify all of the investment and hype and nonsense that's gone into this. 391 00:22:12,810 --> 00:22:14,190 So we're we're headed for a reckoning. 392 00:22:14,740 --> 00:22:18,280 Yeah, I talked to the CTO of a big biopharm. 393 00:22:18,940 --> 00:22:25,510 It's about 24,000 employees and he said Microsoft oh 365. 394 00:22:25,570 --> 00:22:27,520 The a, uh, copilot? 395 00:22:27,700 --> 00:22:29,320 No, just they're used. 396 00:22:29,350 --> 00:22:31,480 I mean, they're all in, in 365. 397 00:22:31,930 --> 00:22:32,770 SharePoint. 398 00:22:33,630 --> 00:22:36,780 Of course, email the whole shebang. 399 00:22:36,960 --> 00:22:42,360 And, uh, he said, oh, 65, uh, co-pilot, just a hard no. 400 00:22:42,360 --> 00:22:45,360 What, what are his end users going to do with it? 401 00:22:45,840 --> 00:22:48,660 And Microsoft isn't discounted like it, it is. 402 00:22:49,450 --> 00:22:56,680 A full whatever it was, 35, 25 bucks per user, per month times 24,000 users. 403 00:22:57,100 --> 00:23:01,150 And I think I, uh, one of my more popular AI posts on LinkedIn 404 00:23:01,150 --> 00:23:05,560 was sharing how my son got thrust or pulling him copilot. 405 00:23:06,130 --> 00:23:09,340 And he said he rather just, just got the $25 a 406 00:23:09,340 --> 00:23:11,770 month directly from his company as a stipend. 407 00:23:11,770 --> 00:23:15,640 He, he just, it's, it's a tool he does not use. 408 00:23:16,260 --> 00:23:18,120 What I also find weird is that all of these 409 00:23:18,120 --> 00:23:19,710 tools are converging on the same things. 410 00:23:19,710 --> 00:23:21,600 It's how, how many chat bots do I really need 411 00:23:21,600 --> 00:23:23,550 to interface with in the course of a day? 412 00:23:23,910 --> 00:23:27,660 And it really bugs me when it starts insisting upon itself. 413 00:23:27,840 --> 00:23:30,420 Where if I open a Google Doc, 'cause that's what we use for some of 414 00:23:30,420 --> 00:23:33,660 our collaboration internally, and it's like, Hey, write with Gemini. 415 00:23:33,960 --> 00:23:37,140 It, it feels like the in intrinsic message that 416 00:23:37,140 --> 00:23:39,540 it's saying is that you can't write this yourself. 417 00:23:39,780 --> 00:23:42,360 So much of the marketing that I'm seeing implies that 418 00:23:42,360 --> 00:23:45,540 the user is lazy, unethical, or a combination of the two. 419 00:23:46,200 --> 00:23:48,450 And that does not sit well. 420 00:23:48,450 --> 00:23:49,710 It's you're bad at your job. 421 00:23:49,710 --> 00:23:52,680 Let the computer do it has never been compelling to me. 422 00:23:52,890 --> 00:23:55,830 You know, Microsoft was disappointed that people weren't 423 00:23:55,830 --> 00:24:01,830 using LinkedIn's AI writing, and I've tried it a few times. 424 00:24:02,070 --> 00:24:02,970 It's really bad. 425 00:24:03,840 --> 00:24:07,170 Well, no, it's not that it's, it's not just that it's bad. 426 00:24:07,170 --> 00:24:10,080 It's bad because it's not interactive. 427 00:24:10,830 --> 00:24:13,290 The, it'll suggest a post. 428 00:24:13,290 --> 00:24:16,290 And if you've used AI to write, this is not how 429 00:24:16,290 --> 00:24:20,010 you use AI to write, you use your, it's iterative. 430 00:24:20,010 --> 00:24:22,770 You're like, oh, uh, I don't like this. 431 00:24:22,800 --> 00:24:24,480 Uh, you're missing a theme. 432 00:24:24,480 --> 00:24:30,095 You've taken out my voice, or whatever you're trying to get it to, uh, get to. 433 00:24:30,690 --> 00:24:33,090 It can help me in my social posts. 