Ep. 65: Claude Knows Kung Fu. Midjourney's Still in the Dojo with v8.1?

0:00:00Intro and the boys are back

Play from 0:00:01 8:88:88 0:00:01 What's up everybody Midjourney Fast Hours podcast episode 65 Rory Flynn, Drew Brucker We're back two weeks in a row, man. We are crushing, crushing the attendance, baby. Wow. We are I know it's been a weird winter and uh you know holiday season for us. There's a lot of stuff going on projects wise that just somehow destroy Fridays, which is when we normally record. But hey, we're back.

Play from 0:00:28 8:88:88 0:00:28 Hopefully for the foreseeable future just doing this. This stuff's only moving at like a, you know, ludicrous rate. So it's like. By the time we get to like when we miss like two or three weeks is I mean, our podcast would be five hours talking about what happened. So try to get this back in on a regular schedule. Sorry for everyone who told their mailman and their dog walkers about us. We didn't show up for a couple weeks. like, y'all told me about this podcast and then it just, I never got a new episode, so I'm out.

Play from 0:00:57 8:88:88 0:00:57 Geez, man. The churn on the new podcast users has gotta be off the charts. We'll have to dig into that. uh Just come back. There's not a lot of places that do it like we do it, so it's cool. That's right. That's right. I think we're going to kind of get into like we had an awesome episode last week that I think really, you know, we went into V8 first impressions, talked through some of the likes, dislikes, what we hope to see kind of thing.

0:01:12Why the podcast got bigger than Midjourney

Play from 0:01:28 8:88:88 0:01:28 I know you had a chance to kind of dig in some more into it since then. So maybe we start there and then we'll expand into some of these other areas. I think we got a lot to talk about. You know, we covered some Claude stuff last week. I think more than we had even thought we were gonna do because I think we were around the hour mark last time when we just started talking about cloud. We're like, I don't know how deep we're gonna get. And then we ended up just ripping another 30 minutes. ah We had some really good comments on that.

Play from 0:01:56 8:88:88 0:01:56 I think people wanna see more of that too. And obviously, we started this with sort of like the Midjourney moniker, Midjourney Fast Hours podcast, right? Focus specifically on Midjourney. We've expanded that into like covering more of the, I would say just like. larger culture of this movement, you know, and kind the other tools that play into the space. And it is all interconnected, right? And so uh I think like going into these other things is super valuable for people.

Play from 0:02:23 8:88:88 0:02:23 People have sort of expressed it. So we'll probably spend more time on Claude today as well. Get into seed dance. I know you had some revelations there and then we can just kind of see where things take us, you know. Yeah. Yeah. Because it's, you know, look, when we started this, there was like a few games. There's a few gigs in town. That was it. It was like a journey and like stable diffusion. Maybe there was flux, flux one. I don't know. There wasn't a lot of options out there.

Play from 0:02:49 8:88:88 0:02:49 And it was like, we talk about like a runway, like Gen three update. It was like, we got text to video. That's cool. Now it's like, you know, I don't even know where to start or where to end. But it's. know scope creep for the podcast podcast create for sure know, that's just the evolution for anyone who's been here since day one. That's where that's where the stuff goes. But we didn't want to be pigeonholed into only talking about one thing because there is too much stuff and you guys are all probably on the same trajectory as us.

Play from 0:03:18 8:88:88 0:03:18 You're not just like, can only use one tool, you can use them all now we all have stuff to get done. So hey, that's what we're doing getting stuff done. But Rory, tell me man, tell me uh what has changed for you with mid-journey since last week. I uh think you were telling me you spent a little bit more time in there. Anything new sort of surface up for you or what were some of your observations?

0:03:25Updated Midjourney V8 take

Play from 0:03:43 8:88:88 0:03:43 I would say that few things. Number one, I started doing like a little bit more expanded ambiguous prompting, meaning like, you know, instead of just like we were doing, think lastly, we're doing war criminal. So like, let me just add like all the crazy things that I can think of from my head and just dump it in here and like no structure or anything. Stuff was coming out much better. know, this mood board stabilized for me a little bit. I was able to use multiple mood boards.

Play from 0:04:09 8:88:88 0:04:09 The one key thing I found with you know, not getting crappy output was just like screw the V7 V8 combo profile and just start a new one. um I just started new ones. I just went and spent the good thing about this is now for training your personal is now it's quick. It's really quick and it can just like scroll, scroll, scroll, hit scroll, scroll, hit scroll, scroll, hit. It's like not. Yes. No, no, no, no, no, no, no, no, no, no.

