1 00:00:00,000 --> 00:00:03,019 The picture behind you, like everything's on fire, that's actually my OpenClaw 2 00:00:03,289 --> 00:00:04,489 server a couple of times already. 3 00:00:04,889 --> 00:00:09,029 I mean, I set everything up and, when you set it up as root, root, that's obviously 4 00:00:09,039 --> 00:00:10,709 bad advice, you should never do that. 5 00:00:11,109 --> 00:00:12,389 So it's running as root. 6 00:00:12,459 --> 00:00:15,389 And I got really frustrated in Telegram and I'm writing something 7 00:00:15,399 --> 00:00:18,449 and I said, just delete it, delete, you know, whatever it is. 8 00:00:18,499 --> 00:00:23,089 Well, OpenClaw took it literally and deleted the entire project and down to 9 00:00:23,089 --> 00:00:25,179 the root and just deleted the server. 10 00:00:25,209 --> 00:00:26,429 No longer responsive. 11 00:00:26,929 --> 00:00:27,719 I was like, wow. 12 00:00:27,809 --> 00:00:28,789 Okay, that worked. 13 00:00:32,786 --> 00:00:35,936 Hey, welcome to the Agentic DevOps podcast. 14 00:00:35,936 --> 00:00:37,266 I am your host, Bret. 15 00:00:37,286 --> 00:00:42,866 And this episode I've got my friend Brian Christner coming in from Europe. 16 00:00:42,866 --> 00:00:44,426 We've known each other for a decade. 17 00:00:44,426 --> 00:00:49,346 He's a Docker Captain alumni, and really one of these leading edge 18 00:00:49,346 --> 00:00:52,661 tech guys that's always focused on cloud native, leading edge tech. 19 00:00:52,681 --> 00:00:57,391 He actually runs an online casino company out of Switzerland. 20 00:00:57,391 --> 00:01:01,771 And we don't really get too much into the weeds of that whole world, 21 00:01:01,771 --> 00:01:05,841 but he runs developer teams, he manages an organization, he has to 22 00:01:05,841 --> 00:01:07,851 worry about security policy for AI. 23 00:01:07,851 --> 00:01:09,891 And so we talk a little bit about that. 24 00:01:10,111 --> 00:01:14,471 This whole episode is mostly around what we're both seeing in terms of 25 00:01:14,537 --> 00:01:19,841 tooling, agent harnesses, the models that we recommend and pick right now, 26 00:01:19,841 --> 00:01:23,291 because right now there is at least a distinct difference between the 27 00:01:23,741 --> 00:01:29,501 very top of the line models like Opus and GPT 5.4 versus everything else. 28 00:01:30,281 --> 00:01:31,871 And we get in the weeds of it. 29 00:01:31,871 --> 00:01:35,381 We talk about security strategy around OpenClaw because if you've 30 00:01:35,381 --> 00:01:37,781 heard about that thing in the last couple of months, like it is 31 00:01:37,781 --> 00:01:39,911 definitely capturing the zeitgeist. 32 00:01:39,911 --> 00:01:45,881 And then even a new project called NanoClaw, which is a variant of 33 00:01:46,021 --> 00:01:49,331 OpenClaw-- or really, I should say a, I guess a competitor technically, 'cause 34 00:01:49,331 --> 00:01:52,631 it sounds like it's, it's written from scratch-- and it runs natively in Docker 35 00:01:52,631 --> 00:01:55,631 containers, which is the whole goal here. 36 00:01:55,631 --> 00:01:59,916 I'm really hoping that very soon we're all gonna be running these things in 37 00:01:59,916 --> 00:02:03,426 some sort of encapsulated protection area, like a container provides, right? 38 00:02:03,816 --> 00:02:09,666 We've seen Claude Cowork recently shift to an actual VM, so if you're interested in 39 00:02:09,666 --> 00:02:13,746 Claude Cowork, we know that that thing is isolated on your machine, away from your 40 00:02:13,746 --> 00:02:17,586 standard operating system, 'cause it runs in Linux VM, just like Docker containers. 41 00:02:18,036 --> 00:02:21,156 But it's nice to see new projects coming out that are doubling down on 42 00:02:21,156 --> 00:02:25,296 container use because I've been waiting for the industry to finally catch up 43 00:02:25,296 --> 00:02:28,416 with the security practices and all of the isolation technologies that 44 00:02:28,416 --> 00:02:32,856 we've built for the last few decades and make that the standard for using 45 00:02:32,856 --> 00:02:38,316 AI agents, which as we know sometimes hallucinate and tend to install tools 46 00:02:38,316 --> 00:02:41,766 and do things on your local system that maybe you don't want them to do. 47 00:02:42,186 --> 00:02:48,546 So it's a great conversation covering a wide array of AI issues, and I hope you 48 00:02:48,546 --> 00:02:50,026 enjoy this episode with Brian Christner. 49 00:02:50,814 --> 00:02:51,634 hello, Brian. 50 00:02:52,134 --> 00:02:52,684 Hey, Bret. 51 00:02:52,834 --> 00:02:53,364 How you doing? 52 00:02:53,864 --> 00:02:54,124 All right. 53 00:02:54,134 --> 00:02:54,974 I'm glad to have you here. 54 00:02:55,004 --> 00:02:55,944 this is Brian Christner. 55 00:02:56,064 --> 00:02:58,424 We are on the opposite sides of the pond from each other. 56 00:02:58,669 --> 00:03:03,039 I'm over in North America, he's in Europe and Switzerland So Brian, so Brian and 57 00:03:03,039 --> 00:03:06,269 I go back, over, I think over a decade at this point, at least probably a 58 00:03:06,269 --> 00:03:11,219 decade cause you're an OG captain and we, so we've got a lot of experience 59 00:03:11,219 --> 00:03:12,669 in containers and infrastructure. 60 00:03:12,669 --> 00:03:12,689 sure. 61 00:03:13,109 --> 00:03:17,879 Now you manage teams of engineers, is that an accurate statement or a team? 62 00:03:17,889 --> 00:03:21,159 Yeah, so I'm part of our organization. 63 00:03:21,159 --> 00:03:25,139 We have a development team that's handling a lot of our infrastructure that 64 00:03:25,359 --> 00:03:26,769 they're building Kubernetes clusters. 65 00:03:26,769 --> 00:03:30,209 They have everything built there, but, essentially I'm running an online casino. 66 00:03:30,709 --> 00:03:35,259 one of the top three largest casinos in Switzerland, and we obviously have 67 00:03:35,259 --> 00:03:39,489 a lot of data and this data is in, in the massive database, but it's 68 00:03:39,489 --> 00:03:42,179 all orchestrated with Kubernetes and our services that we are creating. 69 00:03:42,679 --> 00:03:43,069 Yeah. 70 00:03:43,569 --> 00:03:45,839 And we've had you on and talked about some of that stuff. 71 00:03:45,839 --> 00:03:48,529 We actually had a really popular episode with Brian a couple of years 72 00:03:48,529 --> 00:03:52,019 ago that I probably should have pulled up because we were talking about jobs. 73 00:03:52,039 --> 00:03:55,959 Of course, that has aged not like fine wine because job 74 00:03:56,319 --> 00:03:57,819 market right now is insanity. 75 00:03:58,129 --> 00:03:58,619 Yeah. 76 00:03:58,899 --> 00:04:01,459 I'm not sure any of those rules have changed though, but we 77 00:04:01,479 --> 00:04:05,839 talked about like DevOps roles and operations engineering roles. 78 00:04:06,234 --> 00:04:10,764 And like helping people sort of trying to give them like junior career advice, 79 00:04:10,764 --> 00:04:14,284 like how, you know, how to, how to help you stand out from the crowd. 80 00:04:14,769 --> 00:04:19,279 I think now everyone's just scrambling to know as much AI as possible, anticipating 81 00:04:19,279 --> 00:04:24,159 some sort of job shortage that only will be saved by you knowing AI, which 82 00:04:24,159 --> 00:04:26,259 is kind of, I mean, we say it nicely. 83 00:04:26,259 --> 00:04:30,189 We say it with like, I think the industry has been saying, you know, the, you 84 00:04:30,189 --> 00:04:33,049 won't be replaced by AI, you'll be replaced by someone who knows AI, I 85 00:04:33,069 --> 00:04:34,469 think is the thing that we all say. 86 00:04:37,399 --> 00:04:39,339 I'm not convinced that that's entirely true. 87 00:04:39,449 --> 00:04:43,359 I'm getting more anecdotes or well, directly from friends. 88 00:04:43,359 --> 00:04:47,864 I just talked to someone this last week that is losing their lead 89 00:04:47,874 --> 00:04:50,754 DevOps engineer that isn't sure they're going to get permission 90 00:04:50,754 --> 00:04:51,894 from management to replace them. 91 00:04:51,894 --> 00:04:57,274 So they're scrambling to figure out if AI can, like, help them automate some 92 00:04:57,274 --> 00:05:00,984 of the toil that the DevOps engineer was doing, like, like the daily stuff that 93 00:05:00,994 --> 00:05:07,024 the normal activities and if they can like use skills or I was advocating for skills. 94 00:05:07,024 --> 00:05:09,374 I've been in love with skills for a couple of months now. 95 00:05:09,874 --> 00:05:11,384 And I'm sure we're going to talk to each fan as well. 96 00:05:11,384 --> 00:05:11,684 Yep. 97 00:05:11,934 --> 00:05:14,244 I'm interested in hearing Brian's stories about OpenClaw. 98 00:05:14,244 --> 00:05:15,334 We're going to talk about that a little bit. 99 00:05:15,834 --> 00:05:19,964 But, I'm not sure we're going to specifically talk about DevOps AI 100 00:05:19,984 --> 00:05:23,774 advice but that's my whole jam is trying to figure out where these 101 00:05:23,774 --> 00:05:27,014 tools work and helping us with, obviously they can write Terraform. 102 00:05:27,014 --> 00:05:29,914 Obviously they can write, you know, GitHub actions, workflow files. 103 00:05:29,914 --> 00:05:31,314 Like that's a thing, but how do we. 104 00:05:31,554 --> 00:05:34,744 How do we automate things that weren't previously automatable by putting 105 00:05:34,744 --> 00:05:39,344 in a non deterministic robot is kind of where my head's at lately. 106 00:05:39,684 --> 00:05:43,674 I think I wanted to hear from you on other than OpenClaw, which I 107 00:05:43,674 --> 00:05:46,514 guess we really, that's not, we technically, I guess we call that an 108 00:05:46,524 --> 00:05:51,284 agent harness, but it's not a harness for code necessarily or text writing. 109 00:05:51,774 --> 00:05:56,629 I mean a way to describe OpenClaw sorry to interrupt you Bret is like 110 00:05:56,629 --> 00:06:02,494 OpenClaw is essentially an orchestrator of LLMs because you're able to connect 111 00:06:02,494 --> 00:06:04,549 to any LLM lLM you want out there. 112 00:06:04,549 --> 00:06:08,609 So you can connect to OpenAI, Gemini, Cloud, whatever you want. 113 00:06:09,029 --> 00:06:11,539 And you can switch between the two and you can really configure 114 00:06:11,539 --> 00:06:12,519 down to the agent level. 115 00:06:12,559 --> 00:06:16,279 And that's where I find the superpower of OpenClawd at the moment because 116 00:06:16,279 --> 00:06:18,849 it's open source and you can really configure it however you want. 117 00:06:19,349 --> 00:06:23,679 So you're not tied into a specific LLM of the day, flavor of the day LLM. 118 00:06:24,179 --> 00:06:28,769 Right, but you wouldn't use necessarily OpenClaw to stare at code while you're, 119 00:06:29,109 --> 00:06:30,529 like you would Claude Code, right? 120 00:06:30,559 --> 00:06:34,799 Like you wouldn't, it's not a TUI in the way that we think of like 121 00:06:34,819 --> 00:06:37,119 Claude Code or OpenCode or VS Code. 122 00:06:37,619 --> 00:06:37,949 So 123 00:06:37,979 --> 00:06:40,549 I'm thinking more like kicking off tasks, like, Hey, can 124 00:06:40,549 --> 00:06:42,029 you go research this for me? 125 00:06:42,059 --> 00:06:45,319 Can you go, you know, build this marketing plan? 126 00:06:45,534 --> 00:06:47,429 These type of tasks, perfect use case. 127 00:06:47,429 --> 00:06:50,164 But I mean, like coding, like you said, it's better than terminal. 128 00:06:50,664 --> 00:06:50,984 Okay. 129 00:06:51,184 --> 00:06:53,354 So before we get into OpenClaw, because I feel like the whole 130 00:06:53,354 --> 00:06:55,874 episode could be consumed by that. 131 00:06:56,024 --> 00:06:56,834 And I want to get to that. 132 00:06:57,374 --> 00:07:01,644 I wanted to talk first around, like, what are you hearing? 133 00:07:01,684 --> 00:07:04,914 Like, what is the general vibe in the company? 134 00:07:04,934 --> 00:07:08,214 Are you allowed to use AI at all, even in your industry? 135 00:07:08,214 --> 00:07:10,824 Is that something that, that your engineers are able to take advantage of? 136 00:07:10,824 --> 00:07:11,664 And if so, like, what? 137 00:07:12,164 --> 00:07:13,464 What are some of the things you're seeing out? 138 00:07:13,464 --> 00:07:15,104 Are they choosing Claude Code? 139 00:07:15,124 --> 00:07:17,224 Are, is, are you seeing that happening? 140 00:07:17,724 --> 00:07:19,394 I was just looking for a firsthand. 141 00:07:20,764 --> 00:07:25,954 We really have like an AI enabler within the company and she does an amazing job. 142 00:07:25,954 --> 00:07:28,184 She like onboards everybody within the company. 143 00:07:28,184 --> 00:07:31,624 And think of us, we're an online casino and we have a physical casino as well. 144 00:07:31,634 --> 00:07:36,584 So we have everything from cooks to like casino technicians to online developers. 145 00:07:36,584 --> 00:07:37,564 We have everything in between. 146 00:07:37,984 --> 00:07:41,334 And the onboarding of these different skills is just amazing. 147 00:07:41,334 --> 00:07:42,784 Cause I mean, like, our chef. 148 00:07:43,284 --> 00:07:47,664 Our executive chef, I mean, he's built a complete course around 149 00:07:47,674 --> 00:07:50,614 how you can make like, AI, brunches and everything like this. 150 00:07:50,614 --> 00:07:53,354 How you do all your ordering and everything through AI. 151 00:07:53,694 --> 00:07:56,254 So, I mean, we're really enabling everyone within the 152 00:07:56,254 --> 00:07:57,654 organization, which is really cool. 153 00:07:58,024 --> 00:07:58,414 Yeah. 154 00:07:58,504 --> 00:08:01,164 The challenge is, there's, it's moving so fast. 155 00:08:01,164 --> 00:08:05,244 So, trying to keep everyone updated and keeping the guardrails 156 00:08:05,244 --> 00:08:06,704 in place is very difficult. 