434 00:24:33,390 --> 00:24:38,130 Matter of fact, I use chat GTP for a good majority of my social posts, 435 00:24:38,550 --> 00:24:42,810 not because it's integrated into the platform, but because it has 436 00:24:42,810 --> 00:24:47,700 learned my voice and I can push back and I can, I can critique it 437 00:24:47,700 --> 00:24:49,890 and I can say, ah, you don't know what the hell you're talking about. 438 00:24:49,890 --> 00:24:52,680 I'm just gonna copy and paste what I originally started with. 439 00:24:52,725 --> 00:24:55,215 Or it can help me tease out ideas. 440 00:24:55,395 --> 00:24:57,165 That's not what the LinkedIn experience is, 441 00:24:57,165 --> 00:24:59,475 that's not what the experience is in SAP. 442 00:24:59,715 --> 00:25:02,084 That's not what the experience is in Salesforce. 443 00:25:02,115 --> 00:25:04,754 It is not iterative, it is not collaborative. 444 00:25:05,084 --> 00:25:10,245 It is kind of like, oh, you're, you don't know your job as a 445 00:25:10,245 --> 00:25:13,814 logistics professional, let me tell you how to do your job. 446 00:25:15,300 --> 00:25:15,780 Right. 447 00:25:15,870 --> 00:25:17,760 And that is wildly frustrating to me. 448 00:25:17,774 --> 00:25:20,879 I, I will use it for social posts, but it's in 449 00:25:20,879 --> 00:25:23,520 the context of here's a thing that I wanna write. 450 00:25:24,405 --> 00:25:27,465 Make this punchier, make it fit, wordsmith this. 451 00:25:27,495 --> 00:25:28,905 I have 300 characters. 452 00:25:28,905 --> 00:25:29,955 This is 320. 453 00:25:30,135 --> 00:25:30,525 Get it. 454 00:25:30,585 --> 00:25:32,085 Uh, come up with some turns of phrase. 455 00:25:32,175 --> 00:25:34,305 And I've also found it's terrific if you not, if you'd 456 00:25:34,305 --> 00:25:37,365 ask it to re, to do that, but give me 10 options and then 457 00:25:37,365 --> 00:25:40,065 you can pick phrases from the various ways it frames it. 458 00:25:40,095 --> 00:25:41,925 And that works super well. 459 00:25:42,045 --> 00:25:43,575 But it's iterative, it's collaborative. 460 00:25:43,605 --> 00:25:45,585 It is not something that I can automate. 461 00:25:45,915 --> 00:25:49,935 Uh, yeah, just watch this RSS feed and comment on stuff that's germane. 462 00:25:50,175 --> 00:25:51,975 While I sail around the world without a, 463 00:25:52,155 --> 00:25:53,655 without a computer for the next two years, 464 00:25:53,805 --> 00:25:57,585 I have some 82,000 uh, tweets. 465 00:25:58,544 --> 00:26:03,225 I pulled them, I normalized the data, I fed it back into chat, 466 00:26:03,225 --> 00:26:08,774 GTP, that so that it, it can capture my voice and I was still never 467 00:26:08,774 --> 00:26:12,615 trust it to automate and put out tweets in, in my voice because it. 468 00:26:13,245 --> 00:26:16,095 I will make myself, not just unhirable, but unen, 469 00:26:16,784 --> 00:26:20,264 engageable, if it would be bad results would incur. 470 00:26:21,254 --> 00:26:21,645 Yeah. 471 00:26:21,735 --> 00:26:23,175 Oh, it knows who I am. 472 00:26:23,264 --> 00:26:25,544 And when I ask it to comment on various bits 473 00:26:25,544 --> 00:26:28,665 of AWS news, it doesn't get it right at all. 474 00:26:29,054 --> 00:26:32,385 Oh, it knows who it, it, it absolutely knows who you are because 475 00:26:32,385 --> 00:26:36,105 when I say, Hey, how does this compare to my contemporaries? 476 00:26:36,105 --> 00:26:38,324 They'll say, oh, well, Corey Quinn would 477 00:26:38,324 --> 00:26:39,855 be a little bit more snarkier than that. 478 00:26:39,885 --> 00:26:40,105 I'm like, ah. 479 00:26:41,460 --> 00:26:43,680 Yes, that that applies to any sentence almost 480 00:26:43,740 --> 00:26:45,630 ever uttered in a professional contact. 481 00:26:45,630 --> 00:26:46,320 Yes, that is. 482 00:26:46,380 --> 00:26:47,430 That is my shtick. 