Play from 0:04:36 8:88:88 0:04:36 Yes. Yes. Yes. No, no, no. It's like so much easier now. Yeah. So it's quicker, like as soon as I stabilized those and just started using those exclusively, like everything got better. Less freak outs, less like, is this not even remotely close to what I'm looking for? So yeah, I would just say that's a recommendation, because then my mood board stabilized from that and everything. a good one. I think that's a good one, because like, I don't see, and maybe people do it.

0:04:40Why new personalization profiles helped

Play from 0:05:04 8:88:88 0:05:04 People just don't talk about it if they do, which is, I don't know how much time people really spending with the personalization profiles. And that to me is something you have to spend a good amount of time. It's worth the amount of time. You mess with it one time at a really deep level, maybe one or two times, and you're probably good. You know what I mean? um So spend the time to do that. I mean, that's really one of the primary factors of getting results that you like and good quality results.

Play from 0:05:37 8:88:88 0:05:37 so, know, rate more than they recommend, you know, just keep going. And again, to Midjourney's credit, the way that you, you know, rank these now uh is just such a better experience, right? Like, I mean, I, you know, I mean, just sitting at the hot keys of one, two and three, and then just like... go through 10 and not clicking anything you're just like my god this this sucks but So if anyone's unfamiliar, this is what it looks like now to choose your personalization code.

Play from 0:06:06 8:88:88 0:06:06 You can just be like, all right, yeah, I like this one, or I like this one, I like this one. It's so much easier. You can put it in the four column, the three column. To me, this is so much faster and better. And it's like, all right, I can see stuff also in relation to other things instead of just the two pairs where most of the time you'd go through 10 that you didn't like. And then you'd be like, all right, here's a two pair. And it's like, OK, I like both.

0:06:31The weird Midjourney magic is back

Play from 0:06:32 8:88:88 0:06:32 then I can't pick. Yeah, it's like. it's uh, it does just the one thing but I was I was finding you know was getting way more of like back into what I was trying to do which was like this sort of you know, really weird like digital realism kind of thing and like getting great like great texture and just like great depth and things that I was looking for like there was just To me this week was was more so like I was getting stuff that I was looking for you know, like these these types of sort of really more abstract, but like with the realism baked in, which is sort of how I like things.

Play from 0:07:09 8:88:88 0:07:09 um You know, really just looking for, again, like the mid-journey things that people, that other generators can't do. Like again, that's sort of what we've always been looking for. Like this one got me, I was like, that's cool. Like I'd be, you know, this stuff to me, like I'm throwing, I do a lot of presentations, especially a lot of like stage presentations and things like that. And it's like, like to throw stuff in there that looks like this. So people are just like, what, you know, it may be like a little bit of like, uncomfort, but like, what's he talking about?

Play from 0:07:38 8:88:88 0:07:38 Like, keep the attention versus like a big, big page versus text. Like when you see that come on there or, you know, these, these like cracked quartz faces that I was doing, um, you know, like getting, getting someone's attention with something like this, you know, or like that, where it's just like, what the, that's, that's weird. I haven't seen that today. Basically, it's going back to the same reason we started using this stuff, right?

Play from 0:08:05 8:88:88 0:08:05 why... This one got me the other day. I was like, oh yeah, cool. Like, I like that. Like, that makes me feel weird. You know, like it's... hey Roy is this with a mood board or no? This one is just my profiles attached and this just prompting. Yeah. uh show the prompt too? Because I wanna see uh kind of again, like if the structure or coherent, like what you put in here in terms of what that generated, because again, like going back to the jump from V7 to V8, apparently just a little bit less magical and know, blah, blah, blah, and more natural language or even structured prompts, which.

Play from 0:08:48 8:88:88 0:08:48 I think somebody was saying, why did you guys just use work? It's like we're trying to show that being nerfed a little bit. We've been talking about the templates or the structure for a long time since the beginning, but I'm just curious how something like this worked out for you. So this, I mean, this listened to me, you know, the 85 mil, I thought that was pretty good. Like the tacky sharp with a little bit of like drop off. Had half her face is chipped quartz, half her face is natural skin.

Play from 0:09:16 8:88:88 0:09:16 Well, this is also a, this is also a variation. So if I was going back to the original one, this was the original, um half her face is chipped quartz, half her face is natural skin, nailed it. Like skin has been torn off to reveal broken amethyst crystal interior. It looks like it's listening to that intricate face immersion. So it was like, wanted it to be not like an overlay, but like sort of like part of the whole thing. Jagged quartz sharp rippled textures, it's nailing that.

