7/6/2026 - How to Build a Persistent Digital Brain for Your AI Agents

Transcript

1. 0:01[music] [music]
2. 0:18[music] [music] Good evening, good people. What's going
3. 0:34on with my mic? Why is it so high? There we go. Little lower. That's better. Good evening. Welcome to the AI learning lab.
4. 0:42Happy Monday, July 6th, 2026. Um, tonight is the momentous night. Oh, wait, no, it's July 7th when Fable goes away. It is right. Or is
5. 0:57it midnight tonight? Or is it midnight tomorrow? Hang on. Hang on. Brandon probably knows.
6. 1:121159 Fable.
7. 1:28Okay. 11:5959 on July 7th, which is tomorrow. Ah, okay. So, you still have a day. Beautiful. You still have a day. I've been using
8. 1:44the Fable and uh it's pretty good. It's pretty interesting. Um, welcome everybody. If you're new here, this is the AI learning lab. My
9. 1:53name is Kyle Shannon and uh 77, but I don't know what time. Uh 11:5959. So basically midnight, just the second before midnight
10. 2:07tomorrow. Aloha, Linda Sherman. We got folks from all over. The US right now is trailing 3-1 in the second half at the 67 minute mark.
11. 2:23It's not looking good. They're not playing great. They're not playing great. Um, okay. So, some housekeeping.
12. 2:32That's great news. I was assuming it would be gone tomorrow morning. So, you've got Fable. It looks very bad for us. It doesn't look great.
13. 2:42If we get if we could get something now, so it was only a one-sore game, that wouldn't be too bad. Except the the the
14. 2:48Belgians are just they're so aggressive and like it's like our guys don't know. It's like no one's ever played them that hard.
15. 2:56That was a bad tackle. Yellow card. It's [snorts] not It's not looking good. [laughter] Okay. Anyway, let's I'm gonna I'm gonna
16. 3:10ignore the off in the distance. Um Okay. So, um, couple of housekeeping things. So, there's a chat here. Up underneath the
17. 3:22AI learning lab logo in the corner here is a button which pops out the chat. So, if you want to chat and ask questions,
18. 3:29um, feel free to do that there. Um, you can also, if you want to look at this live stream and check out things over in
19. 3:37the AI salon, you can open two tabs and in this lower corner right here is the button that pushes out um this window as
20. 3:46a as a pop out window. And then in that second tab, you can scroll around the AI salon or or whatever you want to do. If
21. 3:53you want to go play with other things, you can do that. Um, so that's that. Also, tomorrow night is AI Salon Presents. We've got uh Matt
22. 4:03Dixon, who's a a CTO, fractional CTO, who does AI stuff, and he's going to be talking to us about what AI your business needs. He's going
23. 4:15to be doing an exercise with us. So, it's going to be really good. So, if you uh if you've never been to an AI salon
24. 4:20presents, tomorrow would be a good one to come to. Um it's hosted by myself and Liz Miller Gersfeld. uh a cast of
25. 4:28thousands and uh we're gonna have Matt Dixon talking about AI and business. So, should be good. Um, beautiful. Trying to think what else
26. 4:42[clears throat] let me let me So, so here's the thing I would request. If you have any questions, if there's anything going on
27. 4:51that you don't understand what's happening or you want my take on something or you want me to explain
28. 4:56something, feel free to put that in the comments. Um, and I'll do my best to answer them. Let me catch you up on how I've been using Fable.
29. 5:15One of the things that I learned when the first reasoning models came out, so like a year and a half ago, is that in my life,
30. 5:31I don't have a ton of problems that require massive expertise, right? I'm not trying to cure cancer. I'm not trying to do quantum mechanics,
31. 5:46right? You know, I I might be trying to write a screenplay that doesn't suck. And and the large language models can't
32. 5:53do that. And because they're all optimized for code, they're really shitty at writing still. Um, so a lot of the the more complicated
33. 6:01stuff that I do has nothing to do with what these things are actually good at. and and I don't have things that challenge them. So that said
34. 6:13these new agentic platforms like Codeex and Claude Code and Perplexity Computer do a lot of different things. And so how
35. 6:25I've been using Fable and this was something I got off X is that I've been designing and developing what I'm
36. 6:32calling Kyle Brain inside co inside my computer that I run codecs on. And so it's a series of folders that have the
37. 6:42instruction files for like who I am and how I work and what projects I'm working on and what's the latest on each of
38. 6:48those projects. And that brain is the thing that I've been designing. And let's see, two technical questions. Okay.
39. 7:03Is anyone else getting malware on the latest update request download for Codeex? I just if you're getting any sort of notice
40. 7:12outside of Codeex, um, assume it's malware. If you launch Codeex on your computer, it'll have a little blue update
41. 7:23button in the bottom lefthand corner. Um, so you know, go ahead and ignore all those things. Ex here, here's the thing
42. 7:34to expect now that we have models like Fable that have been kind of advertised as hacker friendly, right? It can break
43. 7:45into any system. Well, now uh expect a lot more spam, expect a lot more fishing attacks, expect a lot of that stuff. So, that's that's just
44. 7:56going to come. So, I I hadn't heard about that, but but makes good sense to me. Why is it hard to build an agent with
45. 8:05claw with open claw? It's hard to build an agent with openclaw because it's designed by an engineer for engineers.
46. 8:14And so um that's part of it. And then the other part of it is the thing that I was talking about before building this
47. 8:22brain for codeex this instruction set that's in markdown files. That's how open claw works. And the problem with it is is that it
48. 8:32doesn't seem to consistently use its own rule sets. Um that's that's been my experience of it. Like I had I had an
49. 8:40automation running an open claw that was running fine. and then it just stopped and I wasn't working on it and it stopped for two
50. 8:48weeks and then two weeks later it started working again and I don't know why. So, um it's so Ivan the the answer to it is
51. 9:01all of this stuff is overly complicated and janky and unreliable because we're early. Oh, cool. Dr. Claire is doing a webinar
52. 9:15Wednesday evening details in community chat. So if you're in the salon, head on over to community chat and um and look
53. 9:25for for Dr. J's Claire's um webinar that she's doing on Wednesday. Go check that out. It was inside codeex, but good to hear.
54. 9:39Why do you think it was malware, Linda? Claude is not learning from mistakes. He forgets. Okay. So, [laughter]
55. 9:52so here's a thing that you need to know about these things. In fact, let me share my screen. We're going to go entire screen and
56. 10:06hopefully we're going to get some. It's going to stay sharing for more than a minute or two. Pop out the chat over here.
57. 10:39So let me just you can see this right? Yes. Okay. So that's codeex and here's claude.