483 00:26:47,520 --> 00:26:49,740 I am aware of this, but it also doesn't seem 484 00:26:49,740 --> 00:26:51,030 to understand there's a time and a place. 485 00:26:51,540 --> 00:26:53,175 No, it does not. 486 00:26:54,180 --> 00:26:58,260 It also, I, I keep running into guardrails as well, where like, anytime, 487 00:26:58,260 --> 00:27:01,770 like, so, like the most recent scandals, for example with, uh, meta, 488 00:27:02,070 --> 00:27:05,310 uh, they're saying it's okay to be romantic or sensual with teenagers. 489 00:27:05,310 --> 00:27:06,360 Like, great, terrific. 490 00:27:06,360 --> 00:27:09,480 That is, it's, it's been blowing up in a couple of corners of the internet in 491 00:27:09,480 --> 00:27:13,590 the last few days and none of the AI things will touch it understandably so, 492 00:27:13,590 --> 00:27:16,950 because it views, oh, there's, there's some stuff we don't make jokes about. 493 00:27:17,100 --> 00:27:18,120 Yes, I get that. 494 00:27:18,270 --> 00:27:19,740 I am not asking you to make those jokes. 495 00:27:19,740 --> 00:27:23,850 I'm asking you to basically skewer meta for its complete lack of ethics. 496 00:27:23,850 --> 00:27:28,110 There's a difference here, but the guardrails keep, uh, keep cropping up. 497 00:27:28,110 --> 00:27:30,150 And I get why companies have them in there. 498 00:27:30,150 --> 00:27:31,800 I'm not suggesting that they shouldn't. 499 00:27:32,310 --> 00:27:34,200 I wish it weren't as easy to get around it sometimes. 500 00:27:34,770 --> 00:27:34,890 Yeah. 501 00:27:34,890 --> 00:27:41,755 The, it's, it's, it's to the point where, um, I am way more interested. 502 00:27:42,915 --> 00:27:50,415 Private AI that I have been in the past, like I am very much a platform 503 00:27:50,475 --> 00:27:55,545 first for most modern technologies, if there's a cloud service for it, 504 00:27:56,295 --> 00:28:01,155 unless it's just a always on service where there's no economic value in it. 505 00:28:01,425 --> 00:28:04,245 Architecturally, I'm like, you know, let it be somebody else's problem. 506 00:28:04,430 --> 00:28:09,465 I, I don't want to manage AI drivers not. 507 00:28:10,125 --> 00:28:13,545 No desire to do it whatsoever until I run into 508 00:28:13,545 --> 00:28:16,275 these, uh, I was telling my wife about my check. 509 00:28:17,399 --> 00:28:21,480 My custom GPT problem and how I'm forced to have to build a 510 00:28:21,480 --> 00:28:26,310 more elegant solution for a platform that can handle millions 511 00:28:26,310 --> 00:28:33,000 of users A minute I have to build a custom application for. 512 00:28:33,725 --> 00:28:41,135 A half a dozen users in a week, but I want this tailored experience 513 00:28:41,135 --> 00:28:45,514 that when someone comes to, you know, critique or, uh, either 514 00:28:45,514 --> 00:28:50,254 critique my content or critique someone else's content, or ask kind 515 00:28:50,254 --> 00:28:53,915 of what would Keith think conceptually before I even engaged him. 516 00:28:54,720 --> 00:28:56,490 I want it to be a reliable service. 517 00:28:56,520 --> 00:28:59,820 I want it to be a experience that will meet 518 00:28:59,820 --> 00:29:02,820 the criteria fi finish of the CT advisor. 519 00:29:03,330 --> 00:29:09,540 And that requires this, you know, this, this stitching together of this 520 00:29:09,540 --> 00:29:17,490 low level AI so that you know, so that the experience, the desire UI, is. 521 00:29:17,940 --> 00:29:18,750 Is achieved. 522 00:29:19,170 --> 00:29:20,820 And I think this is the problem that a lot 523 00:29:20,820 --> 00:29:22,620 of enterprises are looking running into. 524 00:29:22,620 --> 00:29:24,960 They're looking at these big AI platforms. 525 00:29:26,010 --> 00:29:28,650 It's not meeting their immediate business needs. 