Play from 0:09:42 8:88:88 0:09:42 Blending with contrasting large and small chunks. We've got the large chunks, the little small chunks. Detailed skin texture, got that. Harsh flash Cleveland, we get that. And I just always like to throw intricate details in there for that. It seems to trigger everything. So coherence wise, I mean, this is a hard prompt to nail. Like for example, I ran this on the other on the other ones on the other platforms.

Play from 0:10:09 8:88:88 0:10:09 Let's see if I have this here. That is just for an example purpose. Yeah, which one are these? What we got? we got Nano, think we got Flux, we got like, Pix, what's it called? Re, what the hell is it? ReCraft. And I think this might be, either Gemini or something like that.

Play from 0:10:35 8:88:88 0:10:35 I forgot. But again, like you can tell what the difference is here, right? Like this one, this one had the most sort of like special sauce to it, if that makes sense, like the most like creativity baked in, but you know. It's not this. It's good. Like I said, that made me feel something. I was like, that's cool. what's interesting is that prompt too. So you didn't specify a background, but the 85 mil I think is defaulting or communicating that this is maybe a studio shot, which I think is sort of interesting, right?

Play from 0:11:13 8:88:88 0:11:13 Cause most of these have sort of that plain studio background, right? It's kind of an interesting observation there. Like I said, once I started going back down that rabbit hole, it was like, okay, this works. I just need to change my profile codes. This one I thought was interesting that I listened to this.

Play from 0:11:39 8:88:88 0:11:39 Yeah. that I remember from 5 to 6, you know? Yeah. Yeah. Yeah. you know revealing intricate gears chrome metal part interiors wiring Pistons micro tech like technical components of an exploded product view yet artistic portrait style harsh flash equipment like I don't know if I listened to the last portion of that but Sort of got this I would like to I'm probably edit the corners here and immerse it more but you know, that's when it started sort of like, okay Let's let's try to let's try to push that a little bit.

Play from 0:12:12 8:88:88 0:12:12 How far can we take this? Right? And then it's like, you know, you start adding, I think I have some style references in here where I started adding some different style references to make it a little bit more like this one. I thought was just fucking cool. Like, I don't know why. It's just like, it's so strange and off that I like it. Look, I like some of this, this imperfection that comes through with some of this stuff.

Play from 0:12:37 8:88:88 0:12:37 Um, and again, like, you know, like this one so weird. I was like, I miss. this stuff, which is why I come back. You know, I miss like getting these weird outputs that somehow feel like you can do something with them. ah But yeah, I mean, that's basically what I've been playing around with here. Nothing, mind shattering like this stuff. I like this, you know, like this makes me feel happy that I can do this again, because if this was gone, I was like, I don't really know what else to like.

Play from 0:13:07 8:88:88 0:13:07 This is what makes me push the button. This is what makes me push very strong, very subtle remix. Let's try that prompt again. Let's add, you know, style references to this. Let's like sort of just go with it. But, you know, that's that's just I think we're. I've been with it now, it's just OK, let me just have fun. I just want to do this. there is an interesting note here. And coincidentally enough, I didn't make off Sours this week.

0:13:24Midjourney 8.1 might be the real V8

Play from 0:13:36 8:88:88 0:13:36 Now I did see David put out something about, and I think I sent you a text about this, but it was like, yeah, we're working on like a different version of V8 right now. I think they heard the noise. I think they heard the fact that if you're going to release a new model, people are waiting on for a year and then it goes into Alf and it's still not hot right out of the gate. Kind like what we were talking about. mean, it's just probably, they probably can't operate like that in the future.

Play from 0:14:05 8:88:88 0:14:05 I mean, that may have worked early on, but that's, can't do that, I think at this point. But now I'm reading sort of these off-sour notes. You might find this interesting if you haven't already heard this. So there's an upcoming model. considered VA beta, but it's likely to be named 8.1 to simplify naming. Internal testing shows strong positive feedback overall expected release early next week, Monday through Wednesday.

Play from 0:14:33 8:88:88 0:14:33 Final, now I'm gonna read that goal. uh No, it doesn't matter. uh 8.1 may have edge cases where it performs. worse than 8.0 but not significantly in most areas. Plan is to keep V8 available for around two weeks post launch. Encourage users to compare outputs and report regressions. Long-term, 8 will be deprecated with 8.1 serving as the main model.