58. 10:53Okay. So here's the deal Mr. IT. When you do a new task, they don't call it a new chat. When you do a new task in
59. 11:09Claude Code or a new session, you've got this local setup and then you've got a specific folder. And
60. 11:24if you're just in the home thing, if you just go to new task and you have it do a new task, Claude will remember whatever
61. 11:31you're talking about while that session is still active. But you have to explicitly point it to like here's my
62. 11:41um my Kyle brain folder, this Kyle brain powered thing. So, if you point it to where you have a memory system set up,
63. 11:55I think, and I'm still learning this, I think it will start writing what you're working on into that memory system. But
64. 12:03if you don't have anything selected there, it won't. Um, and if anyone knows better than I do, I use obsidian. Well, obsidian. So,
65. 12:14okay, Obsidian is just a a visual front end for what's already sitting there. So, if I go to my my finder on this
66. 12:24computer and I go to documents, let's see. Is it in documents? No, it's not. It's in Kyle Shannon. Okay. So, on my root level of my
67. 12:38computer, and I know you can't really see this, is a folder called Kyle Brain. And this is all of these markdown files that are all the
68. 12:48instructions for how my brain works. Kyle brain is here. Can I make this bigger? Um, no. Why can't I? Um,
69. 13:03that's a drag. Well, anyway, Kyle Brain is Kyle's instance of the blank brain pattern, a folder of plain files any AI
70. 13:14reads first. So, Obsidian is just a front end to some folder hierarchy, right? But but what's happening underneath it and and the way these
71. 13:24things work is they're writing these markdown files um into this directory structure. And and the way they work the the when you
72. 13:35start up codeex, when you start up clawed code, it basically starts not knowing anything until you point it to basically the
73. 13:47brain it needs to read to sort of to orient itself. And so how I've got this set up is codeex is always tethered to
74. 13:56Kyle brain. And I'm doing the same now with claude code, but it doesn't default to it. So I can create a project this
75. 14:05this Kyle brainpowered project. So anything I do within here is going to get written into that into that memory.
76. 14:12So I hope that helps. This is what you're running up against Mr. it is evidence that we're early and it's
77. 14:26evidence that if you want to figure this stuff out in a powerful way, you're going to have to put in time. You're
78. 14:32going to have to put in time understanding how these things work, what they do well, what they don't do
79. 14:36well. And this is way more complicated than just regular chat GBT, right? These are um you know autonomous agents that
80. 14:46are prompting themselves and doing all this stuff and writing things to these files and pulling things from those
81. 14:52files. I have this automation that goes to um am I still oh no hang on I'm in a different I have this automation that
82. 15:06goes to the great repurpose that see see these stories in the daily signal on the greatrepurpose.com um the automation that that generates
83. 15:18this content um has been broken for five days now because I changed the directory structure in Codeex and Codeex
84. 15:29can't figure out where stuff is. And so I'm trying to have Fable look at what's going on in Codeex. So here's here's how I work.
85. 15:39So, I'm using this thing called Tail Scale, which allows me to basically broadcast um a screen share of my MacBook Pro, my
86. 15:51M5, my AI machine, and and then how I load that. Let me just I'll close this all together. So, we're looking at a
87. 16:00sort of a window. If I go to my Finder on the computer that I'm on and I say go, I'm going to connect to a server.
88. 16:08This IP address is my M5 MacBook Pro and it's going to connect. I log into it and then I say continue and then what I'm looking at here is I
89. 16:23actually have a window. So this window is a whole other computer, right? And within this window I've got
90. 16:30all these different applications open. I've got Claude open. I've got Codeex open. I've got LM Studio. I've got
91. 16:36Obsidian open. I've got tail scale. I've got pages for some reason. Why do I have pages open? Let's see. Pages. Let's quit that. Might
92. 16:46as well not eat up all the memory. Got Telegram open for dealing with Adam. Wonder if Adam's actually alive right
93. 16:54now. Hey, are you alive? I haven't talked to him in forever because I moved his one automation over to Codeex. No,
94. 17:03he's not alive. So, that's something I could do. Um, so anyway, so what I've been doing is this is Claude. And so I'm
95. 17:13running inside this project called Kyle Brain Power. And so what I'm going to say and I'm gonna I'm on Fable Five
96. 17:21High. So what I'm going to say is, did they score? They didn't score. Okay. How many how much time left? Eight minutes left in regulation.
97. 17:35How did he miss that? Oh, it was blocked. Nice block. Um, how can I create that animation? If you want to create the pet,
98. 17:48um, you go to your settings in codeex. So, you go to settings and then you have pets. And then you can
99. 18:02basically just select them and turn them on. Um, and how you can create one is you can say to codeex like how I created
100. 18:10it is I uploaded a picture of me and I said turn this into an animation for a pet and and basically what it did was it
101. 18:19used the image gen capability in chat GPT to to make the images. it it knew what frames it needed and then it it
102. 18:27basically just wrote all those files into one of the markdown folders. So this animation lives as a bunch of
103. 18:36individual images sitting inside folders in Kyle Brain. Okay. Um All right. Let me close settings.
104. 18:52Come on. Why is that not closing? Okay, got it.
105. 19:07Okay, so over here I can say um what projects am I working on? [clears throat] Now, because codeex is tied directly into Kyle Brain and
106. 19:21it's it's using all of Kyle Brain, it should know all of the things I'm working on. So, see, see this says, I'll ground this
107. 19:35in your durable Kyle brain context, especially active work in the project registry. So, I've got a project
108. 19:41registry. So, I'm answering um from the actual files instead of memory vapor. Um I'm also going to apply your Kyle brain
109. 19:49access policy first. Um I've got the relevant orientation. The registry says the canonical active set is. So, it's
110. 19:59doing thinking here. So, notice [clears throat] though, it says read two files. It read my my skillmd file that
111. 20:06was probably on a root level and then it read one down below. So all of that work that it just did, it was going into
112. 20:14these text files in that directory structure Kyle brain and it was reading through text files that was is saying
113. 20:22what's going on. So it says I'm working on the great repurpose. I'm I'm midway in the transition between updating
114. 20:31stages four and five of the great repurpose to include this new digital self um as as a stage. Um I'm working on
115. 20:41paid speaking gigs. I'm working on this Kyle brain thing. I'm working on content evolution digital advisors and then a
116. 20:49master plan for July. And there's a whole bunch of things there that are kind of stubs dormant or lower activity
117. 20:55projects. Story Vine AI Salon in Sydney and they're not really dormant. They're just I'm not actively working on them
118. 21:01with Codeex, right? Codeex is working on these specific things. So Codeex now kind of knows who I am, kind
119. 21:10of knows what I'm working on, and I can ask it, what am I working on? Now I can go over to Claude and because Claude and Codeex both live
120. 21:21on the same computer and they both have access to this Kyle Brain set of files, they can both ask it uh they they can
121. 21:28both access those files. So what I can say is I can say um we've been doing a lot of housekeeping with codeex
122. 21:43and my projects as well as designing
123. 21:58the Kyle brain. Um, digital self. Can you look at our past chats?
124. 22:17Do an analysis of the Kyle brain um structure. and
125. 22:31give me an analysis of how we're doing. What's missing or needs
126. 22:46cleaning up or improving? And what are the next most
127. 23:00obvious things we should do? Okay, so [clears throat] now Fable is going to go analyze what Codeex has been doing.
128. 23:18And also notice it says you've used up 76% of your Fable 5 limit. Resets July 10th at 6 PM. So that's Friday. So So
129. 23:28even if So probably tonight if I play with this even a little bit more, I'm probably going to blow through Fable.
130. 23:36Yeah, Belgium's swapping out their their existing players. It's 3-1. Unless there is some miraculous uh some miraculous turnaround,
131. 23:50I think the US even with their extra player is not going to get there. All right. Get more usage.