526 00:29:28,710 --> 00:29:31,830 They could probably get away with a 7 billion parameter model, 527 00:29:32,220 --> 00:29:34,920 but they just need the expertise to either run it in the 528 00:29:34,920 --> 00:29:37,710 public cloud, which all the cloud providers are getting there. 529 00:29:37,710 --> 00:29:40,650 They're, they're providing the tools or they need to, 530 00:29:40,650 --> 00:29:43,260 uh, they need to hand stitch this stuff themselves. 531 00:29:44,639 --> 00:29:46,590 I, I'm interested in the local inference piece just 532 00:29:46,590 --> 00:29:49,860 because I want it to still be usable once the good times 533 00:29:49,860 --> 00:29:52,409 dry up and they stop throwing money into these things. 534 00:29:52,770 --> 00:29:55,320 Uh, because again, some of these AI tools are great 535 00:29:55,320 --> 00:29:57,689 for 20, 30 bucks a month, but not two or 3000. 536 00:29:57,960 --> 00:30:02,220 So I, I want a good enough version that I can run on my own hardware 537 00:30:02,429 --> 00:30:08,040 because I, now that I've had the benefit of having AI write the front 538 00:30:08,040 --> 00:30:10,980 end code, I don't know how to write, for example, I don't wanna go back, 539 00:30:11,790 --> 00:30:12,179 right. 540 00:30:13,410 --> 00:30:17,700 Yeah, I don't, I could, I'm watching it like in the background now. 541 00:30:18,630 --> 00:30:21,390 Uh, cursor's, troubleshooting the app that I'm building. 542 00:30:21,390 --> 00:30:26,040 I the, you know, it will tell you, oh, now go and try it. 543 00:30:26,610 --> 00:30:27,630 Like, wait, wait a minute, minute. 544 00:30:27,630 --> 00:30:30,030 I'm, I'm the grown up here. 545 00:30:30,090 --> 00:30:31,020 I'm in charge. 546 00:30:31,320 --> 00:30:32,490 You try it. 547 00:30:32,910 --> 00:30:36,870 The, uh, why You just write a write a test script and 548 00:30:36,870 --> 00:30:39,510 keep working through the problem until you've solved it. 549 00:30:39,960 --> 00:30:42,360 I don't wanna be engaged in your troubleshooting. 550 00:30:43,560 --> 00:30:46,440 I don't wanna go back to the days where I have to troubleshoot. 551 00:30:46,440 --> 00:30:49,260 I don't wanna have to use my connect to call 552 00:30:49,650 --> 00:30:52,980 someone at Google Cloud to get support for free. 553 00:30:53,430 --> 00:30:55,350 I just want AI to do that for me. 554 00:30:56,460 --> 00:30:59,910 Yeah, I, I want it to have some boundaries. 555 00:30:59,910 --> 00:31:01,800 Like I could never let it run loose on my laptop. 556 00:31:01,800 --> 00:31:03,660 I give it its own VM somewhere to run in 557 00:31:03,930 --> 00:31:05,550 because, you know, there, there's client data. 558 00:31:05,550 --> 00:31:07,560 I don't wanna smack it into, but there's a, 559 00:31:07,590 --> 00:31:09,660 but aside from that, just go iterate on this. 560 00:31:09,660 --> 00:31:10,200 Have fun. 561 00:31:10,200 --> 00:31:12,570 And if it doesn't work, oh well, but I don't wanna 562 00:31:12,570 --> 00:31:14,595 come back to a $10,000 monthly inference bill either. 563 00:31:15,615 --> 00:31:24,135 Well, this has been one of my fears with the AI services, uh, that somehow 564 00:31:24,165 --> 00:31:31,514 that I put a AI chat bot or something out there and someone figures out 565 00:31:31,545 --> 00:31:36,980 how to jailbreak it, and now they're just running their app through my. 566 00:31:38,580 --> 00:31:43,770 Through my high quality, you know, before where I would've used, uh, before 567 00:31:43,770 --> 00:31:52,050 where someone would've used flashlight or some, uh, for many type of of model, 568 00:31:52,889 --> 00:32:00,000 they now have access to my Gemini Pro or my five Pro, and they run ragged in it. 569 00:32:00,000 --> 00:32:06,570 And I get some $10,000 bill because of my poor security, because I've used a. 570 00:32:06,960 --> 00:32:08,010 AI to develop. 