Play from 0:15:04 8:88:88 0:15:04 Why even really say that? I don't know. So that's where we're at. I mean, it sounds like, you know, the real V8 that we were sort of expecting is the 8.1, which we don't have yet. So take that for what it's worth. Maybe you're gonna think, hey, you know, I've probably been wasting time on this, blah, blah, blah. I never, and I don't think you do either, look at it that way with these tools because there are lessons to be learned, whether it turns out to be a good thing or ultimately fails, right?

Play from 0:15:35 8:88:88 0:15:35 There's something that... most of the time more than one thing that you can pull away from that that has valuability that stretches on to the new thing or in some different way. So I don't know, I just read that. was like, okay, yeah, we should probably let people know about that. I wasn't sure where they landed on it. And so it sounds like we could have 8.1 as early as next week. And so we'll see what that's about. And also, is that gonna land on the regular site?

Play from 0:16:01 8:88:88 0:16:01 Are we gonna get that just, know, do we have to go to alpha for that one? You know, please just let's just go to Midjourney and it's available there. Again, I think you're with me on this, but like, I think we can't really run those type of experiments at this point anymore, right? I those days are done. Just gotta be a little bit more polished on the new model releases. The features, it's like whatever.

Play from 0:16:27 8:88:88 0:16:27 Just the new models, if you're gonna have a bunch of people that have been loyal and paying for a long time, they're gonna have opinions like we did, where it's like, this isn't what we expected. This is not, yeah. me up when the edit model comes. Like, that's to me is when we're gonna really test this and see if people start to get back in the tool for more than just generating these first images. Because right now it's really just that.

0:16:36Why the edit model really matters

Play from 0:16:52 8:88:88 0:16:52 I think a lot of people are trying to generate whatever seed core image they have, then they're taking it over somewhere else. Mid-Journey needs to pull back on that rope and bring people back into the platform. The edit models, how good are they gonna be? They need to be damn good. You need to get the text. know, like I think you need to have that aspect of not only the editing, but like things with the text and the composition of the text, right? That Nano Banana Pro and two have really figured out because a lot of us, you know, you, me, people listening to this, this general space, they're using this for real work now.

Play from 0:17:27 8:88:88 0:17:27 And, you know, you kind of need the coherence and intelligence level to really be there and the consistency to be there. And like we said for text, doesn't have to be like product. Like it doesn't have to be just product. Like even if Midjourney, like they're saying that, I don't want to be like a advertising company. Like, okay. But I'm trying to do anything else. I can't have a street sign say like, flibber gibbet. Like I need the street sign to say something. Or a sign on a, you know, or the car logo or someone's clothes.

Play from 0:17:56 8:88:88 0:17:56 Like it doesn't have to be like, you know, whatever Nano is doing. Cause that's like a graphic design platform. Everything you want it to be, is. But Even if I'm just doing stuff for storytelling purposes, there's little features that text helps in, like that. Street signs, car logos, t-shirts, things in the background, posters on the wall, whatever. Something. It just can't be gibberish anymore.

Play from 0:18:24 8:88:88 0:18:24 That was passable two years ago, and now people see it even more, they're like, unless you just want to be... doing what I'm doing and doing robot face quartz broken head portraits, then people can be like, that's obviously AI. There's other people that don't want to do obviously AI. They want certain things. But yeah, think that's pretty much all I got on mid-journey at the current moment.

0:18:51Claude skills: what should be a skill

Play from 0:18:52 8:88:88 0:18:52 But yeah. next week. Let's see what comes of that. If it happens, when it happens, we could, you know, sort of dive into that later. But let's go to, uh let's go into Claude. You know, I think we've got a lot of people that were really interested in some of the stuff that we were talking about, if are obviously going through their own experience with this. There's not really a playbook, you know, for this stuff. I mean, this is why I think for a lot of us right here, it's kind of the sweet spot, right?

Play from 0:19:25 8:88:88 0:19:25 We like sort of like having the new tech available and the possibilities available and let our curiosity and sort of the trial and error kind of part of it run its course too. Like there's some fun in that. um You know, but at the same time too, like I think taking myself back to when I really started going deep into it. I mean, there really is not a lot of playbook. I mean, you could go on YouTube and look up somebody, know, here's the first couple of things you should do in club.

Play from 0:19:52 8:88:88 0:19:52 But like, honestly, I think I even go back to some of the things that I read and heard initially. And as I've gotten deeper and deeper and then, you know, kind of like figuring out the optimal way to it's like, those aren't even the best ways to be doing some of this stuff. No one really knows. I mean, there probably are select people, few people that really know. But point is, is I think like, uh we all are kind of experiencing this in a different way and trying to set it up.