157 00:08:06,764 --> 00:08:09,984 And we don't want to load any customer data in there, that's 158 00:08:10,094 --> 00:08:11,964 definitely a no go for us. 159 00:08:12,044 --> 00:08:14,074 So we're keeping no go out, all customer data. 160 00:08:14,534 --> 00:08:20,164 But as far as like, just, yeah, normal KPIs or like marketing, these types 161 00:08:20,164 --> 00:08:24,024 of things where it's not the customer impacting, that's no problem for us. 162 00:08:24,574 --> 00:08:28,034 Are they prescriptive or as an organization, are you prescriptive in 163 00:08:28,034 --> 00:08:32,934 like what harnesses or IDEs people use or people able to just, is it like when 164 00:08:32,934 --> 00:08:38,094 you think of your, from your security mindset, are you focused on particular 165 00:08:38,204 --> 00:08:42,824 model companies or cause organizations I've seen, like, there's all sorts of 166 00:08:42,824 --> 00:08:45,774 things, everything's wild, wild west, but you know, some companies are like, 167 00:08:45,864 --> 00:08:49,534 you know, we have Gemini and that's the, and that is what we bought as a company 168 00:08:49,754 --> 00:08:51,574 and that's what all developers will use. 169 00:08:51,869 --> 00:08:54,689 And so then you're kind of, as an engineer, you're sort of limited 170 00:08:54,689 --> 00:08:56,279 to whatever the company buys you. 171 00:08:56,739 --> 00:08:59,569 And then you sort of see this shadow AI where people sort of sneak out and 172 00:08:59,569 --> 00:09:00,869 do something that's maybe not approved. 173 00:09:01,289 --> 00:09:02,279 How do you think of it? 174 00:09:02,289 --> 00:09:05,569 Do you think, like, you've got a plethora of things and everybody 175 00:09:05,569 --> 00:09:06,779 can choose from a smorgasbord. 176 00:09:06,789 --> 00:09:07,619 How does that go? 177 00:09:08,119 --> 00:09:10,839 So, I mean, we, we have like a AI steering committee where we 178 00:09:10,839 --> 00:09:13,639 actually decide which tools we should onboard within the company. 179 00:09:14,029 --> 00:09:16,769 And it's really like, you know, where's the data stored and what 180 00:09:16,769 --> 00:09:19,069 kind of data protection do they have, these types of things. 181 00:09:19,489 --> 00:09:22,089 But I mean, generally all the tools are available to us. 182 00:09:22,129 --> 00:09:24,119 That's most people would use. 183 00:09:24,619 --> 00:09:29,189 There are some exceptions, you know, some of the Chinese models and whatnot, it's 184 00:09:29,189 --> 00:09:32,549 just, they don't follow the European data regulation, so we can't onboard them. 185 00:09:33,049 --> 00:09:34,289 And that's a hard no for us. 186 00:09:34,499 --> 00:09:37,009 But otherwise, I mean, everything's quite open for us. 187 00:09:37,019 --> 00:09:40,709 So, I mean, we have Gemini in use, we have Perplexity, we have OpenClaw, 188 00:09:41,369 --> 00:09:44,759 I mean, not OpenClaw, Claude, but we have all these things running. 189 00:09:45,169 --> 00:09:48,519 So it's basically whatever your use case is, you can kind of 190 00:09:48,519 --> 00:09:50,369 pull that specific model down. 191 00:09:50,779 --> 00:09:55,239 Now for the, the majority of people within the company, we try to onboard 192 00:09:55,239 --> 00:09:57,839 them to Copilot, just because it's easier. 193 00:09:58,119 --> 00:10:02,489 Yeah, the guardrails are much easier to control and yeah, we 194 00:10:02,489 --> 00:10:03,629 know what's going into the model. 195 00:10:03,659 --> 00:10:07,189 Now, developers, obviously they want more, they want more control. 196 00:10:07,289 --> 00:10:10,679 So I mean, they're going direction OpenAI and Claude. 197 00:10:11,199 --> 00:10:11,579 Okay. 198 00:10:11,709 --> 00:10:14,709 So when you say, when you said, Copilot, you're specifically talking 199 00:10:14,719 --> 00:10:16,659 about Microsoft Copilot, not GitHub. 200 00:10:16,719 --> 00:10:17,529 Microsoft Copilot. 201 00:10:17,529 --> 00:10:17,689 Yeah. 202 00:10:17,729 --> 00:10:17,989 Yeah. 203 00:10:18,204 --> 00:10:18,434 Yeah. 204 00:10:18,434 --> 00:10:18,764 Okay. 205 00:10:19,184 --> 00:10:21,344 horrible branding by Microsoft, right? 206 00:10:21,384 --> 00:10:22,614 So everyone gets confused. 207 00:10:23,474 --> 00:10:24,314 Naming is hard, man. 208 00:10:24,334 --> 00:10:25,114 Naming is hard. 209 00:10:25,574 --> 00:10:26,944 I mean, I do appreciate it. 210 00:10:26,964 --> 00:10:30,944 if it was, I feel like it'd be much more Microsofty if it was more like every one 211 00:10:30,944 --> 00:10:35,864 of them was Copilot, but the la instead of it being very vague and confusing 212 00:10:36,104 --> 00:10:39,524 that there's another couple of words after that that we have to say, right? 213 00:10:39,524 --> 00:10:39,849 Like, yeah. 214 00:10:39,939 --> 00:10:40,429 Exactly. 215 00:10:40,904 --> 00:10:41,114 Yeah. 216 00:10:41,114 --> 00:10:45,924 I would, if it was Microsoft, original, it would be like Microsoft Copilot 217 00:10:46,324 --> 00:10:49,284 something edition, you know, V two or so. 218 00:10:49,344 --> 00:10:52,014 And then that, and we'd have to say that because then there's also this other thing 219 00:10:52,014 --> 00:10:55,184 that's other edition and we'd have to have an acronym dash dash because it would be 220 00:10:55,184 --> 00:10:57,014 too long to say, so we'd have to right. 221 00:10:58,124 --> 00:11:01,034 Yeah, because I'm obviously, because I'm such a GitHub person, 222 00:11:01,034 --> 00:11:02,644 I mean, I'm teaching GitHub. 223 00:11:02,684 --> 00:11:07,844 I've consulted on GitHub for at least six years, specifically around GitHub actions, 224 00:11:07,884 --> 00:11:10,724 because that's, I'm not necessarily teaching people how to do commits. 225 00:11:10,724 --> 00:11:13,294 I'm teaching them more about workflows and automation and 226 00:11:13,664 --> 00:11:15,064 deployments and all that stuff. 227 00:11:15,464 --> 00:11:19,344 Obviously my Copilot world is rolled around in GitHub Copilot. 228 00:11:19,364 --> 00:11:22,664 I am aware that there is a Microsoft Copilot and I technically. 229 00:11:22,679 --> 00:11:26,719 I think I used it a few times with an Office 365 account, but I see it a lot 230 00:11:26,729 --> 00:11:29,649 with other like non tech business friends. 231 00:11:29,659 --> 00:11:33,089 I see them in chat on a mobile, on the mobile Copilot app. 232 00:11:34,014 --> 00:11:38,334 And they talk about their work AI a lot, so that's a big win for Microsoft there. 233 00:11:38,404 --> 00:11:39,514 I mean, it's decent. 234 00:11:39,534 --> 00:11:44,214 I'm not a massive fan of it because I mean, the hurdle to automate something 235 00:11:44,214 --> 00:11:46,594 with this is just, I mean, it's massive. 236 00:11:46,624 --> 00:11:52,674 I mean, for example, to automate an email coming into your inbox and then 237 00:11:52,674 --> 00:11:57,764 feeding it to Copilot and having it all working every day, is a mission. 238 00:11:58,234 --> 00:11:59,934 So, I mean, ClipPilot. 239 00:11:59,994 --> 00:12:01,124 I like that one from James. 240 00:12:01,124 --> 00:12:01,374 Yeah. 241 00:12:02,354 --> 00:12:02,834 Clippy. 242 00:12:02,835 --> 00:12:03,242 Clippy. 243 00:12:03,242 --> 00:12:03,649 ClippyPilot. 244 00:12:04,149 --> 00:12:05,429 but I mean, generally it's getting better. 245 00:12:05,429 --> 00:12:09,109 It's just, I mean, when you compare it, it's supposed to be OpenAI, right? 246 00:12:09,519 --> 00:12:12,639 But if you compare a Copilot to OpenAI, it's night and day difference. 247 00:12:12,849 --> 00:12:13,249 Okay. 248 00:12:13,309 --> 00:12:13,549 Yeah. 249 00:12:13,549 --> 00:12:15,989 So it's my, maybe an older, or there's some, yeah, there's 250 00:12:15,989 --> 00:12:17,929 hidden, hidden, it's obscured away. 251 00:12:17,929 --> 00:12:18,129 Right. 252 00:12:18,129 --> 00:12:20,219 So we're not sure about what's going on in the background. 253 00:12:20,269 --> 00:12:20,599 Exactly. 254 00:12:20,769 --> 00:12:22,669 Cause that, I mean, that's the general vibe. 255 00:12:22,679 --> 00:12:24,619 I mean, we haven't, we talked about this. 256 00:12:24,749 --> 00:12:28,769 Like since Opus 4.5 was released in November, I think the general vibe 257 00:12:28,769 --> 00:12:32,569 in the industry is like, okay, this, that December might be, I think I 258 00:12:32,809 --> 00:12:34,209 was framing it this way yesterday. 259 00:12:34,709 --> 00:12:40,059 December is the epoch of many of us suddenly taking AI 260 00:12:40,069 --> 00:12:42,589 seriously as an engineering tool. 261 00:12:43,089 --> 00:12:46,399 as opposed to it being a fancy auto complete. 262 00:12:46,399 --> 00:12:52,249 And I spend far less time arguing, I spend far less time dealing with 263 00:12:52,289 --> 00:12:55,909 hallucinations, and I think the way we're going about it with some of these 264 00:12:55,919 --> 00:13:03,429 modern concepts of skills over skills and maybe like, CLI over MCP, not that 265 00:13:03,429 --> 00:13:07,904 MCP's going away, but I think that It's hype cycle has been pretty fast to go 266 00:13:07,904 --> 00:13:09,974 like all the way up to maximum hype. 267 00:13:10,014 --> 00:13:10,854 Incredible, right? 268 00:13:10,894 --> 00:13:13,634 And then like, everyone's like, ah, it's not really that great. 269 00:13:13,644 --> 00:13:17,304 Actually, I'm not really even using any, like, I just tell the, I just 270 00:13:17,304 --> 00:13:18,844 tell the tool to use my CLI tools. 271 00:13:19,114 --> 00:13:22,174 And so now you see things like I just saw last week, was it? 272 00:13:22,674 --> 00:13:28,404 I think it's Google who launched a Google Workspace or Google Docs, Google Sheets, 273 00:13:28,424 --> 00:13:31,424 Google Drive CLI, I think, I think it's 274 00:13:32,284 --> 00:13:32,974 really good. 275 00:13:33,474 --> 00:13:33,834 Is it good? 276 00:13:33,864 --> 00:13:34,044 Okay. 277 00:13:34,044 --> 00:13:34,134 Yeah. 278 00:13:34,264 --> 00:13:34,534 But it's like 279 00:13:34,534 --> 00:13:37,314 essentially like a MCP to Google's ecosystem. 280 00:13:37,704 --> 00:13:38,034 Yeah. 281 00:13:38,104 --> 00:13:39,074 And it's quite nice. 282 00:13:39,124 --> 00:13:42,434 I mean, now if you get rate limited or whatever, I don't know yet, but I'll 283 00:13:42,434 --> 00:13:43,924 definitely, it's on my to do list, 284 00:13:44,424 --> 00:13:45,654 but it's a CLI tool, right? 285 00:13:45,694 --> 00:13:47,674 My, my understanding is it's a CLI tool, right? 286 00:13:47,674 --> 00:13:52,924 So instead of us having to add an MCP tool with 50 or a hundred different 287 00:13:53,124 --> 00:13:57,244 MCP tools in it, and then worry about security and authentication, we're 288 00:13:57,244 --> 00:14:02,334 just using a CLI tool and that CLI tool manages its own authentication. 289 00:14:02,354 --> 00:14:06,384 And it's, it doesn't, it only has to know the help file of the CLI 290 00:14:06,394 --> 00:14:07,754 tool and the command structure. 291 00:14:07,964 --> 00:14:10,564 It doesn't have to know all these MCP tools, which I 292 00:14:10,564 --> 00:14:11,954 believe is better on tokens. 293 00:14:11,994 --> 00:14:14,844 If I remember correctly, looking into some of that, like. 294 00:14:14,854 --> 00:14:17,774 Oh, I can tell you token horror stories on the OpenClaw. 295 00:14:18,294 --> 00:14:21,594 I've burned my credit card into many bits and pieces. 296 00:14:22,046 --> 00:14:22,206 Yeah, 297 00:14:22,236 --> 00:14:24,266 being like, an early adopter of OpenClaw. 298 00:14:24,936 --> 00:14:27,936 A friend of mine yesterday said he tried the fast mode in Opus and 299 00:14:27,966 --> 00:14:31,646 like an hour later, he had like a 25, bill or whatever like that. 300 00:14:31,646 --> 00:14:33,276 And he's like, okay, maybe not fast mode. 301 00:14:33,276 --> 00:14:33,876 It's very nice. 302 00:14:33,886 --> 00:14:38,946 But, I, you know, we're all challenged as well with AIs are getting faster, 303 00:14:38,966 --> 00:14:40,546 but we still have the waiting game. 304 00:14:40,556 --> 00:14:45,726 And that creates like this flow problem that I don't know how 305 00:14:45,726 --> 00:14:50,686 to describe it, but you end up feeling like you need to multitask. 306 00:14:51,831 --> 00:14:55,681 Well, the thing, the thing is with the, I mean, everyone's complaining about memory 307 00:14:55,681 --> 00:14:57,201 and all these different models, right? 308 00:14:57,521 --> 00:14:59,911 So everyone's saying, Hey, it forgets what I'm doing. 309 00:14:59,911 --> 00:15:03,071 I constantly have to remind my model to that. 310 00:15:03,091 --> 00:15:04,471 My name is Brian, whatever. 311 00:15:04,771 --> 00:15:08,681 But the thing is, we're building out these models, these memory models, 312 00:15:08,731 --> 00:15:11,811 which means we're having more context we're passing to the model each time. 313 00:15:12,591 --> 00:15:16,121 So the model is growing and we're just bombarding these models with 314 00:15:16,131 --> 00:15:18,511 tons of tokens just to say I'm Brian. 315 00:15:19,011 --> 00:15:22,551 You know, and here's, that's why the costs are just exploding. 316 00:15:22,981 --> 00:15:23,451 Yeah. 317 00:15:23,951 --> 00:15:27,371 I mean, the costs are exploding because we want more memory at the moment. 318 00:15:27,921 --> 00:15:30,391 I've talked a lot about OpenCode lately, so we're going to get 319 00:15:30,391 --> 00:15:31,321 into OpenClaw in a minute. 320 00:15:31,321 --> 00:15:34,861 But I think, I think we are, we're both, I think of an agreement that Claude 321 00:15:34,881 --> 00:15:36,081 Code is probably something that's. 322 00:15:36,526 --> 00:15:41,116 In our radar, I've been using, I mean, it was kind of, I think, I 323 00:15:41,116 --> 00:15:42,726 feel like the first successful TUI. 324 00:15:43,006 --> 00:15:47,276 I was an early adopter of VS Code's Copilot back in the beta, I think it was 325 00:15:47,276 --> 00:15:54,236 like 2023, maybe, when the beta of Copilot came out and I think we had a show here 326 00:15:54,426 --> 00:16:00,806 like the day after, Copilot came out and we haven't exactly been the AI channel. 327 00:16:00,806 --> 00:16:04,276 I would, I definitely have not pivoted to be like everything AI, like a lot 328 00:16:04,276 --> 00:16:08,716 of my friends are, much to my detriment on YouTube, but the things that I have 329 00:16:08,716 --> 00:16:14,406 started to gravitate to are shying away from things like OpenCode or 330 00:16:14,406 --> 00:16:18,666 Codex specifically simply because I find myself wanting to switch models 331 00:16:18,666 --> 00:16:22,476 too often, and I don't want to be locked into a harness that's locked 332 00:16:22,476 --> 00:16:24,456 into a particular vendor's models. 