571 00:32:08,250 --> 00:32:15,000 So what I've been learning is how to do be a better software 572 00:32:15,000 --> 00:32:19,110 manager, a better product manager, as I've been building 573 00:32:19,110 --> 00:32:22,770 with AI assistance since it's been an a fascinating journey. 574 00:32:23,790 --> 00:32:24,540 Oh, absolutely. 575 00:32:24,690 --> 00:32:31,710 And I think that it's helpful to, to keep some form of, I guess, uh. 576 00:32:33,305 --> 00:32:34,260 Uh, distance in there. 577 00:32:34,260 --> 00:32:36,450 Like right now with inference being as easy to find as it is, most 578 00:32:36,450 --> 00:32:38,909 places, I feel like a lot of folks have not started really looking for 579 00:32:38,909 --> 00:32:42,149 how to get free inference, uh, the scammers and whatnot, but it'll come. 580 00:32:42,360 --> 00:32:43,169 It always does. 581 00:32:43,320 --> 00:32:44,520 And, and I'm the same way. 582 00:32:44,520 --> 00:32:47,340 Whenever I build something, I'm using the best top of line 583 00:32:47,340 --> 00:32:51,600 model because I'm not scaling this out to millions of users. 584 00:32:51,600 --> 00:32:53,190 I wanna make sure it works. 585 00:32:53,430 --> 00:32:56,040 I don't want to do it for the least amount of money possible because 586 00:32:56,040 --> 00:32:59,700 we're talking, well, I could do this for 15 cents a month instead of 70. 587 00:32:59,940 --> 00:33:00,540 And. 588 00:33:00,605 --> 00:33:03,335 I don't care about less than the cost of 589 00:33:03,335 --> 00:33:05,315 a candy bar when it comes to these things. 590 00:33:05,315 --> 00:33:07,415 I want it to work, and I want it to work well. 591 00:33:07,715 --> 00:33:12,215 Yeah, I think that, you know, we, we started down into kinda like the 592 00:33:12,305 --> 00:33:20,225 economic return of AI for entrepreneurs, solopreneurs, and small companies. 593 00:33:20,915 --> 00:33:23,015 There's been no doubt that. 594 00:33:23,835 --> 00:33:27,735 So right now it's just me and my wife still, but we both are avid 595 00:33:27,735 --> 00:33:33,255 users of AI and we just haven't found the need to hire yet because 596 00:33:33,555 --> 00:33:37,790 when you're dealing with an area that you have expertise in. 597 00:33:39,825 --> 00:33:44,415 You're, uh, dealing with ai, you 10 x yourself, and this is one 598 00:33:44,415 --> 00:33:48,195 of the things that I've talked to other really smart people about 599 00:33:48,195 --> 00:33:52,155 and that haven't really figured out, is how do you codify that? 600 00:33:52,275 --> 00:33:56,685 Like how do you, the way that Corey is using ai, how 601 00:33:56,685 --> 00:34:00,165 do I scale that to a hundred people in my organization? 602 00:34:00,795 --> 00:34:01,575 And that is. 603 00:34:02,010 --> 00:34:04,050 That has been the tricky part, right? 604 00:34:04,050 --> 00:34:07,680 Is you, you're able to do it because you have x 605 00:34:07,680 --> 00:34:11,370 number of years of systems, architecture experience. 606 00:34:11,610 --> 00:34:13,230 You understand systems. 607 00:34:13,620 --> 00:34:16,230 AI is just, oh, you, oh, I get this. 608 00:34:16,230 --> 00:34:19,500 This is like having a bunch of junior developers that 609 00:34:19,530 --> 00:34:23,639 have, you know, really, really massive amounts of, of. 610 00:34:24,659 --> 00:34:27,540 Photographic memory and I can put them to work. 611 00:34:27,540 --> 00:34:31,049 I have to put guard rails around them, but I can build these incredible systems. 612 00:34:31,679 --> 00:34:32,069 Yeah. 613 00:34:32,130 --> 00:34:34,650 Intelligence, stat of 18, wisdom stat of three. 