Play from 0:20:20 8:88:88 0:20:20 You know, like we were talking about the skills and the plugins, like, should I install this? Should I not? You know, like, what do I do? So maybe we can kind of continue to go down that path, showcase some of the things that we're doing in real time and maybe even, you know, give some tips or experiences that we've had. Yeah, I think just like starting with whatever you're doing, if you do it more than once and it's the same process, it should be a skill.

Play from 0:20:46 8:88:88 0:20:46 Like that's basically it. Like from, that's not basically it. That's just like hard rule for me. If I do something and it could be done again and it's going to be done more than one time, then skill. Sometimes I've like, learned the hard way early on. I was building skills for things that maybe weren't replicable. was like, Oh, this is cool. I just build a skill and never use it. It's just like it's like yeah, I was like, this sick I should build this and then it was like there's no point for that uh so I think it's like a like a exercise in like organizing my own operation as to like what I actually do because I remember sitting to myself like a six months ago being like Can I automate anything?

Play from 0:21:25 8:88:88 0:21:25 I don't think I can because it's like everything is so random But I was like maybe that's just because I'm random and I'm not structured enough and so, you know gonna have Claude like structure you in a way that's like, here's what you do all the time. Here's how you can do it better. ah But the that was sort of like the big learning for me with skills early on was, oh, like, I don't necessarily know what a skill I was treating a skill like a custom GPT, which is like not the right way to look at it.

Play from 0:21:54 8:88:88 0:21:54 I don't think it's more so like a full process than it was just like here's a custom GPT that answers one way that I wanted to answer every single time. Right. because you can work between multiple different tools, you can work within multiple different context sources, like it just depends on what you're pulling from. I also didn't know that Claude code skills were different than Claude just like regular skills. I did not know that for the first like couple iterations like a couple months ago.

Play from 0:22:22 8:88:88 0:22:22 And I was like, why is, why does Claude code not use any of these skills? And I'm like, I have to build them specifically for Claude code. So that was a, that was a little bit. saying. Like a lot of this is just trial and error. then, you know, there's almost the fear too of like, did I like do this right? Like, did I leave things broken and kind of, you know, do things the wrong way, setting it up? And now do I need to go back and audit? I mean, I think that this is something that I think everybody should do because we, yeah, we, a lot of it at first was us really trying to figure it out.

Play from 0:22:53 8:88:88 0:22:53 If you didn't make any mistakes, you're not human. So I think it's just like one of those. things where have it audit things that you've created in the past or every so often. Like again, that could be a scheduled task that you create, you know, because yeah, these files can degrade or, you know, you do things differently now and it's not taking those things into account. mean, there are still several things that you and I were just talking about this before. The consistency of it, you know, I think you and I are both getting like a little bit worse results this week compared to last week and maybe in the week prior.

Play from 0:23:29 8:88:88 0:23:29 Sometimes you gotta reiterate things three or four different times, or I have set up workflows or connected chain sort of things that happen, right? You have two where it's like, it didn't follow at that time. And I recognize, I'm like, why didn't you follow it? Just like you were telling me. And it's like, well, I just forgot to read it. And it's like, How can I never have you do that again? So there still are some of these inconsistencies and voids that exist.

0:23:58Why Claude MD files matter so much

Play from 0:23:58 8:88:88 0:23:58 I think the audit thing is critical. I think to your point, like ah maybe even where I like, I think one thing super valuable, maybe super obvious to some people is just the Claude MD file. wasn't really like a, that wasn't something that anybody was really focused on with anything with GPT, like right with ChatGPT. And you you had your custom instructions and this and that, and that was really all the things that you had.

Play from 0:24:28 8:88:88 0:24:28 Whereas this, there's just so many different layers to it where it's like, yeah, you still have sort of like these user preferences in Claude, but that shouldn't be this deep, dense documentation. That should be reserved for something like Claude MD file. And so, maybe that's a good reminder for some people here, spend some time on the Claude MD file. If you're not sure what to put in there, ask it. I mean, like, I think things very simply like who you are, what you do.

Play from 0:24:57 8:88:88 0:24:57 the type of work that you do, the things that you work on, the projects that you have, words that you like, don't like, the way that you write, your bar for quality, blah, blah, blah, blah, like all of that is good. But ask it, and spend a lot of time on that because that's where everything is starting from with all of your future chats, right? It's pulling from the Claude MD. So if that's not something that you've revisited in a while or spent a lot of time on, I would highly suggest start.