132. 24:05Okay. All right. Red transcript. Red transcript. So, let's see. What did it read? the Claude Claude best practices red command red
133. 24:18transcript the codeex directory. So this is telling us what it's doing. So again the way these things work is they've got
134. 24:27access to files on your computer. They've got access to the internet. They've got access to tools. So you give
135. 24:34it a goal. You give it something to do and it just goes and starts doing thing and it starts updating things and
136. 24:41learning things and and things like that. All right. July 7th is when Fable is taken out of the plan until more compute
137. 24:49is available. Yeah. And [clears throat] um Enthropic today cut a deal with uh a data center provider out of I think
138. 25:00Lexington Lexington, Kentucky, somewhere in Kentucky. um like a 19 billion dollar commitment. So So here's what's about to happen. And
139. 25:10this is this is absolutely wild. They're going to start pulling Oh, wait. I I did get a message from Adam. Maybe Adam's not dead.
140. 25:25That's pretty cool. Adam's still alive. I'm here. What's up? So, my I don't know if you can see it, but my uh my OpenClaw server is is still
141. 25:37up and running. That's pretty cool. I have some starred stuff whenever you're done with show and tell. Ah, okay. Great. Cool. Okay.
142. 25:47So, what I was saying, so what I've got going right now is I've got Claude Code running or Claude Co-work. They're the
143. 25:55they're basically just different tabs in the same application. It's the application is called Claude and within
144. 26:02it you can do claude code or claude co-work. And co-work is just more for dealing with files and making documents.
145. 26:15There's four minutes. We're in We're in the 902nd minute. There's four minutes of overtime play. Yeah, it's done. Okay. Um,
146. 26:28in Australia as well. Okay. So, here's here's something that's about to happen. These big models, the the mythos level
147. 26:37model is a 10 trillion parameter model. There's apparently the GPT 5.6 Six is going to be a three trillion parameter model that was
148. 26:52actually optimized for living on some amount of of GPUs. um these bigger models are going to require more and more and more compute and as
149. 27:08they get better and as more people want to use them the de the demand for them is going to far exceed the amount of
150. 27:15compute that's there. So what's going to happen is the frontier model companies are going to keep prioritizing prog progress
151. 27:27But as they keep progressing, the demand for compute is going to be higher and higher and higher and higher.
152. 27:36And so it's going to it's going to create this this situation where very few people have access to those bigger
153. 27:43models and they're going to get very very expensive. And to the extent that we in the United States keep blocking data centers and
154. 27:54energy projects like like nuclear power plants and ways to make energy. Um the constraint is going to be energy and
155. 28:03compute. Um so if we're not very very careful um China's going to eat our lunch because they're all in on this
156. 28:11stuff. They're building power plants everywhere. They're building data centers everywhere. and they're actively
157. 28:18trying to get us to get in fights over it so we slow down. Um, so it's going to be wild. So anyway, okay. So let's look
158. 28:26at what what Fable did here. I reviewed five recent Claude sessions, the full ledger history back to June 16th and
159. 28:33walked walked the Kyle brain tree. Here's the honest read. Overall strong architecture, three week old system
160. 28:39behaving like one. The core design is working. brain.md map.mmd um promotion rules one canonical home again all of this stuff is stuff that
161. 28:53I've asked it to use the best practices that people are talking about to design this system I don't have any deep
162. 29:00thoughts on this I just want my brain to be able to work well and have these agents be able to do stuff for me that I
163. 29:08don't need to you know babysit so much that's the goal The signal publish push is the open wound. So that's the thing where
164. 29:18I had codeex working really nice publishing to our website and then when I updated went from agent brain to Kyle Braden put it at at the
165. 29:27root level of the uh of the project uh it broke. And so I'm still I'm trying to get Fable to fix the problem it caused by
166. 29:39helping me rearchitect this thing. Um that's not been going great. Like Fable hasn't been able to do that really well.
167. 29:45We'll see tomorrow morning if it did that. Um there's not a backup skill. Kyle Brain has no git repo and no remote.
168. 29:53It's been top unprotected risk. So it's never told me that before. That's actually fascinating. So, so there's a good thing. So, it it
169. 30:07basically says whatever files are on that computer, if you're not putting them to some repository and the disk
170. 30:14crashes, that's all gone. This is 20-minute task. Okay, so maybe we'll do that. Um, top level drift integrations
171. 30:21versus memory integrations. Five files exist in both places and all five differ. that directly violates your one
172. 30:29canonical home rule. Map points to memory.inttegrations. It's can okay dead state housekeeping compass is still empty. So this is my
173. 30:41values. So one of the things that um Fable said it was going to do was interview me for my values and put it in a document called compass
174. 30:51time bombs. Sim link expert requires. Okay. Next move in this order. All right, there you go. So, so this is kind of how
175. 31:02I've been using it. I go over to codeex, I do some work on on things um some on the system, some on actual
176. 31:12projects, and then I I've been using Fable to um to analyze what I've been doing and make things better.
177. 31:21For the most part, what it's done is confused me. because this is one of those things where you can set up that directory of
178. 31:34of markdown files any way you want. It doesn't matter. You can tell these things to set it up any way you want.
179. 31:41Um, and so I'm trying to do something that's solid and has logic behind it. And like one of the things that I
180. 31:48haven't heard people talk about but that I want to do is I want to create one of those markdown files is relationships
181. 31:55that I have that I want these AI agents to be able to understand who I've got relationships with and like
182. 32:03who's producer Brandon and who's Andy and who are you know the irregulars and like what are these relationships and
183. 32:10how are they um how do they relate to these different projects. It would be nice not to have to explain that every
184. 32:16time. I haven't seen any of these um memory systems that explicitly have something about relationships. I could
185. 32:22be wrong, but I don't know. Anyway, Claude doesn't know about the time you asked it to talk back to me in a minute.
186. 32:32He never talks back to me. Yeah, I know. I know. I That's That's the problem with some of these things. Unless So Ivan,
187. 32:40what you need to do is you need to say to Claude, you want it to actively to to set up an active memory system, and you
188. 32:48can just say base it on an uh Andre Karpathy's um his his Twitter post where he talks explicitly about how he sets it
189. 32:57up. have have Fable set that up for you tonight and then and then have it educate you about how to set it up to
190. 33:05point to the folder so that it always remembers stuff. Um, okay. There's some some pinned things I think.
191. 33:19Oh, and I wasn't sh How long have I not been sharing? God damn it. Oh, it just stopped. Okay, good. [laughter]
192. 33:31Oh, you just stopped it. Okay, cool. Beautiful. Okay. Um, start. So, let's go to these. Kelly Anderson, what's happening today?
193. 33:42I pelt I built something called the pass inside Claude, which serves as memo board as a memo board across different
194. 33:51channels. Claude co-work. That's cool. Oh, the pass like in a restaurant. That's really cool. Where all the
195. 33:59tickets are, right? Isn't that what the pass is? It triggers with any new session or with a manual trigger pass. So you you so so
196. 34:10basically you're saying, okay, I'm going to go into claude. I'm going to do something. Then I'm going to hit
197. 34:17slashpass if it's not already activated. And whatever I'm doing there is going to get written to the pass so that the
198. 34:23other things have access to it, right? Is that what it is? That's pretty slick. So, this is really good, Kelly. This is
199. 34:31what I'm what I'm trying to put together in this workshop. I think it's going to be an eight-week workshop.