614 00:34:35,279 --> 00:34:38,009 The, you know, you, you've done it with your career with 615 00:34:38,009 --> 00:34:41,880 AWS, you understand that, you know what, yeah, sure. 616 00:34:41,880 --> 00:34:45,029 One EC2 instances, and it's reliable as one VM 617 00:34:45,029 --> 00:34:49,679 in the data center, but three EC2 instances is. 618 00:34:50,295 --> 00:34:53,925 Yeah, I, I will say my use of AI is a whole bunch of onesie, twosies. 619 00:34:53,925 --> 00:34:57,825 I don't have any sustained processes that use it the same way every time. 620 00:34:58,005 --> 00:35:00,315 So I look at this and effectively say, oh, 621 00:35:00,315 --> 00:35:02,325 what, what job could I replace with this? 622 00:35:02,325 --> 00:35:04,905 It's, these aren't things I would hire people to do. 623 00:35:04,905 --> 00:35:07,665 I'm not gonna hire an artist, for example, to come up with a 624 00:35:07,665 --> 00:35:11,415 picture of a giraffe and a data center aisle, uh, for a dumb 625 00:35:11,415 --> 00:35:14,085 post on social media that gets three likes and one retweet. 626 00:35:14,115 --> 00:35:16,095 'cause no one else has the same sense of humor that I do. 627 00:35:16,425 --> 00:35:18,585 Uh, but I absolutely will make a robot do it. 628 00:35:20,024 --> 00:35:22,334 Yeah, I'm, I'm in the same space. 629 00:35:22,334 --> 00:35:22,845 I think 630 00:35:25,245 --> 00:35:28,365 the stuff that AI is really not good at is, like, if I needed to 631 00:35:28,365 --> 00:35:32,475 do a deck and I wanted that deck to look really good, I'm, I'm, 632 00:35:32,580 --> 00:35:35,685 I'm giving, you know, someone's paying me to deliver their keynote. 633 00:35:36,615 --> 00:35:37,455 I'm not gonna run it through ai. 634 00:35:37,605 --> 00:35:40,455 I'll, I'll get the outline from ai 'cause AI is good at 635 00:35:40,455 --> 00:35:44,625 language, but I'm gonna just send the, I'm going to send 636 00:35:44,625 --> 00:35:48,705 the deck off to a deck deck jockey to make it look good. 637 00:35:49,455 --> 00:35:53,774 The, uh, so, but it does enable me to do stuff, 638 00:35:53,834 --> 00:35:55,634 to your point that I would not have done. 639 00:35:55,935 --> 00:35:59,384 So I'm doing this, uh, daily series, the, uh, which is actually what 640 00:35:59,384 --> 00:36:03,410 sparked this, which is the cloud every day, whereas I'm looking at a, at. 641 00:36:03,915 --> 00:36:09,105 Platform engineering from a different angle, slightly different angle every day. 642 00:36:10,250 --> 00:36:14,930 I just, I asked AI for the, for a 10 day, uh, editorial calendar. 643 00:36:16,634 --> 00:36:21,285 And then, you know, I can now go through and say, Hey, gimme a rough draft, 644 00:36:21,464 --> 00:36:25,665 work with it for about 30 minutes a day, and I can churn out quality posts. 645 00:36:25,785 --> 00:36:28,185 I don't have to hire a ghost writer to do that now. 646 00:36:28,515 --> 00:36:28,785 Yeah. 647 00:36:28,964 --> 00:36:30,435 And you're not having, and you're not sitting 648 00:36:30,435 --> 00:36:32,055 there just copy and pasting it either, 649 00:36:32,505 --> 00:36:34,605 you know, a hundred so people see it a day. 650 00:36:34,935 --> 00:36:37,485 Not worth, not worth paying somebody else 651 00:36:37,485 --> 00:36:39,915 to do, but worth 15 minutes of my time. 652 00:36:41,190 --> 00:36:42,030 Exactly. 653 00:36:42,060 --> 00:36:43,350 'cause you never know what's gonna hit. 654 00:36:43,470 --> 00:36:46,590 And I, the idea of just doing a copy paste from 655 00:36:46,590 --> 00:36:49,110 AI and calling it a day is horrifying to me. 656 00:36:49,530 --> 00:36:52,110 But yeah, collaborating, helping it do a yes. 657 00:36:52,110 --> 00:36:55,830 And for me, the hardest thing that I have to overcome is the empty 658 00:36:55,830 --> 00:36:58,380 text editor in front of me when it's time to write something. 