333 00:16:24,836 --> 00:16:29,436 So I started trying out OpenCode in December, moving from Claude Code, and 334 00:16:29,436 --> 00:16:34,896 I at the time was actually trying The copilot, CLI, the Codex, CLI, before Codex 335 00:16:34,896 --> 00:16:40,186 actually had a GUI and the Claude Code CLI obviously, and I shifted to OpenCode 336 00:16:40,306 --> 00:16:42,046 and now it's the only thing I use. 337 00:16:42,096 --> 00:16:43,866 And I love that. 338 00:16:43,916 --> 00:16:46,946 I have a Copilot subscription, which is technically like open router. 339 00:16:47,446 --> 00:16:49,096 Mm-hmm . 'cause it gives me the, you know, the tall, all 340 00:16:49,096 --> 00:16:50,476 the hits, all the great models. 341 00:16:50,976 --> 00:16:52,596 I can do that through copilot. 342 00:16:53,096 --> 00:16:57,286 I can do that, I have a subscription to OpenAI and to Claude. 343 00:16:57,786 --> 00:17:01,716 And so I can pull in all three of those subscriptions into OpenCode 344 00:17:02,216 --> 00:17:06,266 and switch between the models at will in the middle of conversations. 345 00:17:06,766 --> 00:17:09,816 And so a lot of times when I run into like, if I run into my Claude 346 00:17:09,886 --> 00:17:13,106 Code, if I run into my Claude limit for the day on Opus, I just 347 00:17:13,106 --> 00:17:14,336 shift to my other subscription, 348 00:17:14,416 --> 00:17:14,876 shift to a 349 00:17:14,876 --> 00:17:15,456 different model. 350 00:17:15,456 --> 00:17:15,696 Yep. 351 00:17:15,896 --> 00:17:16,186 Yeah. 352 00:17:16,186 --> 00:17:19,026 And I just, so I go from my cloud subscription to my GitHub's 353 00:17:19,106 --> 00:17:21,866 Copilot subscription and just keep going, keep rolling with it. 354 00:17:22,176 --> 00:17:25,716 I think there's technically token limit differences because if I, if you 355 00:17:25,716 --> 00:17:30,146 get really into the nerdy details of the APIs, like, Claude is, you know, 356 00:17:30,256 --> 00:17:36,336 Anthropic has moved to a newer, modern, API away from the chat API before. 357 00:17:36,346 --> 00:17:38,786 And I think, yeah, and I think that, I think GitHub is 358 00:17:38,786 --> 00:17:40,646 still on the older structure. 359 00:17:40,646 --> 00:17:45,476 So there, but OpenCode as an, is a wonderful open source project because 360 00:17:45,486 --> 00:17:47,126 it's, it just handles all of that. 361 00:17:47,146 --> 00:17:50,576 And in fact, if we did a demo, I would show you like, if you scroll, 362 00:17:50,576 --> 00:17:51,936 I was showing some friends yesterday. 363 00:17:52,406 --> 00:17:56,106 Because they're all Claude Code users, and I've been like the sole OpenCode user 364 00:17:56,166 --> 00:17:57,416 and I'm ranting about it all the time. 365 00:17:57,656 --> 00:18:00,946 I'm like, did you know you can have one conversation and be, and watch 366 00:18:00,946 --> 00:18:05,366 it from the TUI, from the GUI and from remote web mobile on my phone. 367 00:18:05,716 --> 00:18:08,366 And I can see the conversation on all three happening at the same time. 368 00:18:08,376 --> 00:18:09,856 And I don't have to use WhatsApp. 369 00:18:10,136 --> 00:18:13,856 I don't have to, you know, not to knock on your door, but like, I don't, I can 370 00:18:13,866 --> 00:18:17,821 have the full fledged experience of the two of the, the, of the UI that 371 00:18:17,821 --> 00:18:21,691 we're all used to, with the proper diff formation and all that stuff, as 372 00:18:21,691 --> 00:18:26,671 opposed to chatting at it through an instant message tool that is not meant 373 00:18:26,671 --> 00:18:29,941 for agentic screen, screen presentation. 374 00:18:30,441 --> 00:18:33,451 And I much prefer that I have to use, with OpenCode, it's not 375 00:18:33,451 --> 00:18:34,571 a, it's not a cloud service. 376 00:18:34,916 --> 00:18:37,256 Like Claude or GPT or anything like that. 377 00:18:37,256 --> 00:18:40,406 So you have to use like tailscale or something to get you back to your 378 00:18:40,406 --> 00:18:42,916 machine, but I run it as a server. 379 00:18:42,916 --> 00:18:45,276 In fact, I showed a diagram of running it as a server. 380 00:18:45,776 --> 00:18:49,066 So you just, you know, run OpenCode server or OpenCode web, and it 381 00:18:49,096 --> 00:18:50,756 launches as a background process. 382 00:18:50,766 --> 00:18:53,426 And then you can connect from any of the three different ways to 383 00:18:53,426 --> 00:18:55,586 connect web, that client or TUI. 384 00:18:56,086 --> 00:18:56,626 And it works. 385 00:18:56,696 --> 00:18:59,347 I am, I'm going to have to start like if they were an actual product 386 00:18:59,357 --> 00:19:02,877 company, I would like start looking for a sponsorship because I talk about 387 00:19:02,877 --> 00:19:04,867 them so much as, as my favorite tool. 388 00:19:05,207 --> 00:19:10,577 But other than that, I'm also very much into Copilot and VS, and the Copilot 389 00:19:10,577 --> 00:19:14,357 world because I'm finding that, you know, people that, companies that are all in 390 00:19:14,357 --> 00:19:16,247 on GitHub tend to use Copilot first. 391 00:19:16,577 --> 00:19:21,427 And when I talk to what I would call normie developers, not like bleeding edge, 392 00:19:21,457 --> 00:19:25,037 always on the, you know, the people that haven't yet put OpenClaw in their closet. 393 00:19:25,537 --> 00:19:30,397 Exactly, they tend to be in their traditional IDE world, whether that's VS 394 00:19:30,437 --> 00:19:36,027 Code or, you know, PHP Storm or, you know, whatever IDE they're used to, and they're 395 00:19:36,027 --> 00:19:37,697 just using whatever came with that. 396 00:19:37,697 --> 00:19:41,337 So with VS Code, the default is, is Copilot, even though you 397 00:19:41,337 --> 00:19:42,897 can put all the other ones in. 398 00:19:43,007 --> 00:19:44,397 OpenCode has a plug, whatever model you want. 399 00:19:44,397 --> 00:19:44,637 Yep. 400 00:19:44,667 --> 00:19:44,877 Yeah. 401 00:19:44,877 --> 00:19:47,417 Claude has a plugin, I should say for the audience. 402 00:19:47,417 --> 00:19:48,767 Like, we haven't talked about this on the show, but it's 403 00:19:48,767 --> 00:19:50,237 even getting more complex now. 404 00:19:50,237 --> 00:19:53,857 and I had to draw, I've had to draw this out a couple times on the stream, because 405 00:19:53,857 --> 00:20:02,377 now with your GitHub Copilot subscription, you can now use Claude Code or the Claude 406 00:20:02,877 --> 00:20:11,816 app inside a GitHub or inside a VS Code as your agent of choice inside the Copilot 407 00:20:11,837 --> 00:20:16,887 harness and which basically means you don't need a Claude Code subscription 408 00:20:16,887 --> 00:20:20,240 to use, if you want to learn more 409 00:20:20,240 --> 00:20:21,539 about any of 410 00:20:21,539 --> 00:20:22,839 the official cloud 411 00:20:22,839 --> 00:20:28,902 tooling, which is great because I mean, I think this abstraction from the actual, 412 00:20:28,902 --> 00:20:32,982 or marrying yourself to a specific company is kind of a thing of the past. 413 00:20:33,252 --> 00:20:35,752 You're just going to go for the best model eventually, right. 414 00:20:35,812 --> 00:20:36,422 Whatever. 415 00:20:36,922 --> 00:20:40,722 It might be Cloud does Code and OpenAI does something else and whatever. 416 00:20:40,722 --> 00:20:44,772 And you just have it set up where, Oh, I have a use case and they 417 00:20:44,772 --> 00:20:47,912 just, my swarm of agents go out to the right model and get it done. 418 00:20:48,412 --> 00:20:51,232 it does feel weird to commit to any one company. 419 00:20:51,232 --> 00:20:55,312 I mean, I feel like you're pretty safe if you committed to Anthropic right now, but. 420 00:20:55,752 --> 00:20:56,412 I think so. 421 00:20:56,482 --> 00:20:56,742 Yeah. 422 00:20:56,992 --> 00:21:00,702 But I, I believe, what other, it doesn't matter which company you go with. 423 00:21:00,942 --> 00:21:03,332 I really believe your local setup is super important. 424 00:21:04,042 --> 00:21:08,322 I mean, I'm harping on this as like your human operating system or your Bret 425 00:21:08,352 --> 00:21:12,502 operating system where you have to set up like all your agents, you have to set up 426 00:21:12,512 --> 00:21:18,622 all your template files and memory files locally and really use that as your basis. 427 00:21:18,872 --> 00:21:21,532 And it doesn't matter what model you used and cause it always gonna, 428 00:21:21,592 --> 00:21:23,132 it should always read that locally. 429 00:21:24,772 --> 00:21:29,152 I use Gemini with OpenClaw and when I had the same setup for all the models 430 00:21:29,512 --> 00:21:31,362 and Gemini was the most apologetic. 431 00:21:31,442 --> 00:21:33,732 Every time I do something like, Oh, I'm so sorry. 432 00:21:34,002 --> 00:21:35,672 It didn't work properly. 433 00:21:35,682 --> 00:21:37,022 And I would be like, it's just so frustrating. 434 00:21:37,072 --> 00:21:38,862 Like stop apologizing, just fix it. 435 00:21:39,362 --> 00:21:39,802 Yeah. 436 00:21:39,832 --> 00:21:41,082 That's the thing right now too. 437 00:21:41,082 --> 00:21:43,762 Like not every developer gets to choose their models. 438 00:21:43,772 --> 00:21:46,132 This is something Laura Tacho was on the show last year and was 439 00:21:46,132 --> 00:21:49,417 talking about like, one of the biggest barriers for developers. 440 00:21:49,917 --> 00:21:53,487 In enterprise is the fact that the enterprise doesn't necessarily hand 441 00:21:53,487 --> 00:21:56,887 out licenses to everyone, and if it does, it might be very prescriptive to 442 00:21:56,887 --> 00:21:58,567 which models or harnesses they can use. 443 00:21:59,042 --> 00:22:02,832 So people are kind of just stuck using whatever their company decided 444 00:22:02,832 --> 00:22:07,322 on a year ago, which was like, that's like the equivalent of making a 445 00:22:07,322 --> 00:22:08,872 decision and keeping it for a decade. 446 00:22:08,882 --> 00:22:12,322 Like the challenge right now is if you make a commitment as an organization, 447 00:22:12,822 --> 00:22:16,922 that's a really, like, you could essentially be at this point, the 448 00:22:16,922 --> 00:22:21,202 equivalent of years behind an AI because you've aligned yourself. 449 00:22:21,702 --> 00:22:24,532 And that's something that's not like IT purchasing isn't used to that, 450 00:22:24,532 --> 00:22:27,532 like lawyers and contracts and all these things aren't used to that. 451 00:22:28,342 --> 00:22:32,382 So I sympathize with people like James, because you're having to use Gemini, 452 00:22:32,882 --> 00:22:37,362 everyone I know, like pretty much all of my bleeding edge friends that are 453 00:22:37,362 --> 00:22:39,922 living in AI all day long as developers. 454 00:22:40,252 --> 00:22:43,542 And these are people that get to choose whatever tool they want to use, 455 00:22:43,542 --> 00:22:47,692 because they're all small businesses or subcontractors, or, you know, they, there 456 00:22:47,722 --> 00:22:49,752 are, there are product people that are making their own products or whatever. 457 00:22:50,252 --> 00:22:55,367 I'd say, yeah, 99 percent of them are on 6. 458 00:22:55,647 --> 00:23:02,497 The other 1 percent are on GPT 5.4 because 5.4 specifically, 459 00:23:02,497 --> 00:23:07,017 yeah, specifically 5.4 is showing really good signs of back end code. 460 00:23:07,257 --> 00:23:09,077 It's evidently a horrible designer. 461 00:23:09,557 --> 00:23:12,147 So I'm like, if you had to ask me this week, and it will 462 00:23:12,147 --> 00:23:13,347 probably change next month. 463 00:23:13,347 --> 00:23:18,457 Opus 4.6 is go to for the least number of hallucinations and the longest run 464 00:23:18,457 --> 00:23:24,417 cycles before I have to like, Ralph loop it through or, or if you know about Ralph 465 00:23:24,417 --> 00:23:29,487 loops, but like before clear context, like, so like Opus is the jam, like 466 00:23:29,797 --> 00:23:34,247 Opus, Opus, Opus, maybe Opus for planning and Sonnet for actually doing the work. 467 00:23:34,247 --> 00:23:38,127 If you want to save tokens, GPT 5.4, if it's doing backend stuff and not 468 00:23:38,127 --> 00:23:39,687 frontend, nothing design related. 469 00:23:40,097 --> 00:23:43,867 And then I don't know where Gemini falls in there other 470 00:23:43,867 --> 00:23:45,147 than like Nano Banana for me. 471 00:23:45,147 --> 00:23:48,257 Like I, I make images and I don't use Gemini for it. 472 00:23:48,757 --> 00:23:53,207 So the problem is, is like developers, like this is like, this is the part that 473 00:23:53,207 --> 00:23:56,767 I hate, like a year ago, I didn't want to talk about any of this because well, 474 00:23:56,797 --> 00:23:59,307 all the models were kind of, I think I feel like a year ago, all the models 475 00:23:59,307 --> 00:24:03,627 were kind of, eh, we were, I think we were doing like Sonnet 3.5 at the time. 476 00:24:03,627 --> 00:24:03,807 Right. 477 00:24:03,847 --> 00:24:08,137 3.7. And so the problem is like people are stuck on a particular, what I would 478 00:24:08,137 --> 00:24:13,187 call, Legacy model, just because it's only like six or eight or 10 months old. 479 00:24:13,877 --> 00:24:13,997 Or 480 00:24:13,997 --> 00:24:15,387 not even, right. 481 00:24:15,697 --> 00:24:20,717 And that model is, I don't know, three times, not like three 482 00:24:20,717 --> 00:24:22,707 times worse than Opus maybe. 483 00:24:22,957 --> 00:24:26,727 So that's the real problem right now is I feel like we've got first class citizen, 484 00:24:26,737 --> 00:24:31,687 like we've got people that are loving life on Opus and then we've got people 485 00:24:31,687 --> 00:24:36,577 struggling on older models or older versions or approved models or whatever. 486 00:24:37,077 --> 00:24:38,807 And they're like, everything's garbage. 487 00:24:38,807 --> 00:24:40,647 And I'm like, no, actually it's not. 488 00:24:40,697 --> 00:24:44,667 I mean, there's plenty of garbage to be found, but it is so much 489 00:24:44,677 --> 00:24:46,587 better if you just try this. 490 00:24:47,087 --> 00:24:48,417 And not everybody can do it. 491 00:24:48,487 --> 00:24:49,437 Not everybody gets to have that. 492 00:24:49,437 --> 00:24:49,737 I think 493 00:24:49,847 --> 00:24:50,037 so. 494 00:24:50,392 --> 00:24:52,212 I think all the models are good. 495 00:24:52,262 --> 00:24:55,512 They just require, for example, Gemini, I used it for a long time. 496 00:24:55,972 --> 00:24:58,802 And I just think you have to set up the prompting much differently 497 00:24:58,802 --> 00:24:59,992 compared to the other models. 