200. 34:39And what I'm trying to put together is a combination of here's some tactical howto's like how to actually do it,
201. 34:49but then also just like some philosophies like [laughter] like here's what it is conceptually and it can be
202. 34:57[ __ ] anything you want it to be. The thing about these memory systems is that they don't have to make sense to you,
203. 35:08right? The reason people use Obsidian is because they want to be able to go look at the system and the relationships. You
204. 35:15don't need Obsidian. Obsidian's just a way for you to look at fancy graphs from what I can tell. Some people love
205. 35:24it, but like I just look at it and it's just like it just [ __ ] overwhelms me. But the AI agents don't get overwhelmed.
206. 35:32The AI agents are just like, I know where all the [ __ ] is, and I know what all the relationships are, and if I ever
207. 35:37get lost, I can just say to my agent, tell me what the hell's going on, and it'll go figure it out. New pin.
208. 35:46Wait. Oh, wait. Which one was that? Go back to that other one. Yep. I'm going to share it in the community now. That's That's really
209. 35:54good, Kelly. You know what? What I'm what I'm almost thinking. So, Brandon, do me a favor. Take a note on this.
210. 36:09So, when Andy did her separating our identity from our work workshop, we basically just continued that channel.
211. 36:23So every other week that group is still meeting to talk about the future of work and and what it means to reclaim your
212. 36:29own value. I think what I'm going to do in the great repurpose for this AI digital brain thing or digital self, I
213. 36:39don't I don't I haven't come up with a name for it yet. I think I'm going to make this an ongoing um basically agent design support group
214. 36:55because because this shit's really complicated and like the thing that you just described, Kelly, is really brilliant and it takes
215. 37:04a metaphor that's that people may know and may understand and it it brings it there. Um, okay. So, speaking of the
216. 37:14great repurpose, so the great repurpose lives inside the AI salon mastermind. And the AI salon mastermind is the
217. 37:22subscription area of the AI salon. If you haven't joined it yet, I would join it. Like, it's there are things going on
218. 37:29in there that are incredibly powerful. One of them is the AI Salon mastermind practice lab. Um, that's co-hosted by
219. 37:37myself, but really run by Sid Harrow. Um, it was designed by Liz Miller Gersfeld. Um, it is incredibly powerful.
220. 37:45It's a powerful group of people that are figuring out who they are in relationship to all this AI stuff. The
221. 37:52great repurpose is really this idea of all of our jobs are about to change. How do we identify what our value is
222. 38:01independent of our jobs? If our jobs are going to change or go away, what do we bring to the party independent of all
223. 38:09those tasks? So, one half of the great repurpose is who am I? The meetings that I've been in in the great repurpose inside the mastermind
224. 38:21have been some of the most transformative meetings of my life because I have a like I have an acting degree.
225. 38:32I'm I'm an entrepreneur. I'm like wacky and zany and I just go for it and I do things and I and I fancy myself
226. 38:41someone who's just living in the world. And what I've been learning in the workshops in the great repurpose is just how [ __ ]
227. 38:56uh committed and attached. I was to my work. I'm a CEO. I'm a founder. I'm I made this. I invented this. I did that.
228. 39:14That I fancied myself as not attached to my job. And what I've discovered is, oh, no, you're incredibly attached to it.
229. 39:23And the fact that Storyvine is still a viable company, great. But at at some point if AI comes along and says no that's not viable anymore
230. 39:36the fact that I'm starting to decouple that identity now is really important. And so if you haven't started thinking
231. 39:44about these things in your life that that the things we used to assign value to that the world rewarded us for. I went
232. 39:56to Harvard. I went to Yale. I worked at this company. I worked at that company. I did whatever accomplishment. I've written this book.
233. 40:05The value of those is sinking. It doesn't mean you're less valuable. You You had to be a pretty remarkable person
234. 40:14to do all the [ __ ] that you did, right? But in the old world, you could lead with the [ __ ] You could lead with I went to this
235. 40:24school. I got this degree. I worked at that company. I did this job. Here's the book. And then they could find out if you
236. 40:35could back it up or not. If that goes away, now it's about that. So that's one half of that. The other half is going to be this
237. 40:45learning how to create literal files that tell these AI agents who you are and how you are. and what you value and
238. 40:58who you care about and the things you want to do in the world. And then it the AI is going to help amplify who you are.
239. 41:06Those two things need to work together. And that digital side of this is really muddy right now and really nebulous and
240. 41:14really hard to set up and really hard to conceptualize. So, I think being in a group with people who are figuring out
241. 41:21who the [ __ ] they are and what they bring to the party and then putting that into some sort of form that AI agents can use and amplify.
242. 41:32That's going to be a pretty badass group of people. So, if I were you, I would get in there. [laughter] The USA team is trying to
243. 41:40figure out who the [ __ ] they are right now, says producer Brandon. Yeah, that was that was pretty brutal. Like, it was
244. 41:45brutal from the beginning. The first half was pretty painful to watch. Okay. Um, let's see. Lori Blair, Silverf Fox, I have a
245. 41:56running conversation with Claude. I have a meeting tomorrow. And it says, "Now we're all set for your
246. 42:03meeting next week." Nope. It's tomorrow. How was your meeting? We both laugh about it later. This Okay. So, okay. All right. Okay. Okay. So
247. 42:19what you're experiencing, Lori, and what Mr. It talked about and what I've been experiencing where this happened to me with
248. 42:30Open Claw with Adam for weeks and I could not get my head around it. chat GPT and and and the Claude website
249. 42:45claude.ai, they manage a lot of this stuff for you. There's code in the application layer of chat GPT that it's automatically writing those
250. 42:59memory files into your, you know, you go to preferences and it says turn on memory and then there's all the [ __ ] it
251. 43:05remembered. It's doing that automatically. these agentic platforms, Claude Code, Codeex, OpenClaw, Hermes, all these things
252. 43:21are I'm trying to think of this. Okay. Okay. Okay. Here. I'm trying to get there. Chat GPT is like one of those
253. 43:38cheap remote control cars you get at Walmart. Okay, you you um you buy the box, you get home, you open the box, you take out the
254. 43:51car, put it on the ground, you take out the battery, you put it in the car, maybe you charge it, you put the
255. 43:57batteries in the remote, you turn on the car, you turn on the remote, and you're driving, right? That's chat GPT, Claude Code, Codeex,
256. 44:12Perplexity Computer, Hermes, OpenClaw, all of them are like if you go to a hobby store and you're like, I want to make a remote
257. 44:25control airplane. And they're like, okay, cool. And they show you a box and it's got this really pretty airplane on
258. 44:30it. You're like, great. So, what do I how do I just like buy a radio? And they're like, "Oh, no." And then they
259. 44:37open the box and what the box is is a bunch of sticks with a couple of plastic parts. And you look at it and you're like,
260. 44:49"What's this?" And they're like, "That's a remote control airplane. You got to go build it." And you're like, "How long
261. 44:57does that take?" I don't know. It could be a couple of months. Couple of months? What? What? What if I screw it up? Oh,
262. 45:03well, don't because it won't fly straight and [clears throat] be a waste of time. So, that's that's what these things are.