0:25:23Building flexible AI systems, not rigid ones

Play from 0:25:24 8:88:88 0:25:24 Yeah, it also kind of like depends on what you want to do with it. Right? Like, I think that's that's it. Some people like to write with it. Some people like to run their whole business with clause. Some people like to run API's through it. I use it for a lot of different things, but it's it all kind of like that. That's the problem. I'm sure a lot of people are feeling this is just like it's so hard to define yourself and what you do when you want to do everything with it and ever change with it, because then you always have to change everything else.

Play from 0:25:50 8:88:88 0:25:50 Right. So that's yeah. That is like a constant thought for me. I'm glad you said that. Because you don't want it to be, it's like the same thing with mid-journey personalization, right? It's personalized to me, but if I ever wanted to do client work that's not my personalization, then it's like I have to turn it off and figure everything else out, right? It's the same idea with Claude, which is why everything, just like a good process that I use for, if I'm building skills or any sort of workflow or anything, it's basically like, take what we just did.

Play from 0:26:23 8:88:88 0:26:23 and like abstract it down to minimum viable quality, meaning like take all the really specific stuff out, like anything that's related to a brand, product, whatever. Like how do we just like look at the bare bones of this workflow and then build it back out to be more robust, but like not necessarily, like for example, if I ran like forensics on 500 images, right? And then I'm like, oh, I want to take all that data and then turn them into system prompts, right?

Play from 0:26:50 8:88:88 0:26:50 Like. The idea that I would do this for, if I was doing this for a car brand, right? All the data that would have been input would be like car brand and all the little things in the system promptly related to like car photography. But like the idea is I don't want it to just be for car photography. I want the process of extracting from large set of data, creating system prompt for any set of data, right? So then it's like, okay, I have to take all of that into account because if not, then everything is just so narrow.

Play from 0:27:18 8:88:88 0:27:18 You can only use it for car photography, you know, car. system prompts again, right? Like it's just, you to sort of like, in that mindset of like, how do I make this sort of like a, you know what I think about skills like now, which is so weird, man, like the matrix is so on point with everything, which is like, remember when in the beginning, in the one, in the first, in the first one, when he plugs in and he just goes, I know Kung Fu, like that's what a skill is. Like it's just like, I know Kung Fu.

Play from 0:27:46 8:88:88 0:27:46 Like that's how I'm trying to, but that Kung Fu has to be for everyone. You know, like it can't just be for me specifically. Because if you're only doing it for you, then think about that as like an individual project. It's not transferable. get, overwhelming isn't the right word, but it's hard to grasp and it becomes a little bit like, that's right.

Play from 0:28:11 8:88:88 0:28:11 It just becomes a little bit intangible and hard to grasp because then in that sense, you start thinking about the things that you're thinking about, and then your mind starts going in, well, there should be an avenue for this and this and this and this. And then I have to make sure that they're all connected and blah, blah, blah, blah. And did this override or contradict anything? I've already got set up. And then you go through all those motions. There is a lot of back and forth.

Play from 0:28:36 8:88:88 0:28:36 uh reconfigure, like I don't know about you, but a lot of the time I'm spending, especially in the last two weeks has just been around auditing and optimizing things I already have done, you know, that I've already set up, right? Like, you know, I was telling you about like that Go system, the writing system the other day. And it's like, there are a lot of files that go into that and each file is really deep. And if I change something, how does that affect something, right?

Play from 0:29:04 8:88:88 0:29:04 Like if I add, you know, 10 more post-structure architecture types that I like and I add that to a file and now I've got a different number, right? Like how does that impact? So there are just a lot of things that change. so spending time there has been valuable too, because I'm just like, okay, well, what worked for me two weeks ago and it was really good, now with the things that I've added is maybe not so good.

Play from 0:29:31 8:88:88 0:29:31 And so I need to be aware of that number one and then two, have to fix it. Yeah, so I've been I've been thinking about this a lot and it's like how do I get rid of that sort of like have to go fix everything sort of uh situation. So I've been building differently. Like I think about it as if like you're going to do like you as a person are going to do something or you get a project. It's like maybe you get a you know, maybe you get a project that's like off off your normal track.

Play from 0:30:03 8:88:88 0:30:03 Right. Like instead of it. Like, let's just say someone asks you to do video instead of photography, right? Like, when someone asks you to video, you've never done video before, but you have a photography background, right? If you were trying to come up with how you as a person would go do that, you're going to go pull different things. Like, you're going to pull, like, my photographer's knowledge, but then also, whatever films I like, like that little influence there, whatever, like, aesthetic I like from, you know, other directors or things I've seen, that's going to be there.