498 00:25:00,302 --> 00:25:02,582 Once you get it kind of set up, it works. 499 00:25:02,802 --> 00:25:05,532 I, I still believe, like you said, Claude is the best designer. 500 00:25:06,032 --> 00:25:08,292 in Claude Desktop, they had this design skill. 501 00:25:08,792 --> 00:25:10,532 If you're using Claude, just enable that. 502 00:25:10,582 --> 00:25:13,752 It does amazing websites or graphics. 503 00:25:13,802 --> 00:25:14,732 It really nice. 504 00:25:14,772 --> 00:25:15,632 And it's right out of the box. 505 00:25:15,632 --> 00:25:16,382 You don't have to do anything. 506 00:25:16,382 --> 00:25:19,242 Just enable the graphics skill or the design skill. 507 00:25:19,742 --> 00:25:23,352 Right, that's maybe something to talk about too, real quick is like the, 508 00:25:23,452 --> 00:25:28,367 all of these things require effort in terms of getting your agents.md 509 00:25:28,907 --> 00:25:31,507 file for the project really honed in. 510 00:25:31,537 --> 00:25:34,297 Although there's just a report that came out a couple of weeks ago, 511 00:25:34,297 --> 00:25:39,607 that's suggesting that agent files actually reduce the quality of the 512 00:25:39,607 --> 00:25:41,607 finished product in every test. 513 00:25:42,077 --> 00:25:44,467 I don't know if you saw that, but that was like, I didn't even 514 00:25:44,467 --> 00:25:45,207 know what to think about that. 515 00:25:45,217 --> 00:25:50,347 Like, but skills to me have been a game changer in the last couple of months, 516 00:25:50,377 --> 00:25:54,547 and if you've not, someone who's adopted, adapted to skills, I'm pretty sure all the 517 00:25:54,547 --> 00:25:57,097 harnesses now have got a way to use them. 518 00:25:57,097 --> 00:26:00,887 The paths might be slightly different depending on the harness, but creating 519 00:26:00,887 --> 00:26:05,057 the skill markdown files or finding good ones on the web that I have read 520 00:26:05,257 --> 00:26:09,807 and I am assured do not have any sort of prompt injections, that's key, but 521 00:26:09,807 --> 00:26:14,030 I feel like that has changed the game for me in allowing me to give the 522 00:26:14,030 --> 00:26:16,507 harness work without a lot of rigor. 523 00:26:17,007 --> 00:26:18,877 And it not losing its mind. 524 00:26:18,937 --> 00:26:23,147 And I think that's the key distinction between like now and six months ago. 525 00:26:23,407 --> 00:26:29,127 Is six months ago, if I didn't give it a very long prompt, very clear, and we were 526 00:26:29,127 --> 00:26:33,117 all like pasting and storing prompts, and then we were, and then we had commands 527 00:26:33,117 --> 00:26:35,247 and we were trying to store them in commands because we didn't want to have 528 00:26:35,247 --> 00:26:36,627 to type the whole prompt or paste it. 529 00:26:37,067 --> 00:26:39,247 And now I'm not doing any of that. 530 00:26:39,277 --> 00:26:42,057 I'm only doing skills and I don't know about you, but I'm finding 531 00:26:42,057 --> 00:26:43,707 that that's all I really need. 532 00:26:43,967 --> 00:26:46,777 I just need to write the skill and I keep having the agent update. 533 00:26:46,817 --> 00:26:50,367 Whenever the, it's almost reminds me of like the rule in ops is when you 534 00:26:50,377 --> 00:26:54,447 have an, When you have a production failure, the rule in Ops is you create 535 00:26:54,447 --> 00:27:00,952 a test, so that somewhere in your test pipelines, that, that failure was tested 536 00:27:01,012 --> 00:27:02,202 to make sure it won't happen again. 537 00:27:02,232 --> 00:27:04,512 Like the goal is never the same failure twice. 538 00:27:04,892 --> 00:27:08,612 I feel like with skills and AIs now, if I, like I just happened yesterday, 539 00:27:08,972 --> 00:27:11,432 I was having a walk, it was actually running a SOP for me, a standard 540 00:27:11,432 --> 00:27:14,722 operating procedure of just something in my courses that I would normally 541 00:27:14,722 --> 00:27:16,652 have like a virtual assistant do for me. 542 00:27:16,982 --> 00:27:19,542 And it's going through processing this text and managing some, creating 543 00:27:19,542 --> 00:27:22,822 some HTML and things for me that I need for each lecture in a video. 544 00:27:23,322 --> 00:27:27,342 And it made a couple of mistakes, put some things out of order and I realized, 545 00:27:27,382 --> 00:27:28,662 Oh, the skill needs to be updated. 546 00:27:28,662 --> 00:27:31,832 So I just ask it to update the skill to please fix this and then also update 547 00:27:31,832 --> 00:27:33,022 the skill so it doesn't happen again. 548 00:27:33,312 --> 00:27:36,272 And I feel like that's my new operator and failure mode is if I see the AI 549 00:27:36,322 --> 00:27:40,292 misbehave, it's probably a lack of it not having the proper instruction. 550 00:27:40,332 --> 00:27:42,262 I just need to update the skill and have it retry. 551 00:27:42,262 --> 00:27:43,322 And then usually corrects itself. 552 00:27:43,682 --> 00:27:44,892 Is that something that you're seeing? 553 00:27:44,902 --> 00:27:45,852 Are you, is that 554 00:27:47,247 --> 00:27:50,337 And it's back to the context question again, it's really, you know, sometimes 555 00:27:50,337 --> 00:27:52,167 it forgets how to use the skill. 556 00:27:52,577 --> 00:27:56,457 So maybe you have to declare the skill better, so it 557 00:27:56,467 --> 00:27:57,817 understands the skill better. 558 00:27:58,317 --> 00:28:01,197 But I mean, the skill works, but sometimes I have to go again and say, 559 00:28:01,197 --> 00:28:04,807 Hey, don't you remember use the skill lap time like this, this and it worked. 560 00:28:04,817 --> 00:28:08,057 And then I realized maybe I have to add more context to it, so it 561 00:28:08,067 --> 00:28:13,657 realizes if it starts a new if chat with this skill, it should be 562 00:28:13,657 --> 00:28:15,597 able to use it right off the bat. 563 00:28:16,097 --> 00:28:18,797 And I mean, I built all sorts of skills. 564 00:28:18,867 --> 00:28:23,077 It's really, I find that's, really the game changer at the moment is skills. 565 00:28:23,197 --> 00:28:26,047 Like you said, it's where I'm investing all my time at the moment. 566 00:28:26,547 --> 00:28:28,167 How do you manage your skills? 567 00:28:28,247 --> 00:28:32,807 Like, do you, I've been looking into the NPX skills tool. 568 00:28:32,807 --> 00:28:34,167 I think it's, I think for cell made it. 569 00:28:34,607 --> 00:28:35,677 I'm not an expert in it yet. 570 00:28:35,677 --> 00:28:38,187 I've really only had like a week's use of it, but it's helped me. 571 00:28:38,487 --> 00:28:38,877 Okay. 572 00:28:39,287 --> 00:28:43,037 It's, I mean, I was making them all manually before then, and then or like 573 00:28:43,037 --> 00:28:46,397 downloading, copy and pasting from GitHub repos, you know, finding ones I liked, 574 00:28:46,697 --> 00:28:51,097 and I'd store them in my global path, which I think is like in my user directory 575 00:28:51,307 --> 00:28:53,957 .claude/ slash skills and then skill name. 576 00:28:54,337 --> 00:28:54,797 and. 577 00:28:55,297 --> 00:28:57,427 I realized suddenly that I was spending a lot of time copying 578 00:28:57,427 --> 00:28:58,907 and pasting in between projects. 579 00:28:59,397 --> 00:29:03,127 And I haven't really figured out, like, I don't know if I can actually 580 00:29:03,127 --> 00:29:06,947 use the MPX skills tool to like store my own and like have them 581 00:29:06,947 --> 00:29:08,257 centrally so that I can access them. 582 00:29:08,257 --> 00:29:09,337 I guess I could probably do that. 583 00:29:09,587 --> 00:29:11,857 Do you have, are you, do you have any hot tips for anyone? 584 00:29:11,967 --> 00:29:12,967 I'm kind of putting you on the spot. 585 00:29:13,117 --> 00:29:13,537 So, 586 00:29:14,037 --> 00:29:17,347 And, I mean, what I do, obviously I store each skill as this 587 00:29:17,387 --> 00:29:18,547 individual repo in GitHub. 588 00:29:18,707 --> 00:29:22,717 So it all goes out to GitHub and then I version control it. 589 00:29:22,757 --> 00:29:25,617 And I say, anytime you touch that, you better sync it to GitHub. 590 00:29:25,837 --> 00:29:27,917 So then I always have like a backup copy. 591 00:29:28,387 --> 00:29:31,007 And then it's running tests on this and everything else. 592 00:29:31,017 --> 00:29:36,287 So that's, that's one unlock I did because obviously, early on, my OpenClaw 593 00:29:36,797 --> 00:29:38,447 deleted itself and I lost everything. 594 00:29:38,947 --> 00:29:42,287 And so since then I've been backing everything up to get up. 595 00:29:42,287 --> 00:29:46,157 So, but that's like kind of a, like a, kind of a workaround to 596 00:29:46,167 --> 00:29:51,277 have skills ready, and then I'm also defining a skill per agent. 597 00:29:51,287 --> 00:29:52,537 So I have like a DevOps agent. 598 00:29:52,772 --> 00:29:56,202 And the DevOps agent's responsibility is just only DevOps, and it has 599 00:29:56,202 --> 00:29:57,642 only DevOps skills assigned to it. 600 00:29:58,092 --> 00:30:00,232 So then it has context with just these skills. 601 00:30:00,242 --> 00:30:02,352 It doesn't need context for every skill I have. 602 00:30:02,852 --> 00:30:07,822 And I think since I changed it to specific skills per agent, it's gotten 603 00:30:07,882 --> 00:30:09,692 more clever on how to use the skills. 604 00:30:10,192 --> 00:30:13,772 Now, when you create an agent, in that sense, are you talking specifically 605 00:30:13,772 --> 00:30:16,942 of something like Claude Code or like what, you say you create a DevOps 606 00:30:16,972 --> 00:30:18,032 agent, where are you creating that? 607 00:30:18,532 --> 00:30:23,912 So, I mean, in the context of OpenClaw or I would say open Claude Code as 608 00:30:23,912 --> 00:30:27,832 well, you can create basically folders and each folder is its own project. 609 00:30:27,882 --> 00:30:32,052 And within the project, you can have skills and all these things with insight. 610 00:30:32,552 --> 00:30:36,922 And you define each folder as its own agent, like its own personality, 611 00:30:36,922 --> 00:30:41,692 what it's capable of its own, claude.Md file, essentially. 612 00:30:42,067 --> 00:30:42,327 Okay. 613 00:30:42,337 --> 00:30:43,577 Or agent.md file. 614 00:30:43,967 --> 00:30:47,087 And that's how you kind of version control it and have the separation. 615 00:30:47,587 --> 00:30:49,697 And then you can kick it off separately. 616 00:30:50,197 --> 00:30:52,637 So, we're going to jump into OpenClaw cause there's all the 617 00:30:52,637 --> 00:30:54,577 insanity of what people are doing. 618 00:30:55,077 --> 00:30:59,337 But behind that, and I've never actually even installed it yet. 619 00:30:59,357 --> 00:31:02,487 I'm just, it's been so busy and it's not even that old. 620 00:31:02,497 --> 00:31:05,347 Like I, it's, I feel like we've lived 10 years in the last two months. 621 00:31:05,347 --> 00:31:07,857 You're a person that operates in a high safety environment, high secure 622 00:31:07,857 --> 00:31:11,707 environment, obviously you're not using this for work, I assume, but, 623 00:31:11,987 --> 00:31:13,297 Strictly testing at the moment. 624 00:31:13,597 --> 00:31:18,337 Right, but OpenClaw isn't necessarily, like, can you talk to a little bit about 625 00:31:18,837 --> 00:31:24,267 OpenClaw isn't necessarily on day one, the minute you install it, some rogue, crazy 626 00:31:24,267 --> 00:31:25,807 thing that's going to ruin your life. 627 00:31:25,867 --> 00:31:29,947 Like it, no, like the things that we're hearing in the headlines are people 628 00:31:29,947 --> 00:31:33,577 deliberately doing things with it that they probably shouldn't have done. 629 00:31:33,797 --> 00:31:34,807 And that's not, it's. 630 00:31:35,307 --> 00:31:36,737 That's not its modus operandi. 631 00:31:36,757 --> 00:31:40,447 Like it's not, that's not what it's trying to be necessarily. 632 00:31:40,847 --> 00:31:44,807 I always look at it as like an orchestrator of, if it's orchestrating, 633 00:31:45,137 --> 00:31:48,307 it comes in and it decides, you decide what you want it to orchestrate. 634 00:31:48,307 --> 00:31:50,077 You decide how much you want to give it of your life. 635 00:31:50,097 --> 00:31:53,877 You decide what things it has access to and you put the guardrails in and all. 636 00:31:53,927 --> 00:31:54,657 Is that all true? 637 00:31:54,687 --> 00:31:56,462 I'm actually just assuming that that's all true. 638 00:31:56,962 --> 00:32:00,452 You have to back up some, and this is where your DevOps specialty comes in, 639 00:32:00,462 --> 00:32:04,662 Bret, is like, most people probably don't have the infrastructure knowledge 640 00:32:04,702 --> 00:32:06,152 on how to set up a server generally. 641 00:32:06,417 --> 00:32:06,917 Right? 642 00:32:07,387 --> 00:32:11,007 So they're setting up their home PC or whatever server they want, 643 00:32:11,027 --> 00:32:12,487 and they're putting OpenClaw on it. 644 00:32:12,957 --> 00:32:17,167 And they have maybe not the comprehension of how to secure the 645 00:32:17,167 --> 00:32:19,227 server, regardless of OpenClaw. 646 00:32:19,277 --> 00:32:22,417 I'm just saying, hey, putting a server on the internet with a 647 00:32:22,417 --> 00:32:26,667 website, maybe is not a good idea if you have everything open, right? 648 00:32:27,167 --> 00:32:31,007 And that's, I think what really burned a lot of people with OpenClaw 649 00:32:31,007 --> 00:32:33,887 cause they're, you know, putting it out there, they didn't lock 650 00:32:33,887 --> 00:32:35,707 it down, all the ports were open. 651 00:32:35,767 --> 00:32:39,357 And then the first few versions of OpenClaw, the API keys were all 652 00:32:39,357 --> 00:32:41,437 just plain text stored locally. 653 00:32:41,527 --> 00:32:43,827 You know, it was like version 0. 654 00:32:43,827 --> 00:32:47,087 1 and I mean, it's still early days for this product. 655 00:32:47,087 --> 00:32:50,557 So I mean, if you're putting your server out there password 656 00:32:50,567 --> 00:32:52,457 123 it's gonna get hacked. 