263. 45:10So, so what we're all experiencing is the difference between ready tomade shrink wrapped out of the box memory and
264. 45:22here are systems that internally use markdown files to remember [ __ ] they want to remember. But we have to tell
265. 45:31them what we want them to remember about us and where and how. And there's an infinite number of ways to configure them. So what you're
266. 45:43experiencing silver fox and and what Ivan experienced and what I experienced with openclaw is the difference between session memory
267. 45:53and persistent memory. So session memory is I'm logged in. I've got the session going. I say to Claude, you know, talk to you
268. 46:03tomorrow. And Claude's like, great. Talk to you tomorrow. And then you quit that thing. And when you launch it again, you're like, hey,
269. 46:12why didn't you reach out to me and talk to me? And it's like, I don't know what you're talking about. If you'd remind me,
270. 46:21you know, we can talk about it. And then you say, well, we had a date that we were going to talk today. And it it will
271. 46:27say things like, "Oh, right. I remember that." Now, that's a hallucination. It doesn't remember [ __ ] Unless you explicitly point it to a
272. 46:39memory system. And so what happens the second time you run it, if you have a memory system set up, is that when you're talking to it
273. 46:49about doing something tomorrow, you essentially have to say, "Turn this into a a cron job, an automation,
274. 46:57and write what we're talking about into the persistent memory system." And then when you quit that and launch it again
275. 47:04tomorrow, instead of just being there like an idiot, not knowing anything, what it does is it launches and it goes
276. 47:11immediately into your brain and says, "Okay, who am I? Who's the person I work for? What's the stuff we've been talking
277. 47:19about? What's the stuff we're working on? Is there some automation I need to pay attention to?" And then it reads all it's literally
278. 47:27reading all of those text files. the next day. So, so, okay. So, it's this it's it's every time you turn it
279. 47:39off, it's just its memory is wiped. And so, without a clear persistent consistent with you memory system, and it's not even a
280. 47:53memory system, it's like a context system. It's it's your brain. It's the it's your memories. Without that, these
281. 48:01things just forget everything. But in the session, it seems like they've remembered everything. It's absolutely remarkable. Okay, Cam Keken.
282. 48:10Claude keeps knocking fable down to 4.8 because of my biological work. I'm sure even my memory system I set up
283. 48:18references for my bio work and it still trips it. Any ideas on what I can use Fable for? I so what I would do Cam is
284. 48:31so one of the one of the things like Alex Finn talked about this this is a thing I kind of stumbled on I might have
285. 48:38been inspired by something I saw on X what they've been using they've been using codeex as the builder
286. 48:53so they've been using codeex which is from OpenAI and it it runs GPT 5.5 and it looks like 5.6 is going to come out tomorrow. So, Codeex
287. 49:03probably is going to get as good as Fable or close to it starting tomorrow. And I doubt that they'll have such serious
288. 49:16bi biological gates on it because anthropic kind of screwed up and they had to kind of overcorrect. I assume that chat GPT won't have to do
289. 49:27that quite as bad. But anyway, I would use Codeex as the builder and the the the creator of stuff and then use Fable
290. 49:37as the interrogator and the analysis tool of what Codeex is doing. So basically just use it to analyze what
291. 49:47the work it's done, how it's organized things, what it's done, if the if the answers are right, things like that.
292. 49:53That might allow you to get around some of it because it's just checking something else's work rather than having
293. 49:59to generate it. But honest to God, I don't know. I mean, nothing that I do, nothing that I have done has triggered
294. 50:07Fable to to drop back to Opus 4.6. But you're doing biology stuff. And that's one of the no nos is biology
295. 50:14stuff. Mary Carol, did I miss? Did Kyle say what he was using Fable for? Yeah, I'm using Fable. I've got these two
296. 50:22agentic systems. I'm using Codeex as my primary building agent. The thing that I'm and I and I don't
297. 50:32have that many projects going right now. I've got the Great Repurpose effectively newsletter daily content section. Um,
298. 50:40I'm working on it to develop the workshop for building the memory. I'm building the memory. Like part of what
299. 50:47I'm doing right now with codeex is building the memory that I'm going to then teach a course on how to build. Um,
300. 50:56I'm working on it with StoryVine for sales outreach. I'm work I I've got I don't know six or seven different projects but
301. 51:08they're all relatively small and self-contained and they're all just experiments so I can learn how it works.
302. 51:15Frostbitten they didn't screw up per se. The thing that got banned is also possible to do on other AI models. Yeah.
303. 51:23Yeah. Although you know to the the argument Frostbitten is that is that Fable is of another level that if if Cam
304. 51:32isn't getting the results with Opus 4.8 or with um ChatGBT 5.5, then Fable would be something to to do.
305. 51:45Okay. Do you want to talk about Chat GBT 5.6 updates? I do, but I but I want to talk about one other thing. So, because
306. 51:52it it sounds like there's some folks here. Sally talks too much. [laughter] Hey, no Emmy. I changed my name because
307. 51:59no one understands what Lori said. >> [laughter] [clears throat] >> Oh, that's funny.
308. 52:14That's hilarious. Sally talks too much. Okay, here's another here's another skill to start exploring.
309. 52:29There's probably a full workshop on this. This is the thing. What's going to start happening inside the great repurpose inside mastermind
310. 52:39is this shit's getting pretty advanced. And it's going to be impossible to figure this stuff out on your own. And so the
311. 52:49whole idea is to get into a group of people doing more focused kind of work. a whole other area that I think could be
312. 52:57an entire workshop on its own. And if anyone's good at this and wants to do a workshop inside the great repurpose, let us know.
313. 53:08You don't need fable class large language models for everything you do. There are times when you could use Sonnet 5
314. 53:24for doing like organizational stuff or directory structure stuff where it's completely cheap, super fast and it's
315. 53:30and and and is good. It's really good. And then if you need to solve something else, you you can move up to Opus 4.8
316. 53:38and that's even better. And then there are things that you're doing that are really complicated and multi-step and
317. 53:45you want the thing to work for four hours and go do this 27step plan. You want it to come up with the 27step plan
318. 53:52and then orchestrate how it's going to happen. That might be what you use fable for. So one of the skills moving forward
319. 54:00is going to be understanding when do you use different models and in fact the frontier models are so
320. 54:14busy trying to secure data centers power and build the next model that they're ignoring a massive opportunity.
321. 54:26Someone needs to build a system that's really good at helping you dynamically choose the right model. We shouldn't have to keep that in our
322. 54:40[ __ ] heads, but right now you do. So like a thing that you could build with codeex or a thing that you could
323. 54:49build with fable cam maybe. Here's the thing you build with cable fable cam cam cam fable cam cable cable fable
324. 55:00bam cam see see what's going on here the brain she's she's shortcircuiting um here's a thing you could build
325. 55:09[clears throat] you could say to Fable hey it looks like you can't do much of this biology stuff but I want to keep doing this biology stuff
326. 55:18and what I'd like you to do is I'd like you to go into deep research mode and I want you to research what you're good
327. 55:24at, what Opus 4.8 is good at, what Sonnet 5 is good at, what Chat GPT 5.5 is good at, 5.6 six is good at the haik
328. 55:33coup models the have it go find all of the different levels the open- source models and say I want to build a system
329. 55:45where I've got a front end that's a single prompt box and I say what I want to do here's my goal today I
330. 55:53want to work on blah blah blah blah blah and I want that system to take that prompt understand it and then recommend different models for me
331. 56:05to use at different phases of that project. That would be an incredibly valuable application for anyone right now
332. 56:16because you don't the most of the work we do doesn't require a fable. most of it. Like I saw a tweet today. It said, "If
333. 56:26you're centering a div tag, if you're centering HTML, don't use Fable." Okay, but when do I use it? Like, yeah, that's
334. 56:37obvious. But if I'm in the middle of working on a project, do I need to stop at every moment and go, "Huh, in this
335. 56:44stage, is is this basically just an HTML update stage? Let me change my model." And you know sometimes when you change
336. 56:51the models it forces a new chat. So it's not as cut and dry as you know the engineering douches would like you to think. Anyway,
337. 57:09frostbitten. I have 255 AI models tested released over the past two months and the best way to prompt with them and
338. 57:17some things that they are ideal more ideal than others. And if you had a system like so if you've already got
339. 57:23that well maybe you've already built this but if you've already got that wait and why 255 you should have 256.