Play from 0:30:32 8:88:88 0:30:32 Like, all these little, like, different pieces that influence you. Like maybe there was something, you know, maybe it was like your job as a social media coordinator back like eight years ago where you had to film things on your iPhone, like all these little like requisite pieces that will go into you completing that job. Right. So I look at Claude a lot of the times is the way people build this stuff is like, you're an expert accountant, you've, you you do this and you do exactly the right thing all the freaking time.

Play from 0:30:58 8:88:88 0:30:58 Right. And it's like, then Claude becomes your, your valedictorian. out of high school, you're straight A student in college, you're graduated with a master's in accounting, just went right into accounting, never did anything else. It's like, that's all it knows. So it's like, there's no other it's like not reality. Like if the reason we're so complex, right? Yeah. you're acting, if you just do that, it's a universal global experience versus maybe what you're kind of hinting at is like situationally, how does it act, right?

Play from 0:31:32 8:88:88 0:31:32 It might act as a valid Victorian and this way in these circumstances or when you're doing these formats or you've got these types of relationships, right? Like compare that to like a social post versus an email. or slack mess, right? Like different tones, different casualness, different layer of depth, right? Different objectives maybe even to your point. yeah, like even that, you can't have this global universal, just like our personalities, they're multi-dimensional, right?

Play from 0:32:03 8:88:88 0:32:03 Like we've got different triggers and it's kind of the same way, right? Is what I think you're probably getting at. because you you react to things differently or or approach things differently because of your previous experience. Now, like me, there's certain things like when I take things from when I was a bartender, like no one's going to tell Claude to go act like a bartender when figuring out how to write an email. Right. But like I'm pulling from that because that's, you know, some sort of like patience, like how to figure out how to upsell people, like how to like make sure people aren't getting too drunk, but keep them entertained.

Play from 0:32:34 8:88:88 0:32:34 But like there's like this whole little like Yes. all of knowledge there that you're not tapping into with AI, which is like the sort of creative layer. All this to say is that like two things. Number one, if you're doing things that are very system and process oriented, and it's like all data and like whatever, like, you can be the valedictorian. If you have to do creative stuff, like you need a little bit extra spark for this, or else everything will be 100 % like flatlined, exactly the same. So like I've been building instead of like these rigid sort of pipelines, like do A, B, C, D.

Play from 0:33:04 8:88:88 0:33:04 Like, and you do this every time and this is how it works. And if something changes, then I have to break the whole thing. Whereas it's like, I only talk to like, or fix one thing, which is like my sort of orchestrator. He looks at the project and says like, Oh, what skills do we have that I need for this? I'll pull those in and use it for this project. So it's like, I just operate that consistently as to like how to think about things or like, this project worked.

Play from 0:33:29 8:88:88 0:33:29 This product didn't blah, blah, blah. So it keeps like updating, like the roster. the roster, the general manager, it's like roster management, like how everything runs. And then like when something works really well, it's like, okay, then we have our little column like my little agent builder on the back end that just goes, okay, cool. That thing worked like 98%. Let's put that back in the in the knowledge base. So when something similar comes up, just use that exact process or those skills, right?

Play from 0:33:55 8:88:88 0:33:55 So it's redeployable based on like, just the skills. that you have and not like, here's my one vertical pipeline, here's my other vertical pipeline. Like, because this is, here's an example. I have visuals, I did this in a presentation the other day. Like, this is my problem with N8n. Like, I don't have a problem with, know people like to use it, but like, it's so linear, meaning like, this gets triggered, then this goes to here, this goes to here, this goes to here.

Play from 0:34:25 8:88:88 0:34:25 If something breaks here or you come in, I already have the Twitter analytics. that I need like you can't you have to go still run it again right or like one of the API's is off here. You're screwed right so it's more so like been thinking more so like OK you know brief comes in we have our product manager production manager like reads the brief you know if it says like here's competitor photos go build me this like it'll curate the team and then go execute and then basically deliver everything if it's a different example it'll just.

Play from 0:34:59 8:88:88 0:34:59 hold different skills, right? So it's just, it's a way to be redeployable and like less sort of like hard coded, if that makes sense. Um, cause I think that's, that's something that probably people are running into where you're like, I built these very intricate systems, but like they require so much maintenance all the time. Uh, I'm just trying to work on if it's just, can isolate it then too, like which skill didn't work. All right, let me update that one.

Play from 0:35:25 8:88:88 0:35:25 Like, and then I'm just working on the roster, right? So it's, it's a lot less. maintenance, if that makes sense. It's easy to deploy agents. It's easy to pipelines. You just keep building the skill sets. Again, things that you're going to do more than one time. Yeah. your own skills and plugins too, you know, and give it, how often is this triggering and firing? How often have I used this? You know, like is it triggering and firing when it should?