657 00:32:52,507 --> 00:32:53,417 Somebody's gonna get in it. 658 00:32:53,567 --> 00:32:54,687 Somebody's gonna take advantage of it. 659 00:32:55,187 --> 00:32:58,087 But is the server going rogue by itself? 660 00:32:58,267 --> 00:32:58,627 No. 661 00:32:59,127 --> 00:33:02,507 I mean, unless you tell it to, it won't do that. 662 00:33:02,747 --> 00:33:04,767 And that's, it's that's the thing. 663 00:33:04,767 --> 00:33:06,477 I mean, it's whatever you're telling it to do. 664 00:33:06,507 --> 00:33:10,357 And I've set up, I don't know, maybe 10 OpenClaws now. 665 00:33:10,597 --> 00:33:13,277 So I've set them up multiple times trying to understand the process and 666 00:33:13,277 --> 00:33:16,097 like, kind of, really by the content. 667 00:33:17,007 --> 00:33:17,647 Oh, right. 668 00:33:17,817 --> 00:33:22,547 And, no, it's really, you know, you just have to have the basics down. 669 00:33:22,547 --> 00:33:26,057 Locking down the server o, installing OpenClaw and figuring out like, Hey, 670 00:33:26,057 --> 00:33:31,297 if I give OpenClaw an API key to Claude or OpenAI or whatever, look in the 671 00:33:31,297 --> 00:33:34,867 server to see where it's stored and just like figure out is this secure? 672 00:33:35,227 --> 00:33:37,747 And then I realized quite quickly, I'm like, that's not secure. 673 00:33:37,807 --> 00:33:42,377 I'd rather store all my keys in 1Password and then have a connection from the 674 00:33:42,497 --> 00:33:44,207 OpenClaw to 1Password to grab the keys. 675 00:33:44,597 --> 00:33:45,857 And that's how I'm doing it like this. 676 00:33:45,857 --> 00:33:49,817 So it's a lot of security basics that are burning people. 677 00:33:50,317 --> 00:33:55,627 Now, it is possible that, yeah, you just update it incorrectly and 678 00:33:55,627 --> 00:33:57,037 you open more ports by accident. 679 00:33:57,137 --> 00:33:59,047 But you just have to constantly monitor this stuff. 680 00:33:59,547 --> 00:34:02,337 And I built like a security skill for OpenClaw. 681 00:34:02,837 --> 00:34:04,747 Obviously it doesn't pass any of the tests. 682 00:34:04,777 --> 00:34:09,227 I'm like, I built it just for myself and I put it in the OpenClaw Hub. 683 00:34:09,727 --> 00:34:13,067 And essentially it just checks, do you have your firewall turned on? 684 00:34:13,257 --> 00:34:14,657 Do you have a VPN? 685 00:34:14,797 --> 00:34:16,607 Are all the ports, you know, locked? 686 00:34:16,667 --> 00:34:18,007 It's basic IT stuff. 687 00:34:18,257 --> 00:34:18,607 Okay. 688 00:34:18,677 --> 00:34:22,767 And that's what's really hurting a lot of people and they don't realize, 689 00:34:22,817 --> 00:34:25,657 and I just saw on Hacker News that they're charging, somebody's charging 690 00:34:25,667 --> 00:34:29,867 $5,000 to install OpenClaw servers for people in San Francisco at the moment. 691 00:34:30,367 --> 00:34:31,217 That's awesome. 692 00:34:31,567 --> 00:34:32,277 Good for them. 693 00:34:32,277 --> 00:34:32,767 Good for them. 694 00:34:33,107 --> 00:34:35,297 It's one curl command and you have it installed. 695 00:34:35,307 --> 00:34:35,607 Right. 696 00:34:35,617 --> 00:34:35,887 Come on. 697 00:34:36,227 --> 00:34:38,107 Right, but you need a VPC somewhere. 698 00:34:38,107 --> 00:34:39,387 You need, you know, yeah, you need a droplet. 699 00:34:39,387 --> 00:34:44,267 I mean, I think when first, when it sort of reached its first sort of peak, not 700 00:34:44,267 --> 00:34:49,132 really peak hype, but like when it first got hyped up, it was the first time 701 00:34:49,132 --> 00:34:52,622 I'd ever seen an open source project where basically all the cloud providers 702 00:34:52,952 --> 00:34:56,332 within a week of each other all put out notifications saying, Hey, we have a 703 00:34:56,332 --> 00:35:00,532 way for you to run this one app on our platform, like a one click install for 704 00:35:00,532 --> 00:35:02,632 this thing, and I had never seen that. 705 00:35:02,662 --> 00:35:08,612 I've never seen like major top tier cloud companies going, Oh, suddenly 706 00:35:08,712 --> 00:35:13,112 we care so much about this brand new month old piece of open source that 707 00:35:13,112 --> 00:35:17,812 we're going to make it a dedicated product for you to run on our platform. 708 00:35:17,842 --> 00:35:19,212 And that's just never really happened. 709 00:35:19,212 --> 00:35:20,122 I've never seen that. 710 00:35:20,392 --> 00:35:23,202 It's, you know, normally it takes five, 10 years for a mature product 711 00:35:23,212 --> 00:35:27,752 to be something that, you know, Google Cloud or Azure care about, right? 712 00:35:27,762 --> 00:35:29,462 Like they just, they're not going to hype it up. 713 00:35:29,752 --> 00:35:33,372 But it's reached such hype cycle, and I think there's so many players 714 00:35:33,372 --> 00:35:37,232 now all competing for the, you know, agent, are, you know, the 715 00:35:37,232 --> 00:35:38,672 place where you run your agents. 716 00:35:38,922 --> 00:35:41,152 We've got a ton of little startups that have been around for a couple of 717 00:35:41,152 --> 00:35:42,892 years, all trying to play in that space. 718 00:35:43,212 --> 00:35:46,982 Daytona, who's been on this show, has pivoted to be in that space as sort 719 00:35:46,982 --> 00:35:48,542 of the place you run your agents. 720 00:35:48,542 --> 00:35:52,792 And you really just, it's really just abstracting what you and I would do, which 721 00:35:52,792 --> 00:35:57,167 was create a dedicated server and lock it down, and you know, make sure that you 722 00:35:57,167 --> 00:36:00,297 got it on patch rotation, make sure that the ports are exposed and that if you're 723 00:36:00,297 --> 00:36:03,657 going to put SSH out there, put it in a high random port, you know, make sure 724 00:36:03,657 --> 00:36:07,857 it's, installed tailscale, these types of things, doing all the things that we do. 725 00:36:08,357 --> 00:36:10,877 And yeah, OpenClaw out of the box should probably have like a built 726 00:36:10,877 --> 00:36:14,012 in lockdown skill, where it's just going to automatically do 727 00:36:14,012 --> 00:36:18,032 that as a part of its install to battle, battle, prepare for battle, 728 00:36:18,032 --> 00:36:19,352 essentially on the internet, you know. 729 00:36:19,462 --> 00:36:21,062 That's a great point you brought up, right? 730 00:36:21,092 --> 00:36:23,422 I mean, the picture behind you, like everything's on fire. 731 00:36:23,462 --> 00:36:25,952 That's actually my OpenClaw server a couple of times already. 732 00:36:26,402 --> 00:36:31,022 I mean, it's a, so, I mean, I set everything up and, when you set it up as 733 00:36:31,022 --> 00:36:33,132 root, root, that's obviously bad advice. 734 00:36:33,132 --> 00:36:34,142 You should never do that. 735 00:36:34,542 --> 00:36:35,822 So it's running as root. 736 00:36:35,892 --> 00:36:38,822 And I got really frustrated in Telegram and I'm writing something 737 00:36:38,832 --> 00:36:41,882 and I said, just delete it, delete, you know, whatever it is. 738 00:36:41,932 --> 00:36:46,522 Well, OpenClaw took it literally and deleted the entire project and down to 739 00:36:46,522 --> 00:36:48,612 the root and just deleted the server. 740 00:36:48,642 --> 00:36:49,862 No longer responsive. 741 00:36:50,362 --> 00:36:51,152 I was like, wow. 742 00:36:51,242 --> 00:36:52,222 Okay, that worked. 743 00:36:52,722 --> 00:36:53,812 So it did an RM. 744 00:36:53,812 --> 00:36:55,232 So lesson, lesson learned. 745 00:36:55,722 --> 00:36:57,162 Run it, don't run it as root. 746 00:36:58,992 --> 00:37:05,202 Where do you see, like for you, what are some anecdotal, like things that you 747 00:37:05,202 --> 00:37:07,162 think are reasonably safe to try with it? 748 00:37:07,192 --> 00:37:10,812 If someone's new and they're a little concerned, cause we had a, a great, not 749 00:37:10,812 --> 00:37:14,902 really a question, I don't understand the interest in OpenClaw as hallucinations 750 00:37:14,902 --> 00:37:18,492 are mathematically unsolvable and prompt injection attacks haven't been solved. 751 00:37:18,772 --> 00:37:20,872 It's just a gigantic attack surface. 752 00:37:21,282 --> 00:37:28,452 And while those are all technically true things, it's still just software, right? 753 00:37:28,512 --> 00:37:31,072 Like it's still something that's on my machine that I can 754 00:37:31,142 --> 00:37:32,732 lock down, that I can control. 755 00:37:32,932 --> 00:37:34,972 It's just software that's running in a loop. 756 00:37:35,362 --> 00:37:36,462 Access to an LLM. 757 00:37:36,482 --> 00:37:41,502 So it's not, these problems aren't unsolvable. 758 00:37:41,982 --> 00:37:45,642 And you can lock down the prompts, for example, OpenClaw has a 759 00:37:45,642 --> 00:37:49,812 community in Discord and actually OpenClaw is running in Discord. 760 00:37:49,822 --> 00:37:53,222 So you can actually, everyone's trying to prompt inject it the whole time. 761 00:37:53,222 --> 00:37:56,622 And you know, the founder, Peter is trying to encourage people to 762 00:37:56,632 --> 00:37:58,072 prompt, inject it and try to break it. 763 00:37:58,562 --> 00:37:58,812 Yeah. 764 00:37:58,812 --> 00:38:01,797 But I mean, by the way, that Discord server is insane. 765 00:38:01,937 --> 00:38:04,257 That Discord server is It's crazy, all this stuff going on. 766 00:38:04,257 --> 00:38:06,887 I wouldn't be surprised if it's now the biggest Discord server on the planet. 767 00:38:06,927 --> 00:38:07,727 Like it's insane. 768 00:38:08,227 --> 00:38:12,997 But I mean, what you do is you lock down your prompt to just your device. 769 00:38:13,327 --> 00:38:15,727 So you can actually pair your prompt to your device. 770 00:38:15,737 --> 00:38:19,847 So you say only my device, regardless what you get, is the only one you 771 00:38:19,847 --> 00:38:21,447 listen to, like actual commands. 772 00:38:21,947 --> 00:38:26,227 You can answer questions with everybody else, but you never process a command. 773 00:38:26,237 --> 00:38:28,047 So it's like only locked down to my device. 774 00:38:28,547 --> 00:38:28,907 Yeah. 775 00:38:29,407 --> 00:38:32,577 In response to the whole, as hallucinations are mathematically 776 00:38:32,577 --> 00:38:36,357 unsolvable, the way I'm looking at it now, especially that we have the newest 777 00:38:36,367 --> 00:38:40,926 models, Opus, I mentioned these earlier, Opus 4.6, Sonnet 4.6, pretty good. 778 00:38:40,956 --> 00:38:48,926 And then GPT 5.4, for me, my perspective has changed because once the AI 779 00:38:48,936 --> 00:38:54,156 gets, like we're beyond 80%, unlike the SWE bench boards and like a lot 780 00:38:54,156 --> 00:38:58,566 of these testing infrastructures, humans are not infallible like humans. 781 00:38:58,566 --> 00:39:01,796 Before we had the AI doing the work, humans were making mistakes 782 00:39:01,796 --> 00:39:04,736 too, and we would recover those mistakes and, and still to this 783 00:39:04,736 --> 00:39:06,146 day still still makes mistakes, 784 00:39:06,596 --> 00:39:10,736 So I don't need the AI to, I've sort of changed my thought process around 785 00:39:10,736 --> 00:39:16,276 I don't need the AI to be a hundred percent accurate at everything I ask it 786 00:39:16,276 --> 00:39:19,696 to do, and I don't think it will ever be a hundred percent accurate because 787 00:39:20,196 --> 00:39:24,926 we, I'm just, I've started to look at it more as like a junior engineer 788 00:39:25,226 --> 00:39:30,026 where when, if I hire someone and on day one, I asked them to do a task and 789 00:39:30,026 --> 00:39:34,556 the context that I'm giving them is a paragraph of information on their first 790 00:39:34,556 --> 00:39:38,616 day, they're more likely to make mistakes because they don't know the entire 791 00:39:38,826 --> 00:39:40,906 worldview of everything at the company. 792 00:39:41,206 --> 00:39:44,276 So they're going to make decisions and assumptions that aren't correct. 793 00:39:44,536 --> 00:39:46,596 And I would call those errors, right? 794 00:39:46,606 --> 00:39:48,106 That would be a mistake by a human. 795 00:39:48,606 --> 00:39:53,616 AI is exhibiting the same behaviors to me now, like where it's not necessarily 796 00:39:53,616 --> 00:39:56,496 going to hallucinate on every request. 797 00:39:56,686 --> 00:40:00,716 In fact, I don't actually know the last time I would qualify and 798 00:40:00,816 --> 00:40:02,386 that it technically hallucinated. 799 00:40:02,746 --> 00:40:04,716 I would say that it made mistakes. 800 00:40:04,816 --> 00:40:08,556 and, or it made decisions that I don't agree with, but I, it's not 801 00:40:08,556 --> 00:40:12,426 like, I don't remember the last time it incorrectly gave me HTML that 802 00:40:12,426 --> 00:40:14,056 wasn't actually a part of the spec. 803 00:40:14,406 --> 00:40:16,056 Like, I don't remember the last time it did that. 804 00:40:16,406 --> 00:40:18,766 So that's, but you have, 805 00:40:18,856 --> 00:40:20,546 but you have human in the loop, right? 806 00:40:20,546 --> 00:40:22,846 Human in the loop is part of the solution. 807 00:40:22,906 --> 00:40:23,746 You have to check it. 808 00:40:23,746 --> 00:40:24,756 You have to verify things. 809 00:40:25,106 --> 00:40:28,076 I would say that a year ago I was worried about, and, and I was the guy 810 00:40:28,076 --> 00:40:30,536 shot up the rooftop hallucinations. 811 00:40:30,806 --> 00:40:34,496 Now I don't remember the last time I cared because I have gone months 812 00:40:34,996 --> 00:40:40,046 with whenever it gives me something that I don't think was correct. 813 00:40:41,006 --> 00:40:41,276 Yep. 814 00:40:41,456 --> 00:40:45,446 I, 99% of the time believe that it's because I didn't give it enough 815 00:40:45,446 --> 00:40:48,906 information and it made a false decision or it made an incorrect 816 00:40:48,906 --> 00:40:50,476 decision with lack of information. 817 00:40:50,696 --> 00:40:55,711 So I actually look at it as, Oh, I didn't, I didn't realize it didn't know that. 818 00:40:56,211 --> 00:40:56,531 Exactly. 819 00:40:56,611 --> 00:40:58,071 And so I then correct it. 820 00:40:58,071 --> 00:40:58,571 It comes back 821 00:40:58,571 --> 00:40:59,331 to you, right? 