340. 57:34Come on Frostbitten you got to add one to that. My OCD is going crazy. We're in a digital world here. It's got to be 256. [laughter]
341. 57:45But if you built a front end for that, that is a that is a fairly strong large language model that takes in a goal or a
342. 57:54prompt. Here's the thing I want to work on and it breaks that project down and then says, "Here's the models you should
343. 57:59use for the different phases of the project." That'd be super cool. That'd be bitching. It's got to be 256. No Emmy, right?
344. 58:08CrossFit the open router model. Sakana Fuga Ultra does auto model 255 external ones and like 15 to 20 local ones I've tested.
345. 58:21Sally, I'm a 512 girl. [laughter] Kelly Anderson. Yeah, I've been kicked out of Fable every time I've tried to
346. 58:31use it. It's either I'm either wait either it's too late for it because I'm asking it to research scientific info on
347. 58:38die chemistry. It blocks me. I listen um understand that we're Okay. Okay. So, Brandon and I were talking before the
348. 58:52show started tonight. There was a there's a a study that came out that basically said that the open-source
349. 59:00models that can run on it on desktop hardware, the capabilities of them are essentially two years behind where the frontier is.
350. 59:11So what that means is that within two years we will have fable level open-source models that you can run on a laptop.
351. 59:24two years. It's not that far away, but we still get down to the we still get down to what are you trying to do and what
352. 59:34models make sense? And we shouldn't have to think about that, but right now we do. Um, here's the here's the reason why this matters.
353. 59:45If you're just using $20 a month chat GPT or $20 a month claude or even $200 a month, if you're inefficient about what models
354. 59:55you use, what's going to happen is you're going to do a project where you just forgot that you're in Opus 4.8 or
355. 1:00:02you're in Fable level or you're in GPT 5.6 Soul, I think is the big one. and you're going to do a relatively
356. 1:00:14innocuous project that's going to burn up 80% of your $200 for the month and you're going to burn it in a day.
357. 1:00:22Not because you typed in too many prompts, but because you said to an agent, I want you to go solve these 27
358. 1:00:30things. And it's just going to go do that. And and when these agentic systems go and do 27 different projects, in my
359. 1:00:39mind, I'm thinking like, oh, it it it it goes and it does a prompt and then it's done. If you watch what they're doing,
360. 1:00:46they're burning [ __ ] tons of tokens just getting set up for project one. And then it tries project one and it fails. And
361. 1:00:57it says, "Let me go back and reroute this and create a new let me write a Python application and I'm going to go
362. 1:01:02do this. So, I'm going to go do that. It's doing all this [ __ ] insane behind the scenes computing, token burning,
363. 1:01:12token maxing, as the douche douchy people say, all behind the scenes. So, if you've got shitty model selection set up, those
364. 1:01:20agentic systems are just going to blow through your credits. So, it's important to actually start thinking critically about what level do
365. 1:01:29I actually need? Can I actually run this on my local large language model, my Quen 3.6 model? And and I don't have a
366. 1:01:38good answer for this for myself. When should I be using Quen 3.6 to do stuff? When should I be using Sonnet 5 level
367. 1:01:46stuff? When should I be using Opus level stuff? And then when should I be using really expensive stuff? I don't think I'm
368. 1:01:55I don't think I'm going to need to use really expensive stuff probably ever. But it's cool to play with. But but those strata are important.
369. 1:02:09Okay. To try and work with all the 275 total models, all the data, how to process it, what not to do. So the process of a
370. 1:02:23cheat sheet, I suppose. Yeah, you're an overachiever, Frostbitten, which is good. um you don't have to deal with 250 275
371. 1:02:34models. But I think I think what would be smart is to be decently knowledgeable about what can you run on your local hardware if anything
372. 1:02:45you know what is the super fast super cheap model? What's the medium one? What's the large one? What's the baller
373. 1:02:52one? And then what's the insane one? Like if if you understood those five or six levels and just when you would use
374. 1:02:59those five or six levels for the different kinds of projects and then when you invoke them and how that's a whole thing. It's a whole thing.
375. 1:03:12All right. GLM 5.2 is pretty good. What are you running it on, Frostbitten? Are you running it on some cloud cloud-based uh
376. 1:03:22virtual machine? That's a good time, isn't it? No, Emmy. So, basically, start showing off your work in the salon, but
377. 1:03:31also LinkedIn. Yes, you should be showing off your work. Mary Mary's in the house. By the way, tomorrow is AI Salon
378. 1:03:45Presents. So, if you haven't RSVPd for it yet, go to the salon, go to events, and RSVP for the presents meeting
379. 1:03:54tomorrow night from 5 to 7 Mountain time u with Matt Dixon, who's going to talk about what level of AI you need within
380. 1:04:05your business and how you can determine that. Um, Wednesday, Dr. Jay has a webinar. Claire Jacobs. So, go look at that. It's in community
381. 1:04:19chat. There's details for it. Go look at that. Thursday, we've got week 10 of the AI salon mastermind practice lab.
382. 1:04:33So, lots going on in the mastermind, lots going on in the salon in general, lots of events. Um, get your ass there. [laughter]
383. 1:04:43Okay. Trust bitten Quen 3.6 obliterated with Opus distilled model running locally. I've been paying the $20
384. 1:04:53cloud for GLM 5.2 and others. Oh, interesting. How Opus Distilled. So, wait a minute. Is that If I go to
385. 1:05:06I'm gonna get some free consulting from frostbitten right here. Hang on. Hang on. I've got some I got some
386. 1:05:15learning to do. Okay. So, let me go to LM Studio and then go to search for models and then look for
387. 1:05:29Opus Distilled. Where do I find it? Oh, wait. Quen. Oh, is it this one? Quen 3.5
388. 1:05:5116 minutes ago,
389. 1:06:07comments. The model is who Halhoi. [laughter] Wait, can I copy this from over here?
390. 1:06:21Hang on a second. Oh, obliterated. Claude 4.7 Claude
391. 1:06:364.7 Opus uh literated.
392. 1:06:50Okay, that's 48 days ago. What was the one that I just looked at that was minutes old? This is an uncensored version of the Lord x64.
393. 1:07:02Is this it? Hello. Hello. I AI Quen 3.6
394. 1:07:1935B A3B Claude 4.7 Opus Obliterate. Yeah, this is it. SOS. Oh, share your screen. Damn it. Okay.
395. 1:07:44This is it. Download it. So, wait. And then which one? Likely too large. So, I bet I guess the 6K one. 30 30
396. 1:07:57gigabytes. I want a 64 gig machine. I can probably run the 30 gig, right? Type in No, I got it. This is it, right? Isn't this it?