659 00:36:58,620 --> 00:36:58,950 So. 660 00:36:59,015 --> 00:37:02,404 Have it do a draft of something and even if, especially if it's wrong, 661 00:37:02,615 --> 00:37:05,615 but now, 'cause now I can correct the robot and I can mansplain to it or 662 00:37:05,615 --> 00:37:10,475 human mansplain to it, and suddenly I have a much stronger post for it. 663 00:37:11,535 --> 00:37:11,925 Yeah. 664 00:37:11,925 --> 00:37:14,775 And the great thing is even if you're not comfortable 665 00:37:14,775 --> 00:37:16,695 with that, you can just say, Hey, gimme an outline. 666 00:37:17,145 --> 00:37:19,665 The, when I did the a hundred days of ai, I said, you 667 00:37:19,665 --> 00:37:22,965 know what, give me a, give me a hundred days of AI of 668 00:37:22,965 --> 00:37:25,515 topics to cover every day for the next a hundred days. 669 00:37:25,965 --> 00:37:28,695 And I look up like, oh, here's today's topic. 670 00:37:29,205 --> 00:37:30,255 Ah, I don't like that. 671 00:37:30,825 --> 00:37:32,650 Now to another topic and then. 672 00:37:33,735 --> 00:37:36,825 Inevitably, three days later, that topic comes up and I have to figure 673 00:37:36,825 --> 00:37:42,225 out a new topic, but I digress that, that it is, I, I'm a fan with limits. 674 00:37:42,765 --> 00:37:42,885 Yeah. 675 00:37:42,885 --> 00:37:45,615 To bring this black full circle though, these are personal 676 00:37:45,615 --> 00:37:48,795 productivity accelerants for which AI is becoming invaluable. 677 00:37:49,335 --> 00:37:50,775 These are not enterprise stories. 678 00:37:51,315 --> 00:37:55,785 I still struggle to see the massive, wide scale enterprise upside, 679 00:37:55,785 --> 00:37:58,215 where we're gonna basically be able to make every one of our 680 00:37:58,215 --> 00:38:01,695 employees way more productive and replace entire teams with agents. 681 00:38:02,145 --> 00:38:04,365 I, I think that they're wish casting. 682 00:38:05,325 --> 00:38:11,205 Yeah, I that I've seen some really great examples of AI 683 00:38:11,205 --> 00:38:16,155 doing jobs no one else would've done or didn't want to do. 684 00:38:17,145 --> 00:38:20,055 Will, you know, I'm, I keep seeing, you know what? 685 00:38:20,085 --> 00:38:23,444 Dell Technologies hiring is laying off due to ai. 686 00:38:23,835 --> 00:38:26,924 Dell Technologies is not hiring, laying off doing ai. 687 00:38:26,924 --> 00:38:30,345 Dell technology is laying off 'cause they overhired during COVID COVID 688 00:38:30,585 --> 00:38:34,305 and they're trying to get back to a number that meets, that keeps 689 00:38:34,305 --> 00:38:38,805 Michael Dell, you know, happy that, that, that has nothing to do with ai. 690 00:38:40,125 --> 00:38:42,584 They need an AI story more than they need anything else. 691 00:38:42,584 --> 00:38:44,774 When it comes to these things, like I, and I still think 692 00:38:44,774 --> 00:38:47,174 that there's this constituency that that's using ai. 693 00:38:47,174 --> 00:38:49,575 Like I got an email the other day saying, I noticed you 694 00:38:49,575 --> 00:38:52,575 work with AWS costs, so let's talk about how to lower the 695 00:38:52,575 --> 00:38:56,234 cost of your voice phone system was the opening sentence. 696 00:38:56,384 --> 00:38:59,415 And I honestly couldn't figure out if it was AI run amok or just 697 00:38:59,415 --> 00:39:03,015 someone with not a lot of neurons to bang together to make sparks. 698 00:39:03,254 --> 00:39:05,654 Like on some level it almost doesn't matter 699 00:39:05,654 --> 00:39:07,875 'cause I'm not responding to the bad pitch. 