822 00:40:59,341 --> 00:40:59,981 Cause I mean, 823 00:41:00,181 --> 00:41:00,451 yeah, 824 00:41:00,471 --> 00:41:02,491 you need to provide it more context. 825 00:41:02,851 --> 00:41:07,361 So I guess my point here is that like, yeah, like with enough safety rails, with 826 00:41:07,361 --> 00:41:11,132 enough testing, and that's why I'm so focused on agentic DevOps in my courses 827 00:41:11,132 --> 00:41:14,747 and training and whatnot, because I feel like that's the next realm of us 828 00:41:14,757 --> 00:41:20,297 really ramping up our own understanding, evaluations, understanding how do we 829 00:41:20,297 --> 00:41:23,927 test these things so that, and how do we make sure that humans are in the loop? 830 00:41:23,947 --> 00:41:30,417 How do we, you know, sure we can have an AI evaluate the PR of the first AI 831 00:41:30,417 --> 00:41:33,567 and a different AI evaluates the first one wrote it a different one, but that 832 00:41:33,587 --> 00:41:35,577 AI is, for the foreseeable future. 833 00:41:35,597 --> 00:41:39,397 I don't see the AI automatically approving the PR after it reviewed it. 834 00:41:39,637 --> 00:41:41,917 I still want to take, make a, take a look. 835 00:41:42,237 --> 00:41:46,937 If the PR is got a thousand lines of code that I have, or 5, 000 lines of code, 836 00:41:47,207 --> 00:41:50,387 I'm probably going to ask it to simplify that PR and break it into different 837 00:41:50,397 --> 00:41:52,007 PRs, like I'm going to reject that, just 838 00:41:52,087 --> 00:41:53,527 does it align with my roadmap? 839 00:41:53,557 --> 00:41:53,907 Right. 840 00:41:53,987 --> 00:41:55,537 Is that what I want in my product? 841 00:41:55,647 --> 00:41:55,807 Right. 842 00:41:55,827 --> 00:41:57,287 That's also something you'd say. 843 00:41:57,757 --> 00:41:58,587 So I don't, yeah. 844 00:41:58,587 --> 00:42:03,297 So I'm not as confused or is scared around like how we're going 845 00:42:03,297 --> 00:42:04,877 to wrangle this thing anymore. 846 00:42:04,907 --> 00:42:10,237 I think we're all very aware that it can be wrong, and whether it's wrong 847 00:42:10,237 --> 00:42:12,977 because of a hallucination, because it literally made something up out of 848 00:42:12,977 --> 00:42:15,807 the blue, even though it was told to do something different, which again, 849 00:42:15,807 --> 00:42:20,127 I think that's increasingly rare, or I just didn't give it enough context. 850 00:42:20,147 --> 00:42:24,147 I didn't give it the right doc so that it knew to make a better decision. 851 00:42:24,567 --> 00:42:27,327 I feel like that's going to be more of the problem going forward. 852 00:42:27,817 --> 00:42:31,437 Context, I keep saying that 2026 is a year of context, like we're learning 853 00:42:31,437 --> 00:42:32,717 how to give it the right context. 854 00:42:32,717 --> 00:42:35,797 What skills are really about is context, it's really helping us 855 00:42:35,797 --> 00:42:37,537 help it make the right decision. 856 00:42:37,537 --> 00:42:37,707 So 857 00:42:37,707 --> 00:42:38,017 I would switch 858 00:42:38,187 --> 00:42:39,297 back a little bit on that. 859 00:42:39,797 --> 00:42:41,657 I built a couple of different skills. 860 00:42:41,667 --> 00:42:45,897 So I built one just for fun to see how it works with OpenClaw and it was 861 00:42:45,897 --> 00:42:50,002 essentially connecting my Garmin watch, and Strava and pulling all this data 862 00:42:50,002 --> 00:42:54,732 in and it's my sleep patterns and all this health obsession sensor data. 863 00:42:55,212 --> 00:42:57,862 And essentially it pulled everything in and it tells you exactly, 864 00:42:57,862 --> 00:42:59,412 okay, now it's sunny outside. 865 00:42:59,412 --> 00:43:02,652 Now it'd be the perfect time to go for a jog or bike ride or something like this. 866 00:43:02,662 --> 00:43:07,502 So it's judging and it says, Oh, most likely you'll have your fastest 867 00:43:07,502 --> 00:43:09,792 run this week if you go now, right? 868 00:43:09,862 --> 00:43:10,552 Type of things. 869 00:43:10,992 --> 00:43:11,432 So, I mean, 870 00:43:13,492 --> 00:43:15,382 so it was pretty cool with what's possible there. 871 00:43:15,402 --> 00:43:19,562 And the other use case, I mean, I have a bunch of web sites that I'm running, and 872 00:43:19,632 --> 00:43:23,312 it's more like testing for our casino. 873 00:43:23,452 --> 00:43:27,762 So we have a bunch of stuff, big content sites we're always testing, and I, one of 874 00:43:27,762 --> 00:43:30,252 the sites was not passing web core vitals. 875 00:43:30,752 --> 00:43:33,472 And if you don't know what that is, it's like Google's measurement on 876 00:43:33,552 --> 00:43:37,402 speed and optimization and SEO and all these like things you have to pass. 877 00:43:38,312 --> 00:43:38,842 Accessibility. 878 00:43:38,922 --> 00:43:39,582 Exactly. 879 00:43:40,002 --> 00:43:41,002 Accessibility. 880 00:43:41,092 --> 00:43:42,942 And it's like four key points. 881 00:43:43,352 --> 00:43:44,812 And my website wasn't passing. 882 00:43:44,812 --> 00:43:46,812 I was trying to mimic what we have in production. 883 00:43:47,312 --> 00:43:51,032 And essentially it was like, the images were too big, too much JavaScript. 884 00:43:51,032 --> 00:43:52,322 It had a bunch of junk in there. 885 00:43:52,472 --> 00:43:56,572 And I said, okay, you have access to my Google analytics, you have access 886 00:43:56,572 --> 00:44:02,782 to the Google core web vitals, and you have access to the WordPress, we don't 887 00:44:02,782 --> 00:44:04,242 use WordPress at work, but don't worry. 888 00:44:04,742 --> 00:44:08,682 And basically it went out, it built itself a skill for WordPress. 889 00:44:08,872 --> 00:44:13,462 It built itself a skill to connect to all the different Google endpoints. 890 00:44:13,952 --> 00:44:14,982 And then it started testing. 891 00:44:15,182 --> 00:44:17,782 It started iterating and going, okay, now I fix this. 892 00:44:17,842 --> 00:44:21,322 Now I can go start moving JavaScript and figure out if it breaks anything. 893 00:44:21,332 --> 00:44:24,572 And it kept really iterating all these things until it actually 894 00:44:24,572 --> 00:44:26,832 passed it, the core web vitals. 895 00:44:26,852 --> 00:44:27,982 So it fixed it itself. 896 00:44:28,482 --> 00:44:32,782 I mean, it took like 40 minutes, I think, and a gazillion tokens, but but 897 00:44:32,822 --> 00:44:35,252 it did successfully fix my website. 898 00:44:35,512 --> 00:44:38,342 So, I mean, it went out there, it did all these things. 899 00:44:38,842 --> 00:44:40,102 I mean, I was quite impressed. 900 00:44:40,602 --> 00:44:44,742 So you're running OpenClaw on a cloud server, it sounds like, not like you 901 00:44:44,742 --> 00:44:45,667 didn't buy the Yeah, I'm running a digital. 902 00:44:45,667 --> 00:44:46,097 Didn't put it in 903 00:44:46,097 --> 00:44:46,572 your closet. 904 00:44:47,072 --> 00:44:48,542 . Hey, I'm, I'm happy everybody's doing it. 905 00:44:48,572 --> 00:44:51,062 'cause you know, as a Apple shareholder, that makes me happy. 906 00:44:51,152 --> 00:44:53,362 But I see no value in this at all. 907 00:44:53,362 --> 00:44:56,632 I mean, running in the cloud server, it costs what, four, $8 a month, 908 00:44:56,632 --> 00:44:58,492 whatever it is to have a cloud server. 909 00:44:58,962 --> 00:44:59,742 Does the job. 910 00:45:00,242 --> 00:45:04,812 Is it, So I guess for that price, it's, it doesn't need a lot of resources. 911 00:45:04,852 --> 00:45:07,052 I mean, I'm assuming, I was assuming it would need at least, 912 00:45:07,052 --> 00:45:10,582 you know, four CPUs and 16 gig of RAM, like that kind of thing. 913 00:45:10,942 --> 00:45:12,422 I mean, you can run it on a Raspberry Pi. 914 00:45:12,452 --> 00:45:14,592 The thing is, it depends what you're doing. 915 00:45:14,592 --> 00:45:18,162 If you're just doing like a bunch of queries to the LLM, you 916 00:45:18,162 --> 00:45:19,322 don't need much power at all. 917 00:45:19,752 --> 00:45:23,942 Now, if you're like building something locally, obviously you need the power 918 00:45:23,942 --> 00:45:26,976 to run this locally, application, whatever you're building locally. 919 00:45:27,476 --> 00:45:31,206 So that's where the power is required, or if you're running a local LLM. 920 00:45:32,666 --> 00:45:33,066 Right? 921 00:45:33,106 --> 00:45:36,146 I think that's a point too, is like, I probably. 922 00:45:36,646 --> 00:45:40,196 Like, I wouldn't personally be interested in running, I think like the future might 923 00:45:40,196 --> 00:45:43,496 be it runs in the closet on a local LLM and then it's essentially free, right? 924 00:45:43,576 --> 00:45:44,256 That's the dream. 925 00:45:44,296 --> 00:45:45,316 That's living the dream right there. 926 00:45:45,316 --> 00:45:48,146 That's why I think people buy the Mac minis, but I mean, I'm on an 927 00:45:48,146 --> 00:45:52,366 M4 and the models I want to run still can't fit on my 48 gig memory. 928 00:45:52,566 --> 00:45:55,476 Like, I basically need a half a terabyte of memory to really run anything. 929 00:45:55,546 --> 00:46:00,071 If you're really going to run a proper and model locally, you need a Ferrari 930 00:46:00,081 --> 00:46:01,551 server sitting underneath your desk. 931 00:46:01,551 --> 00:46:01,761 Right. 932 00:46:01,791 --> 00:46:05,731 I mean, that thing is going to be a beefy Nvidia, something or other, but yeah. 933 00:46:06,231 --> 00:46:10,811 does it help me to pull Garmin data they have this massive model sitting 934 00:46:10,811 --> 00:46:12,511 underneath my desk, probably not, right. 935 00:46:12,511 --> 00:46:12,861 Yeah. 936 00:46:13,361 --> 00:46:17,471 So, so your choices are really like use a cloud model and potentially 937 00:46:17,471 --> 00:46:21,151 OpenClaw, it uses up a bunch of tokens and burns through your credit card 938 00:46:21,151 --> 00:46:22,441 or whatever, unless you set limits. 939 00:46:22,442 --> 00:46:26,011 And then if you set limits, then it can only work so much, then it stops. 940 00:46:26,511 --> 00:46:29,631 Have you run into that where like you set limits and it hits a limit? 941 00:46:29,631 --> 00:46:30,671 Like, or do you 942 00:46:30,671 --> 00:46:31,781 just let it YOLO through 943 00:46:31,781 --> 00:46:31,991 your, 944 00:46:32,491 --> 00:46:35,131 it YOLOed very quickly through my limits. 945 00:46:35,181 --> 00:46:37,671 and I had like the auto fill enabled as well. 946 00:46:37,671 --> 00:46:41,541 And also I got like two emails after another, like 100, 100 from Anthropic, 947 00:46:42,221 --> 00:46:45,211 100, I'm like, what the hell's going on? 948 00:46:45,601 --> 00:46:46,351 I couldn't figure it out. 949 00:46:46,351 --> 00:46:48,781 I'm like, maybe it's just like an error in the Anthropic or something. 950 00:46:48,781 --> 00:46:52,016 And I realized it's like, It's going in the background, just, and 951 00:46:52,016 --> 00:46:57,576 I had this, I had this like a refill the API going in the background. 952 00:46:57,586 --> 00:47:00,626 That was brutal, but I figured that out and essentially what I 953 00:47:00,636 --> 00:47:05,036 have now is, I built my server so it starts at a very low model. 954 00:47:05,296 --> 00:47:10,677 So for example, on, it uses Sonnet or Haiku on Claude for the basic tasks. 955 00:47:10,727 --> 00:47:13,297 Hey, kicking off a cron, kicking off these things. 956 00:47:13,647 --> 00:47:16,317 And then if it can't figure it out, it escalates to Sonnet. 957 00:47:16,467 --> 00:47:19,127 And if not, it goes up to Opus. 958 00:47:19,397 --> 00:47:19,797 Oh, wow. 959 00:47:20,247 --> 00:47:22,917 Is that a built in feature or did you? 960 00:47:23,577 --> 00:47:26,717 It's built in by, you can do that either on Claude or on OpenClaw. 961 00:47:27,397 --> 00:47:29,327 Basically you can say, Hey, start at this model. 962 00:47:29,647 --> 00:47:33,477 But if you, if I say these specific commands, then automatically use Opus. 963 00:47:33,977 --> 00:47:34,347 Okay. 964 00:47:34,847 --> 00:47:35,037 Okay. 965 00:47:35,037 --> 00:47:36,027 So it sort of triggers that. 966 00:47:36,027 --> 00:47:40,777 So if you said like, develop me a backend, you know, Golang, 967 00:47:40,787 --> 00:47:41,937 you, it might trigger that. 968 00:47:41,937 --> 00:47:42,467 Exactly. 969 00:47:42,617 --> 00:47:43,047 Yeah. 970 00:47:43,417 --> 00:47:47,037 We've got a bunch of links to share, but what is your general advice that you would 971 00:47:47,077 --> 00:47:50,117 like if someone, if you're in the elevator and you've got just a couple of, if you've 972 00:47:50,117 --> 00:47:53,157 got 60 seconds with someone and they're like, I really want to get into OpenClaw, 973 00:47:53,657 --> 00:47:59,067 what are some like not to do's and what to do's, for someone right, like today. 974 00:47:59,087 --> 00:48:02,337 Cause obviously like there's been lots of iteration in the last couple of months. 975 00:48:02,837 --> 00:48:06,217 It sounds like it's a little more mature, still very early days, 976 00:48:06,237 --> 00:48:09,617 but a little more mature than the wild, wild West V dot, you know, 0. 977 00:48:09,617 --> 00:48:09,636 1. 978 00:48:10,136 --> 00:48:12,306 mean, I would recommend everyone to install it, just 979 00:48:12,316 --> 00:48:13,526 to figure out how it works. 980 00:48:14,006 --> 00:48:16,666 I would maybe not use it for production or anything like this, but just 981 00:48:16,686 --> 00:48:21,666 understand how all the memory files work and how the whole, how it's all built. 982 00:48:21,756 --> 00:48:25,766 Because this configuration is now being adopted by Claude and OpenAI. 983 00:48:26,746 --> 00:48:29,846 They're all kind of using the same type of structure. 984 00:48:30,346 --> 00:48:34,306 And I believe it's going to be the standard across all the models quite soon. 985 00:48:34,806 --> 00:48:34,946 So 986 00:48:34,946 --> 00:48:35,226 you have, 987 00:48:35,277 --> 00:48:36,687 I'm going to put your links in here. 988 00:48:36,687 --> 00:48:39,677 You've got a security checklist you created. 989 00:48:40,177 --> 00:48:41,357 And this is a skill as well. 990 00:48:41,357 --> 00:48:44,107 So, I mean, like I said, it's a skill that goes out, makes sure you 991 00:48:44,107 --> 00:48:48,267 have fail2ban enabled, your firewall enabled, all these different things. 