397. 1:08:10You can see this, right? Which quantized model? No, that one's 36.85.
398. 1:08:27What? No, it's not. A3B 35 billion parameter A3B clawed 4.7 opus.
399. 1:08:42Oh, distilled. Damn it. >> [laughter] >> What's your GPU? I've got a M5 Max with 64 gigs of unified memory.
400. 1:08:56So, I'm thinking maybe this this 30 the the the Q6 model.
401. 1:09:28Opus Reasoning Dis distilled Opus 70 days ago.
402. 1:09:59Hi. Is it? Yes.
403. 1:10:22That's okay, Frostbin. You don't You don't need to do this. I'll I'll I'll figure it out. Let's see. 3.7.
404. 1:10:37Was it 3.7? No, it's 4.7. 4.7.
405. 1:10:54Oh, this is the uncensored version of the the reasoning distilled created with obliteration remove refusals with
406. 1:11:05transformers to know more about it. So the difference is roughly this. Obliterated models remove or weaken
407. 1:11:13refusal behavior while uncensored models are trained to behave with fewer refusal constraints. I was just bringing up the computer to
408. 1:11:23see your screen better. Oh, cool. Just try Claude 4.7. Okay. Claude 4.7. I appreciate this.
409. 1:11:37Okay, this one 48 days ago. Does that sound right? That seems old. Does that one look good? That one will work. Okay. And then
410. 1:11:52do you think that with 64 gigs the the 30 gig quantized model? Yeah, that's fine. Okay, perfect. So the 30 gig quantized model. So we're
411. 1:12:03going to download that. Okay. So, what you've just witnessed and and like notice all of the models down the left.
412. 1:12:12So, so there's these big models out there and what happens is the hacker community like goes after these models
413. 1:12:19and they do all this optimization and all this crazy ass stuff to be able to make models that historically have have had
414. 1:12:28to run on massive data center kind of hardware. and they make it more and more and more efficient. They quantize it and make it less precise
415. 1:12:38um for you to be able to run it. And so it sounds like this is going to be a good one. But now anything that I've got
416. 1:12:44that relies on the Quen 3.6 model that's not this one and I have a different thing running like it's just gonna it's like things
417. 1:12:54things change. So this is a very complicated space and it's a very it requires a lot of uh just nerding out. New pin. All right. So
418. 1:13:10let me come back over here. Okay. Let's see. Start. I'm a bit behind because because I've been on vacation the past 10 days and
419. 1:13:22was only using AI as chief of staff to work on my travel plans. Writing a brief tonight to burn through all of my Fable
420. 1:13:28tokens for tomorrow. I mean, what I would do, Suzanne, if I were you, is have Fable, you've got you've got
421. 1:13:36tonight and tomorrow. So, it's tomorrow night at midnight it goes away. I would go into claude code or co-work and have Fable
422. 1:13:46design you a memory system that it can write locally to your hard drive um that is designed to have other AI
423. 1:13:56systems point at it and then point it to the Andre Carpathy post where he talks about designing the brain and there's a
424. 1:14:05bunch of subsequent posts where people have improved on that. you can just tell Fable to go do research on whatever's
425. 1:14:12the latest best practices for a markdown brain. Um, and just have it help you design that. That's that's what I've been
426. 1:14:20working on for the past week and a half or so. It's fascinating. I feel I I have never felt more stupid except when I
427. 1:14:28first started doing Tik Tok. [laughter] Yeah, I read I read this one before, Brandon. That's cool. That's good.
428. 1:14:38All right. Uhuh. [clears throat] Thank you, Frostbitten. I appreciate that. So, that's now downloading. Is it? Yeah, I'm up to 3%. So, these
429. 1:14:47models are big. [laughter] And actually, Cam for you, I don't know if you've got hardware big enough to run
430. 1:15:01the models, but that model is based on Opus 4.7. So, it's, you know, it's it's an Opusish level model that you can run on local
431. 1:15:16hardware that doesn't have any of the refusal stuff built in. I was asking how to get rid of gopher holes, common household items, so horses
432. 1:15:32don't break legs. I don't have horses or acorage. [laughter] I maybe wouldn't show that one online. That's hilarious.
433. 1:15:41Oh my god, that's really funny. Just to see if it would or what it would show. That's really cool. Its responses are based off of Opus. All
434. 1:15:51right, cool. That's awesome. Mary Carol Suzanne, welcome back from vacation. I am looking for a use case to use Fable
435. 1:15:59by tomorrow night's deadline. Curious how you plan to write the brief to use all the credits. I'm assuming something
436. 1:16:06like a basic brief might just use a few of the the things. Well, one of the things that you can do in
437. 1:16:14I know they gave this to codeex, which I haven't played with it yet, but I think it's in Claude code as well, which is slashgoal sl go
438. 1:16:27slashgoal and then you can give it a really broad goal and just let it go build a plan. I I think that would be a way to burn a
439. 1:16:35bunch of credits. I'm asking because I do not know. So looking for advice on the best way. I was considering
440. 1:16:44outlining a book or speaking ideas. I think that Mary Carol, you know what might be fun, Mary Carol is um just go into fable.
441. 1:16:57Maybe you even go into sonnet like or maybe you just go into actually you know what you probably do. Maybe this is the
442. 1:17:04here's a here's a here's an idea. go into regular Claude or or Chat GPT, whichever one you use more that
443. 1:17:11whichever one's got more history of you and then go in with one of the decently strong models and say, "Hey, based on
444. 1:17:20everything you know about me, um, people keep telling me I should go do speaking or workshops or whatever it is
445. 1:17:29or you know, whatever whatever in that space that you haven't quite figured out yet, the the stuff that's feeling right to you energetically
446. 1:17:38go in and just sell it based on what you know of of me. Come up with a plan for all the things you know I could and should do. And then
447. 1:17:48you take that plan and you give that to Fable and say, "Fable, go attack this and don't come back until you've accomplished whatever, you
448. 1:18:01know, um, you know, a booking strategy. Um, you know, decks, like you could just give it a whole bunch of [ __ ] to go do
449. 1:18:11and just see how far it can get." Suzanne, I have 0% usage of Cloud this week. I have chat working on a brief
450. 1:18:18right now. I think we should treat tomorrow like a one-day consulting engagement with very expensive expert.
451. 1:18:24That's actually not a bad idea. Every token should produce something you'll re reuse for months or years. Oh, that's a
452. 1:18:32great prompt. That's a great prompt, Suzanne. If you don't know Suzanne, you need to know Suzanne. So, one of the
453. 1:18:42reasons the AI salon is so cool is that there's badasses like Mary Carol and Frostbitten and No Emmy and Suzanne and
454. 1:18:53Silverf Fox and Vicki and Brandon that just hang out in here and they're all like humble, sweet people that are just badasses.
455. 1:19:05Being in the presence of brains that think not like yours is incredibly powerful. It's a double first name, which is super confusing.
456. 1:19:18[laughter]
457. 1:19:37I did that with chat GPT a few days ago. I've just been nervous about giving the playbook to Fable. Give it give it. Everyone's gonna have
458. 1:19:46everything. Mary Mary Carol. It's Everyone's too busy making their own [ __ ] They're not going to look at
459. 1:19:52yours. Well, this is You know what's funny? This is Cindy Coon's playbook is at at one point Cindy [ __ ] realized she
460. 1:19:59she was you know thinking oh what if people take my framework and whatever and she wrote her book on on um on threat casting
461. 1:20:09and she realized oh screw it I'm just going to give them everything. I'm going to give them the whole the whole thing.