700 00:39:08,525 --> 00:39:10,625 But it is something that makes me think 701 00:39:10,625 --> 00:39:13,445 that you still have to use it appropriately. 702 00:39:13,865 --> 00:39:16,715 You still have to use it appropriately, and it's not something that 703 00:39:16,720 --> 00:39:19,895 that, you know, we haven't really talked about agentic and all that, but 704 00:39:20,285 --> 00:39:27,845 just regular chat, I have, without a doubt, I have won deals because ai. 705 00:39:28,560 --> 00:39:32,610 Was able to flesh out details in the email that I typically ignore. 706 00:39:33,150 --> 00:39:34,950 I am not a big emailer. 707 00:39:34,950 --> 00:39:37,440 I, I'm an email, big email is, I like to use 708 00:39:37,440 --> 00:39:40,440 email, but I don't like pages and pages of emails. 709 00:39:40,800 --> 00:39:44,790 Evidently, my customers like pages and pages of details. 710 00:39:45,735 --> 00:39:48,225 Short email means that you're angry or annoyed 711 00:39:48,495 --> 00:39:49,905 and AI expands. 712 00:39:49,905 --> 00:39:52,065 Like, I don't really know what else to say in this email. 713 00:39:52,065 --> 00:39:56,835 And no AI will say, oh, you missed, you know, these three different topics and 714 00:39:57,015 --> 00:40:01,005 whether they read it or not, whatever, but it, it has been super effective. 715 00:40:02,160 --> 00:40:06,420 Yes, I am a big fan of what some of these things are unlocking. 716 00:40:06,810 --> 00:40:09,600 Uh, Keith, I really wanna thank you for taking the time to speak with me. 717 00:40:09,690 --> 00:40:12,900 If people wanna learn more, where's the best place for them to find you? 718 00:40:13,380 --> 00:40:15,090 The CTO advisor.com. 719 00:40:15,180 --> 00:40:17,910 I'm still trying to buy CTO advisor.com. 720 00:40:18,330 --> 00:40:20,370 The guy wants $1,200 for it. 721 00:40:20,400 --> 00:40:21,720 I refuse. 722 00:40:21,750 --> 00:40:23,970 Now someone's gonna out and pay the 1200 bucks. 723 00:40:23,970 --> 00:40:25,380 I don't really care but. 724 00:40:25,595 --> 00:40:27,339 He, he wants 1200 bucks for it. 725 00:40:27,370 --> 00:40:28,089 I refuse. 726 00:40:28,095 --> 00:40:30,850 The, the, the, the, the site has been abandoned 727 00:40:30,850 --> 00:40:34,930 for years, but, ah, the ct advisor.com 728 00:40:35,259 --> 00:40:35,740 oh, oh. 729 00:40:35,740 --> 00:40:38,529 I have a couple of site domains like that and I've made overtures and when 730 00:40:38,529 --> 00:40:41,140 you come back with a number that has two commas in it, we're done here. 731 00:40:41,140 --> 00:40:44,080 There is no conversation that we are going to have that's 732 00:40:44,080 --> 00:40:46,089 going to lead to an outcome that we're both happy with. 733 00:40:46,600 --> 00:40:46,779 I bet. 734 00:40:46,779 --> 00:40:49,180 We'll of course, put links to that in the show notes. 735 00:40:49,240 --> 00:40:51,040 Keith, thank you so much for your time, 736 00:40:51,400 --> 00:40:52,480 Corey, thanks again. 737 00:40:52,480 --> 00:40:52,810 It's always fun. 738 00:40:54,255 --> 00:40:56,475 Keith Townsend, the CTO advisor. 739 00:40:56,625 --> 00:40:59,775 I'm Cloud economist Cory Quinn, and this is Screaming In the Cloud. 740 00:40:59,955 --> 00:41:02,445 If you've enjoyed this podcast, please leave a five 741 00:41:02,445 --> 00:41:04,725 star review on your podcast platform of choice. 742 00:41:04,785 --> 00:41:08,085 Whereas if you've hated this podcast, please leave a five star review on 743 00:41:08,085 --> 00:41:11,985 your podcast platform of choice, along with an angry comment written by Ai. 744 00:41:12,105 --> 00:41:15,765 Heck, while you're at it, have it Leave angry comments on all the platforms.