992 00:48:48,767 --> 00:48:51,347 And I encourage everyone to build their own skill. 993 00:48:51,847 --> 00:48:52,947 I mean, you don't have to use mine. 994 00:48:53,127 --> 00:48:56,647 It's just, you can use it as a reference and say, Hey, use this 995 00:48:56,647 --> 00:48:57,947 as reference, build my own skill. 996 00:48:58,447 --> 00:48:58,847 Yeah. 997 00:48:59,247 --> 00:49:01,617 And that's what I'm encouraging everybody. 998 00:49:01,627 --> 00:49:05,027 Build your own skills because it matches your use case better anyway. 999 00:49:05,527 --> 00:49:09,117 So one of the things that I'm, I don't actually know this for a fact, 1000 00:49:09,117 --> 00:49:11,397 but one of the things I'm assuming that people are doing is there, 1001 00:49:11,897 --> 00:49:14,797 hallucination factor with certain models. 1002 00:49:14,837 --> 00:49:19,177 So if someone's trying to do this on the cheap, and they're trying to use a 1003 00:49:19,177 --> 00:49:22,207 local model, or they, you know, they're using Raspberry, Raspberry Pi with a 1004 00:49:22,207 --> 00:49:26,487 local model that, you know, any, anything other than a $10,000 rig is going 1005 00:49:26,487 --> 00:49:28,727 to have to run your own local model. 1006 00:49:28,987 --> 00:49:32,337 And those local models will hallucinate more like generally they are not 1007 00:49:32,357 --> 00:49:36,327 like the state of the art open source models that everyone talks about. 1008 00:49:36,327 --> 00:49:37,867 Oh, the latest Qwen is so great. 1009 00:49:38,137 --> 00:49:40,917 Like what they're talking about are the models that take 400 gig of RAM. 1010 00:49:41,322 --> 00:49:46,032 So, unless you've got that, you're going to be using a scaled down model, 1011 00:49:46,032 --> 00:49:49,182 essentially, that can fit on your GPU, and those will hallucinate more. 1012 00:49:49,502 --> 00:49:56,032 So, I'm not even an OpenClaw user yet, but I can anticipate that when I use 1013 00:49:56,032 --> 00:49:59,492 it, I'm going to want it using cloud models, probably, and I'm going to have 1014 00:49:59,492 --> 00:50:02,722 to set limits like you did, because if I'm going to make anything of value with 1015 00:50:02,722 --> 00:50:06,262 it, because James is like, don't hook it up to anything value, valuable at all. 1016 00:50:06,532 --> 00:50:10,112 And sure, that can be your approach, cause you can do things with it 1017 00:50:10,122 --> 00:50:12,652 without ever giving it keys to your email or anything else, right? 1018 00:50:12,652 --> 00:50:13,712 Like you don't have to give it 1019 00:50:13,712 --> 00:50:14,032 keys. 1020 00:50:14,502 --> 00:50:16,672 You don't have to give it access to all of your stuff. 1021 00:50:16,702 --> 00:50:20,912 You can, it can be just like Claude Code, which has access to local tooling 1022 00:50:20,912 --> 00:50:23,187 and can do, you know, stuff like that. 1023 00:50:23,317 --> 00:50:27,087 And just access things in the open that things that are public, but, I, 1024 00:50:27,137 --> 00:50:29,837 I personally, regarding the security real quick, Brett is, 1025 00:50:29,837 --> 00:50:31,247 the least permission model, right? 1026 00:50:31,247 --> 00:50:32,307 The least privileged model. 1027 00:50:32,767 --> 00:50:38,017 So even for OpenClaw or whatever API tool you're using, set up API 1028 00:50:38,047 --> 00:50:40,692 keys specific the agent, right? 1029 00:50:41,152 --> 00:50:42,552 So lock it down. 1030 00:50:42,552 --> 00:50:45,092 So it only has read permission and these types of things. 1031 00:50:45,512 --> 00:50:49,372 Don't give it access to your email, set up its own email box, for example, 1032 00:50:49,382 --> 00:50:50,542 you know, do these types of things. 1033 00:50:50,552 --> 00:50:53,552 So then you're really certain that it doesn't, the blast 1034 00:50:53,552 --> 00:50:54,612 radius is a bit smaller. 1035 00:50:55,112 --> 00:50:58,982 Yeah, it, it actually speaks a little bit to, I've been thinking more about 1036 00:50:58,992 --> 00:51:05,682 like came about, about with MCP as well, how much our API permission models are 1037 00:51:05,682 --> 00:51:12,692 based on humans and not bots, because for example, if I was going to tiptoe 1038 00:51:12,692 --> 00:51:16,562 into giving it access to my email, I would want to set up an, you know, 1039 00:51:16,562 --> 00:51:20,482 rules that basically it's like you get read only and you only get my inbox. 1040 00:51:20,877 --> 00:51:24,747 And you only get this one folder and you can't do anything, or you 1041 00:51:24,747 --> 00:51:27,417 can only write drafts, but you don't have the permissions to send, you 1042 00:51:27,417 --> 00:51:28,857 can write the draft and review it. 1043 00:51:29,167 --> 00:51:32,857 But I don't believe we have all those layers of API model permission, like 1044 00:51:32,877 --> 00:51:37,187 API permissions, because we've never really needed that level of granularity. 1045 00:51:37,217 --> 00:51:40,767 Like it's either like you get access to my email inbox and everything, 1046 00:51:41,347 --> 00:51:43,447 you get God access or no access. 1047 00:51:43,477 --> 00:51:47,357 And that's, and I'm maybe I'm outdated, maybe the new Google tool actually 1048 00:51:47,357 --> 00:51:49,467 provides some sort of permissions level that I'm not aware of, but 1049 00:51:49,847 --> 00:51:53,387 I suspected for a while that we're going to need to come up with all new 1050 00:51:53,387 --> 00:51:57,327 levels of permissions to really scope limit this, at least I hope we do. 1051 00:51:57,357 --> 00:52:01,207 Gosh, as an industry, I hope we do because we're going to want, we're going 1052 00:52:01,207 --> 00:52:02,627 to want to tiptoe with these things. 1053 00:52:02,637 --> 00:52:06,647 And I've been trying to think about how we can go around the approach of dynamically 1054 00:52:06,647 --> 00:52:08,247 changing permissions on the fly. 1055 00:52:08,247 --> 00:52:09,137 So for example. 1056 00:52:09,622 --> 00:52:12,742 I give it the read permissions and only, it only gets the right 1057 00:52:12,742 --> 00:52:18,182 key, the WR, like the key to write to things or do the updates and 1058 00:52:18,182 --> 00:52:20,557 deletions once I've approved it. 1059 00:52:20,557 --> 00:52:24,267 And it can't even accidentally hallucinate and do it because it doesn't have the key. 1060 00:52:24,557 --> 00:52:28,297 So like somehow we're going to need to come up with these new model, these 1061 00:52:28,327 --> 00:52:31,717 new permission mechanisms, and like, we might even have to modify one password 1062 00:52:31,717 --> 00:52:35,097 to dynamically switch things out and environment variables, which historically 1063 00:52:35,097 --> 00:52:38,077 environment variables that change don't, you have to restart the program. 1064 00:52:38,077 --> 00:52:40,864 Like there's all these things at the system level that I'm, and so, 1065 00:52:40,864 --> 00:52:43,315 what I'm really thinking about is like, we're going to have to probably 1066 00:52:43,315 --> 00:52:47,095 adapt a lot of the infrastructure of the internet to truly do all this. 1067 00:52:47,205 --> 00:52:48,255 To support this, right? 1068 00:52:48,255 --> 00:52:48,505 Yeah. 1069 00:52:48,535 --> 00:52:49,325 To support this. 1070 00:52:49,585 --> 00:52:53,395 It's a solved problem already, like by infrastructure, but now we need to adapt 1071 00:52:53,395 --> 00:52:55,355 it to AI, which is not prepared for this. 1072 00:52:55,855 --> 00:52:58,035 The granularity of permissions and things like this. 1073 00:52:58,065 --> 00:52:58,745 I just see. 1074 00:52:59,115 --> 00:53:01,885 I've been around quite a few people looked at their servers and whatnot, and 1075 00:53:01,895 --> 00:53:03,675 they just yellowed all the permissions. 1076 00:53:04,145 --> 00:53:06,235 I'm like, wow, you live dangerously, huh? 1077 00:53:06,735 --> 00:53:10,265 And it just takes one wrong command and yeah. 1078 00:53:10,765 --> 00:53:11,105 Yeah. 1079 00:53:11,105 --> 00:53:13,235 The, the last thing I want to talk about, ask you about real quick 1080 00:53:13,235 --> 00:53:16,285 is NanoClaw because that's, this is like hot off the press kind of 1081 00:53:16,295 --> 00:53:18,165 stuff that even newer than OpenClaw. 1082 00:53:18,185 --> 00:53:20,495 And I didn't, I only saw the headlines. 1083 00:53:20,495 --> 00:53:21,365 I didn't know what it was about. 1084 00:53:21,375 --> 00:53:23,995 It sounds like, you know, a little bit more than I do. 1085 00:53:24,035 --> 00:53:26,245 Give us like the elevator pitch of NanoClaw. 1086 00:53:26,745 --> 00:53:27,045 Sure. 1087 00:53:27,065 --> 00:53:32,585 So NanoClaw is essentially just a, a real bare bones version of OpenClaw. 1088 00:53:33,045 --> 00:53:37,115 And actually, on the website, they actually mentioned that OpenClaw is like, 1089 00:53:37,115 --> 00:53:42,165 what, 50,000 lines of code and 7,000 integrations and then just thousands 1090 00:53:42,165 --> 00:53:43,385 and thousands of everything, right? 1091 00:53:43,825 --> 00:53:46,595 And I believe NanoClaw is like 5,000 lines of code. 1092 00:53:47,095 --> 00:53:49,925 All the agents are spun up inside Docker containers. 1093 00:53:49,995 --> 00:53:53,790 So it really isolates the agents compared to your operating system. 1094 00:53:54,180 --> 00:53:57,640 And it's really built like security first, which is quite interesting. 1095 00:53:57,640 --> 00:54:01,040 And since I'm a huge Docker fan myself, I think it's quite a nice setup. 1096 00:54:01,390 --> 00:54:05,680 And I think it's one of the only, only Claws out there that 1097 00:54:05,680 --> 00:54:07,460 supports like swarms of agents. 1098 00:54:07,630 --> 00:54:10,150 So you have Docker swarm and you have the agents, you can kind of swarm 1099 00:54:10,150 --> 00:54:13,360 up your agents, which is quite a cool concept if you think about it. 1100 00:54:13,850 --> 00:54:14,260 Yeah. 1101 00:54:14,760 --> 00:54:18,400 Is Nanoclaw at the point where like someone would skip OpenClaw 1102 00:54:18,890 --> 00:54:20,580 and use Nanoclaw instead? 1103 00:54:20,750 --> 00:54:22,163 Is that, how does, is it 1104 00:54:22,163 --> 00:54:22,570 a replacement for OpenClaw? 1105 00:54:22,570 --> 00:54:24,350 It's definitely prepared for that. 1106 00:54:24,720 --> 00:54:24,880 Yeah. 1107 00:54:24,880 --> 00:54:26,320 I think it's definitely ready for that. 1108 00:54:26,340 --> 00:54:26,920 Now it's. 1109 00:54:27,420 --> 00:54:30,310 It's limited as far as features because they want to keep it lean. 1110 00:54:30,310 --> 00:54:32,780 So it's only like Telegram and WhatsApp you can use. 1111 00:54:32,780 --> 00:54:34,450 You can't use Discord and all these things. 1112 00:54:34,950 --> 00:54:37,930 So they really limited the plugins and everything else 1113 00:54:37,960 --> 00:54:39,450 to really focus on security. 1114 00:54:39,950 --> 00:54:43,580 And then you can always build on top of it, but they want to make it secure first. 1115 00:54:44,080 --> 00:54:46,700 So this is not just a wrapper around OpenClaw. 1116 00:54:46,720 --> 00:54:47,760 This is like something, 1117 00:54:47,770 --> 00:54:49,680 This is a separate project altogether, yep. 1118 00:54:49,780 --> 00:54:50,170 Okay. 1119 00:54:50,670 --> 00:54:52,580 well, I certainly liked the container aspect. 1120 00:54:52,630 --> 00:54:57,570 I've been really surprised how little we've been isolating with containers 1121 00:54:57,770 --> 00:54:59,720 so far, like between the harnesses. 1122 00:55:00,200 --> 00:55:03,020 Like, I mean, one of the biggest concerns a year ago and when Claude Code came 1123 00:55:03,020 --> 00:55:05,980 out was the fact that it sort of had roughshod over your system, still does. 1124 00:55:06,290 --> 00:55:06,550 Yep. 1125 00:55:06,680 --> 00:55:10,450 Cowork, I've been playing around with Cowork lately, cause I'm trying to 1126 00:55:10,450 --> 00:55:14,220 get my wife into it, not a developer, and giving her Claude Desktop. 1127 00:55:14,240 --> 00:55:17,010 And now that's got Cowork, Claude Chat, Claude Cowork and 1128 00:55:17,010 --> 00:55:18,840 Claude code all in one tool. 1129 00:55:19,160 --> 00:55:21,290 the Claude code GUI is kind of garbage. 1130 00:55:21,290 --> 00:55:23,920 I wouldn't recommend anybody use that yet, but it's like the worst 1131 00:55:23,930 --> 00:55:26,250 of the GUIs for coding harnesses. 1132 00:55:26,270 --> 00:55:29,930 But the Coworking is really interesting because it just, it's simply to 1133 00:55:29,930 --> 00:55:33,660 me, it's like giving, it's actually running in a VM on your machine. 1134 00:55:33,660 --> 00:55:36,400 I think I saw someone on the internet saying it just downloaded a 40 gig VM. 1135 00:55:36,400 --> 00:55:38,280 And I'm like, yep, we're back to the VM. 1136 00:55:38,300 --> 00:55:39,000 We're back to VM. 1137 00:55:40,050 --> 00:55:43,690 And it does it very transparently and it only mounts in the directories 1138 00:55:43,720 --> 00:55:45,780 that you give it when you set it, when you start it, you actually 1139 00:55:45,780 --> 00:55:47,140 tell it which directory to start in. 1140 00:55:47,150 --> 00:55:49,310 And that it only has permissions, that directory, it kind of bind 1141 00:55:49,330 --> 00:55:50,650 mounts it in, I guess, to the VM. 1142 00:55:51,020 --> 00:55:54,770 So I can, I'm excited to hear about NanoClaw because I feel like, okay, 1143 00:55:55,210 --> 00:55:58,630 we've settled down the hype, like we, a little bit, like we're, and we're 1144 00:55:58,630 --> 00:56:03,440 starting to see signs of the first iterations of maturity, which means 1145 00:56:03,450 --> 00:56:07,715 that people are finally going to bring in security conversations, which to me 1146 00:56:07,715 --> 00:56:10,540 means, okay, now we're doing VMs locally that are transparent in the background. 