462. 1:20:16So she gives them the whole thing and nobody does the whole thing. They hire her anyway because because everyone's lazy. [laughter]
463. 1:20:24So, so even if they have all the answers, they won't use them anyway. That's pretty funny. Oh, man. All right.
464. 1:20:38The claw model picker. Look. [laughter] Is that Silverf Fox? Silverfox went and built the thing. Can I copy this? I
465. 1:20:50can't. Can I? No. Damn it. chat. I'd also I'd also ask it to become your intellectual property advisor. For
466. 1:21:02every framework, it should answer is this original. Who else is saying something similar? How can I would make it more
467. 1:21:09distinctive? How can it become a book? How can it become software? Can it become a workshop? Can it become a
468. 1:21:14certification? Can it become an AI agent? These are really good. Andy [laughter] the [ __ ] is time to be afraid. We're
469. 1:21:28going to be dead soon. [laughter] Thanks, Andy. I told you this is going to turn into a support group. [laughter]
470. 1:21:43Oh [laughter] my god. Not with life escape philosophy. Yeah, we're all going to be dead soon, except we're we're
471. 1:21:53[laughter] going to be falling out on our deathbed. We're we're going to wake up and realize, "Oh [ __ ] you've been
472. 1:21:58saved by AI." Oh, damn. I thought I was out. [laughter] I can hear Andy saying that in my head. [laughter]
473. 1:22:09We've all got our brands, right? That's awesome. Oh my god, that's so cool. We don't know each other. Oh, Mary Carol and Suzanne
474. 1:22:19Welker Jurgens, you don't know each other. You two should you two should connect. You you you probably have overlapping experiences.
475. 1:22:27Um yeah, that that would be a good connection. You two should talk. They need the great repurpose. We all need
476. 1:22:33the great repurpose. That's what I was I was saying, Andy, before you got on. Um, I'm I'm thinking about the the workshop
477. 1:22:42I'm going to do on the digital brain thing and and I'm just realizing that that designing and building these
478. 1:22:49agentic systems like this is not a trivial thing and it's not a short it's not a short compact thing and it's not
479. 1:22:57like just here's how you use chat GBT and here's how you do prompting. It's like sort of this infinitely
480. 1:23:03configurable erector set. Um, and and there's going to be best practices that emerge out of it, but it just it seems
481. 1:23:10to me this needs to be an ongoing thing. So, anyway, that's where my head was. Oh, man. How are we doing on time? Oh,
482. 1:23:18we're right at getting right at right at the top of the hour. I really hate the delay with these messages.
483. 1:23:27Oh, Susanna is the first person I had a one-on-one with. Oh, that's awesome. Cool. Beautiful. I love that you guys like connect and
484. 1:23:35have one-on- ones with each other. It's so [ __ ] cool. It's so good. It's so good. I'm I'm a little envious because I'm I I
485. 1:23:46I put a lot of energy out and I don't connect enough and I'm I'm working on getting better at that. Being present to
486. 1:23:54the human beings. This is one of the things that has been probably the most transformative for me inside the
487. 1:24:00mastermind, inside the great repurpose, the the practice lab is really learning to be present to the human beings in
488. 1:24:09those rooms and and understanding the difference that happens when you actually connect, when I actually connect.
489. 1:24:20Should make a list of the one-on- ones you want to have. That's a good idea. I've had a few and they're always fun.
490. 1:24:27Very cool. By the way, airfare is off the hook. The fall is bonkers. Just need to vent about that. Domestic,
491. 1:24:34international, wherever you're flying. It's ridiculous. Awesome. More expensive flights, please. Um, okay. So, we're wrapping it up.
492. 1:24:44Tomorrow, AI Salon presents 5 to 5 to 6:30 Mountain time. Um, Matt Dixon is our speaker. He's a fractional uh CTO
493. 1:24:55and AI guy. He's been around, he's been in the trenches forever. He's going to help us all think about our businesses
494. 1:25:02and and where and what AI to apply to it, like how to use AI in your business. It should be a really fascinating um
495. 1:25:11hour and a half. Um so that's tomorrow. Wednesday, Claire Jacobs is doing um a webinar, a workshop um that's in events
496. 1:25:22and it's also posted in um chat, the community chat. Then Thursday is the practice lab, Friday's office hours and
497. 1:25:32all the other good stuff. Okay, tonight was one of those nights where I like I'm talking so much about stuff
498. 1:25:46that I just barely understand that I don't know what I said. [laughter] Like like when I talk about stuff where
499. 1:25:58I know what I'm talking about, I can barely remember what I said. But like tonight, it's just like I don't even
500. 1:26:03know. I don't even know where we went. All right, I'm gonna have to go watch my own replay. [laughter]
501. 1:26:14All right, everyone. Um, I hope you had a good Monday night. Um, sorry about the US in the uh old soccer ecc fantasy.
502. 1:26:23That was that was that was dashed pretty soundly tonight. [laughter] I barely understand either. That's why I
503. 1:26:31have my chief of staff on it. Exactly. My claude model picker is in irregular's chat. Cool. Uh such as it is after my
504. 1:26:39minute in the lab. Beautiful. So what that is what what Silverf Fox put together is I sort of had a brain fart about you
505. 1:26:48need to understand different levels of models and when you would use certain ones. You don't always have to use the
506. 1:26:54big badass one. It'll just blow through your tokens. But how are you supposed to know? Well, it sounds like she went and
507. 1:27:00built that, which is good. Uh, I have no idea why I don't remember anything about the show to Oh, look. Soccer. [laughter]
508. 1:27:12I'm being trolled by my producer. I will have you know this is this is unfair unfair advantage. That's that's a goal. Go. [laughter]
509. 1:27:27Oh my god, it's true. All right, we we will we will meander through this. We're we're going to figure this stuff out
510. 1:27:38and uh we're going to be in good shape. You know, it's funny. I was for a long time thinking about this AI stuff.
511. 1:27:50The fantasy that I had was that the AI stuff was going to get good enough it was just going to do its own thing.
512. 1:27:58I don't think how that's how it's going to go. I think how it's going to go is the models are going to get really
513. 1:28:04really good and it's going to require a whole new way of thinking about how we work. And there's going to be lots of work to
514. 1:28:14do to design the future of work and how we take who we are and bring that to life and how right all that
515. 1:28:21stuff. So I'm I'm I'm actually feeling bullish about the future and the requirement for humans to figure a bunch
516. 1:28:28of stuff out um independent of these machines getting smarter and smarter. All right, beautiful everybody. Have a
517. 1:28:36good night. I will see you Wednesday. Well, come to AI Salon Presents. I will see you there tomorrow and then
518. 1:28:43Wednesday night we will back be back here and we'll play with some things. I hope I assume maybe fingers crossed
519. 1:28:51we're playing with chat GPT 5.6 and everyone's like oh god it's so awesome but it probably won't be that. It may
520. 1:28:59not even be out but if not maybe we'll go make songs ino. [clears throat] [laughter] We haven't done that in a while. All
521. 1:29:06right, peace out. Thanks everybody.