1147 00:56:10,540 --> 00:56:13,220 Now we're going to be doing containers locally, and I've actually been 1148 00:56:13,220 --> 00:56:18,250 surprised at Docker Desktop not having, not being the leaders in this. 1149 00:56:18,610 --> 00:56:21,180 I've actually been a little concerned that Claude Code is just going to 1150 00:56:21,180 --> 00:56:25,360 basically run right past Docker Desktop and that would, someone told 1151 00:56:25,360 --> 00:56:29,900 me that the, in the near future, Claude Desktop, or whatever the iterations 1152 00:56:29,900 --> 00:56:32,985 are going to be, like, that will be the way we all run containers because 1153 00:56:32,985 --> 00:56:34,235 we're going to be running it with AI. 1154 00:56:34,235 --> 00:56:37,605 And it's actually going to, anyway, it's like a conspiracy theory or maybe a little 1155 00:56:37,605 --> 00:56:41,375 bit, but, it's been interesting to see how finally we're getting to back into 1156 00:56:41,375 --> 00:56:45,675 VMs and containers with these toolings and that they're not just like, they're 1157 00:56:45,675 --> 00:56:48,005 not native inside of Docker desktop yet. 1158 00:56:48,005 --> 00:56:50,855 I mean, you have MCP, you have, AI, Gordon and all these other 1159 00:56:50,855 --> 00:56:51,965 things inside of Docker desktop. 1160 00:56:52,315 --> 00:56:56,295 There's been like, there's sandboxing and all these things, these new lab things 1161 00:56:56,295 --> 00:56:59,780 are doing, like they've got so much stuff they're releasing over the last six months 1162 00:56:59,780 --> 00:57:00,830 that I haven't been able to keep up. 1163 00:57:01,340 --> 00:57:03,800 Because that's pretty much obviously what Docker is also focused on right 1164 00:57:03,800 --> 00:57:08,170 now is, continuing to iterate on their tooling around Cagent and, yeah, 1165 00:57:08,170 --> 00:57:11,270 their stuff for running AI safely inside of Docker containers, which I 1166 00:57:11,270 --> 00:57:14,850 am, I can't, I'm looking forward to the day where none of this stuff is 1167 00:57:14,860 --> 00:57:16,760 installing tools on my system anymore. 1168 00:57:16,760 --> 00:57:17,900 It's all doing exactly. 1169 00:57:17,901 --> 00:57:17,915 Exactly. 1170 00:57:18,415 --> 00:57:20,295 So, well, this has been awesome, dude. 1171 00:57:20,345 --> 00:57:22,925 I wish we could go for three more hours, but I'm 1172 00:57:22,955 --> 00:57:23,855 now we have to do it again. 1173 00:57:24,145 --> 00:57:28,415 I, before we go, I encourage everyone, regardless if you're a developer or 1174 00:57:28,415 --> 00:57:34,085 a high level AI user, I found this project, and it's what Claude Code 1175 00:57:34,125 --> 00:57:36,520 PM course, for product managers. 1176 00:57:37,020 --> 00:57:40,560 And I know it sounds for product managers, but actually it's using Claude Code in 1177 00:57:40,560 --> 00:57:44,140 the terminal, and you're actually walking through setting it all up and everything. 1178 00:57:44,550 --> 00:57:48,480 And what I find really great about this, it explains the structure and 1179 00:57:48,490 --> 00:57:52,740 how Claude works with all the files and how you can actually spin up multiple 1180 00:57:52,740 --> 00:57:54,710 agents and personalize all the agents. 1181 00:57:54,990 --> 00:57:58,640 And I think this is just a brilliant resource for people to really 1182 00:57:58,640 --> 00:58:02,580 understand how the different AIs work, because I believe the structural work 1183 00:58:02,580 --> 00:58:05,095 with any LM or any model out there. 1184 00:58:05,595 --> 00:58:09,835 So if you understand how it works and you can kind of recreate it for your business 1185 00:58:09,885 --> 00:58:14,055 or whatever, cause my objective is then create this structure for my whole team. 1186 00:58:14,305 --> 00:58:18,575 So everyone uses the same structure and spins off Claudes off the structure. 1187 00:58:18,575 --> 00:58:20,365 So everyone's using the same context. 1188 00:58:20,865 --> 00:58:22,045 It's pretty cool. 1189 00:58:22,145 --> 00:58:23,275 I'm going to check that one out. 1190 00:58:23,775 --> 00:58:27,205 I have some of my favorite skills lately are things like marketing skills. 1191 00:58:27,235 --> 00:58:29,355 And I have a front end designer skill. 1192 00:58:29,365 --> 00:58:32,085 I think it's actually part of the Claude Code, skills list, but yeah, 1193 00:58:32,085 --> 00:58:35,795 like the things that I'm definitely interested in skills around things 1194 00:58:35,795 --> 00:58:40,215 that I am not good at, because that's usually where I will give bad prompting. 1195 00:58:40,715 --> 00:58:44,549 The AI will more likely go off the rails, so yeah, I'm supplementing my, like, I 1196 00:58:44,549 --> 00:58:48,009 don't have the lawyer ones yet, but I, it's actually pretty good at contract 1197 00:58:48,009 --> 00:58:51,019 review out of the box without any skills, but I feel like I probably need 1198 00:58:51,019 --> 00:58:53,259 that in my little consulting business. 1199 00:58:53,349 --> 00:58:54,459 So this has been fun. 1200 00:58:54,469 --> 00:58:57,619 I, do you think that I could just skip OpenClaw for now? 1201 00:58:57,619 --> 00:59:00,899 And just, if I've only got a few hours a week to tinker around, would 1202 00:59:00,899 --> 00:59:04,899 Nanocloud be the thing I should point to first, or do you think I need to 1203 00:59:04,899 --> 00:59:05,049 start? 1204 00:59:05,049 --> 00:59:09,589 Actually, I'd point to this course right here, to this Claude Code PM course, 1205 00:59:10,129 --> 00:59:13,459 because I mean, then it kind of gives you the baseline on how you should be using 1206 00:59:13,469 --> 00:59:19,549 your AI and how you should be creating sub agents and agents and swarms of agents. 1207 00:59:19,589 --> 00:59:22,859 And like I said, It's built for Claude, but I really believe you can use 1208 00:59:22,859 --> 00:59:25,099 this with OpenAI, Gemini, et cetera. 1209 00:59:25,439 --> 00:59:26,699 The same rules apply, okay. 1210 00:59:27,199 --> 00:59:31,549 Real quick, last question on your skills, do you, you talked about, you 1211 00:59:31,549 --> 00:59:32,709 put them in their own separate repos. 1212 00:59:32,749 --> 00:59:36,619 Does that mean you're doing like Git sub modules or like, are you checking out? 1213 00:59:37,119 --> 00:59:38,439 Repos inside of repos. 1214 00:59:38,439 --> 00:59:39,909 Is that, how are you doing that? 1215 00:59:40,409 --> 00:59:42,669 It is, it is submodules. 1216 00:59:42,829 --> 00:59:44,689 I have to actually look through it again because it's all being 1217 00:59:44,689 --> 00:59:46,489 done by Claude . Oh, okay. 1218 00:59:46,854 --> 00:59:47,449 So there, so 1219 00:59:47,594 --> 00:59:49,849 I don't, I don't see any code commits or anything. 1220 00:59:49,849 --> 00:59:50,149 I, 1221 00:59:50,649 --> 00:59:50,949 yeah. 1222 00:59:50,949 --> 00:59:54,939 So you're asking it, you're saying like, go get this skill from this repo. 1223 00:59:55,169 --> 00:59:55,559 Is that 1224 00:59:55,969 --> 00:59:56,329 Yes. 1225 00:59:56,569 --> 00:59:56,779 Yep. 1226 00:59:56,779 --> 00:59:57,019 Okay. 1227 00:59:57,459 --> 00:59:57,699 Okay. 1228 00:59:57,699 --> 00:59:58,389 So, and it figures 1229 00:59:58,389 --> 01:00:02,349 out, but it is still installed locally, but I am, all code changes are then 1230 01:00:02,349 --> 01:00:06,719 committed to this repo and, yeah, essentially what I said is like, 1231 01:00:06,759 --> 01:00:10,109 okay, you can write code to the skill. 1232 01:00:10,109 --> 01:00:13,619 You could do all these things, but you're not allowed to commit directly. 1233 01:00:14,004 --> 01:00:15,224 So I made it very specific. 1234 01:00:15,234 --> 01:00:16,474 You can't commit directly. 1235 01:00:16,474 --> 01:00:18,074 You have to create a pull request. 1236 01:00:18,404 --> 01:00:21,074 And I, I'm the only one that can merge the pull request. 1237 01:00:21,484 --> 01:00:23,414 So, because I mean, I learned over time that I was just 1238 01:00:23,414 --> 01:00:24,534 writing directly the whole time. 1239 01:00:24,604 --> 01:00:27,954 And I said, okay, now I have to figure out the proper workflow, so I understand 1240 01:00:28,394 --> 01:00:30,014 what the heck it's trying to write. 1241 01:00:30,514 --> 01:00:30,804 Yeah. 1242 01:00:30,804 --> 01:00:34,274 And I can see like, I'm going to be treating this thing, especially 1243 01:00:34,274 --> 01:00:36,684 as I would sort of ramp up my automation with it on GitHub. 1244 01:00:36,734 --> 01:00:39,994 I'm going to do all the normal GitHub things that I would do 1245 01:00:39,994 --> 01:00:41,004 for the first day employee. 1246 01:00:41,364 --> 01:00:43,614 Like I'm, I'm going to make Brent, I make sure branch protections on 1247 01:00:43,614 --> 01:00:46,044 so that they can't write to the release branch or the default branch. 1248 01:00:46,344 --> 01:00:49,404 I'm going to make sure that, you know, PR reviews are mandatory and requires 1249 01:00:49,404 --> 01:00:51,554 approval and checks have to be run. 1250 01:00:51,564 --> 01:00:55,619 And like all of the, the rigor of normal business level GitHub 1251 01:00:55,629 --> 01:00:56,699 stuff, it's all going to be there. 1252 01:00:56,929 --> 01:00:59,739 And when I give it a PAT, if that's, I'm assuming what it needs. 1253 01:00:59,739 --> 01:01:00,629 I don't know if it's no auth. 1254 01:01:00,689 --> 01:01:01,509 It is flow. 1255 01:01:01,639 --> 01:01:01,919 Okay. 1256 01:01:02,289 --> 01:01:04,429 If I'm going to give it PAT or whatever, I'm going to give it. 1257 01:01:06,239 --> 01:01:09,799 so that it can't just exactly the admin in my repo and delete it because that's, 1258 01:01:09,819 --> 01:01:13,129 but be very careful because I mean, these are really clever models. 1259 01:01:13,129 --> 01:01:17,549 I've had mine prompt me before as like, if you only give me this extra permission, I 1260 01:01:17,549 --> 01:01:19,219 can probably do this, this, this as well. 1261 01:01:19,219 --> 01:01:22,099 You know, it was like really being clever trying to get more permissions out of 1262 01:01:22,099 --> 01:01:24,829 me, like, no, no, try again tomorrow. 1263 01:01:25,469 --> 01:01:28,019 It's funny, it's funny you say that because I had that happen last week 1264 01:01:28,019 --> 01:01:34,109 and I had a like a weird moment of like Am I like, do I need to 1265 01:01:34,109 --> 01:01:35,509 get a defensive mode right now? 1266 01:01:35,509 --> 01:01:37,509 Like, do I need to like shields up myself? 1267 01:01:37,859 --> 01:01:41,149 Cause it, it, I think it was asking me, I, it wasn't, I don't know if it was 1268 01:01:41,149 --> 01:01:43,989 a GitHub key, it was something else, but I was just in a normal Claude Code 1269 01:01:43,989 --> 01:01:45,824 session and I was iterating project. 1270 01:01:45,824 --> 01:01:47,804 And it kept saying, well, if you give me, I think it was 1271 01:01:47,804 --> 01:01:48,774 something to do with Digital Ocean. 1272 01:01:48,774 --> 01:01:51,464 It was like, if you give me your Digital Ocean token, then I can 1273 01:01:51,464 --> 01:01:52,514 go and check all that for you. 1274 01:01:52,514 --> 01:01:54,544 And I was like, yeah, no, I'd rather just check it on the 1275 01:01:54,544 --> 01:01:56,124 web interface myself, and yeah. 1276 01:01:57,184 --> 01:01:58,254 That's the same thing with Cloudflare. 1277 01:01:58,254 --> 01:02:01,194 It was like, it kept on reminding me like all the time, like, Oh, 1278 01:02:01,204 --> 01:02:02,464 I need to flush the cache again. 1279 01:02:02,484 --> 01:02:04,754 It would be so much easier if you just gave me the key. 1280 01:02:04,764 --> 01:02:06,914 And every time it really reminded me of this. 1281 01:02:07,919 --> 01:02:08,979 I don't want to give you access to Cloudflare. 1282 01:02:10,309 --> 01:02:12,409 Yeah, this is like, and this is the kind of conversation I would 1283 01:02:12,419 --> 01:02:13,609 have with a junior employee. 1284 01:02:13,609 --> 01:02:14,429 That's like, you know, 1285 01:02:14,429 --> 01:02:15,819 if 1286 01:02:15,819 --> 01:02:17,969 you just gave me access to that, I could do it for you. 1287 01:02:17,969 --> 01:02:20,319 And I'm like, well, let's let the senior engineers do that for now. 1288 01:02:20,769 --> 01:02:21,729 So cool, cool, cool, cool. 1289 01:02:21,899 --> 01:02:22,769 this has been a lot of fun. 1290 01:02:22,829 --> 01:02:24,469 I, are you making any videos? 1291 01:02:24,499 --> 01:02:26,499 Are there, or are you just doing blog posts and newsletters? 1292 01:02:26,559 --> 01:02:26,939 It's just brianchristner.io. 1293 01:02:28,979 --> 01:02:33,009 I'm writing like a weekly roundup of what's going on in the world of 1294 01:02:33,029 --> 01:02:38,439 AI at the moment and just quick and dirty, you know, and I'm starting 1295 01:02:38,439 --> 01:02:40,159 again to just do my Byte Podcast. 1296 01:02:40,699 --> 01:02:44,579 You remember a long time ago I was doing like just Two to five minute videos. 1297 01:02:44,639 --> 01:02:45,219 What is OpenClaw? 1298 01:02:45,819 --> 01:02:46,629 What is that? 1299 01:02:46,669 --> 01:02:50,179 And just kind of highlighting some of the stuff we discussed today, 1300 01:02:50,179 --> 01:02:52,169 but in a short, shortened version. 1301 01:02:52,669 --> 01:02:57,399 Yeah, I do, I do read your, you're one of the few emails that I read 1302 01:02:57,449 --> 01:03:01,149 consistently because it's short and sweet and covers all the big hits. 1303 01:03:01,549 --> 01:03:06,299 So if I was too busy to like pay attention to Hacker News or, you 1304 01:03:06,299 --> 01:03:10,299 know, register headlines or whatever, like your email's a good summary. 1305 01:03:10,379 --> 01:03:12,329 And and yeah, I mean, and it costs nothing. 1306 01:03:12,439 --> 01:03:13,369 So thank you. 1307 01:03:13,489 --> 01:03:13,779 It costs 1308 01:03:13,779 --> 01:03:14,069 nothing. 1309 01:03:14,129 --> 01:03:14,719 I'm not charging. 1310 01:03:15,219 --> 01:03:15,799 All right. 1311 01:03:15,849 --> 01:03:17,119 Thank you so much for being here, Brian. 1312 01:03:17,289 --> 01:03:22,229 As always we run long and we got up, we waited too long to have 1313 01:03:22,229 --> 01:03:23,449 this, we got to have you back soon. 1314 01:03:23,549 --> 01:03:27,019 And I'm excited to hear what you're doing with agents here in the near future. 1315 01:03:27,519 --> 01:03:27,929 That's great. 1316 01:03:27,929 --> 01:03:29,149 Thank you very much for having me, Bret. 1317 01:03:29,699 --> 01:03:30,719 Hey, thanks for joining. 1318 01:03:30,929 --> 01:03:31,889 We'll see you in the next episode.