Play from 0:35:52 8:88:88 0:35:52 Do I have to do anything special for it to trigger a fire? Like have I given you enough context as to when this should trigger fire, right? Like all of that, you could even, you know, again, going back to your point, once you've got to a certain point where you've got several of these skills and plugins, like how often am I using these? I like to sort of like rate them against each other too. Like I love the idea of just like on a scale of 0.0 to 10.0, right?

Play from 0:36:17 8:88:88 0:36:17 I want you to completely, you know, carefully and thoroughly read through each of my skill files or whatever skill file, you know, and then based on how I use these things, the quality of this file, the optimization of this file, the consistency, blah, blah, blah. I want to bring it to a 10, but give me its current rating, right? And You know, you're of rich. You're initially going to find skills that are just shit. There are a lot of skills out. You may think all these skills out there are good that are on GitHub.

Play from 0:36:43 8:88:88 0:36:43 They're not. They're not. uh And you know, you can. This is where it's like I kind of call this like the skyscraper technique takes back to the SEO days. But it's like if you find a skill in theory that is good, that is relevant to you, have Claude go through it, grade it. What would be essential to to elevate this to the next level and build on top of it to make a better version? Right? Like so much of this is the skyscraper thing. Like with SEO, that was the approach from a content perspective because you're like, how do I rank number one?

Play from 0:37:13 8:88:88 0:37:13 Well, there was a long period of time there was like the best strategy was to go through and see what ranks on the first five results, go through each of those, what content do they provide, and then just go even harder, like provide something that they don't in addition to what they already do. Right? Skyscraper tech. Now, that to me is a good play here with the skills and the plug-in stuff too. It's like you don't have to just take someone plug in and plug in and play.

Play from 0:37:39 8:88:88 0:37:39 You could take the best parts of it and build a better version, right? It's like you're trading parts and you can build custom skills because you might also have skills that overlap or contradict each other. You know that that was another thing that I ran to. It's like I've got four skills for this. It's only running you know one or two. I probably don't need these other two. You blah blah blah so. um There's a lot you know with these with these sort of like files. that really is kind of like this new thing that a lot of us are running into versus maybe originally what we were doing on, you know, ChatGPT and so forth, right?

Play from 0:38:13 8:88:88 0:38:13 about it a different way, right? Like it's like these little project folders of knowledge that just like tell people what to do. It's like an operating system that you have to build. like the one thing, my favorite new skill, this one I built this week because I was like, this is great. This is like the best thing that's ever happened to me. Like I built a lot of, basically my job is building system prompts now, right? Like if somehow. That's how I this is where I netted out in my life.

0:38:38An edge-case detector skill

Play from 0:38:42 8:88:88 0:38:42 But it's like, a lot of times what I'm doing is like one input image. I need a system prompt to be able to read that and redeploy in a million different ways, right? The system prompt has to be super scalable. Sometimes it's like 1500 lines of code. It's huge. But the one little skill that I've built is sort of like, which isn't going to execute everything. It's just like a way to get me from point A to point like, or point one to point like 1.7 really fast so that I can do like the last, you know, point three percent there or less 30 percent, which is called the edge case detector.

Play from 0:39:19 8:88:88 0:39:19 So basically like it looks like whatever product category I'm putting in and then like thinking of all the different like ways that things could be input so that I like I don't have to like test and then like I find something else. It's like damn, I need to redo the whole system prompt again. Find another thing. Retest it. So it's like here. If like I'm doing jewelry, it's like, well, typically you'd get like a picture of, know, just the jewelry and it's studio. It's like, well, what happens if it's on a person?

Play from 0:39:46 8:88:88 0:39:46 What happens if it can't see it? What happens if it comes in a grid? What happens if there's multiple products? What happens if there's multiple products in a grid and then a person stacked on the bottom? What happens if that? So like it sort of lays out all the variables for me from the beginning and we can work through what to do with them. And then I can go build it, which is just like, I've never done that before. It's normally the other way. I find a bump in the road. you know, fix it, keep going, like keep testing. But it's like, let's look at all the possible things that are going to happen first and then screws.

Play from 0:40:15 8:88:88 0:40:15 And now that's like a I take that that skill, right, which is like edge case detector and like and then just like redeploy it on other things like, what else am I not looking for? You know, like these are the other thing like the road. But it's basically like a built in prediction, like road bump prediction. Even if it's not right all the time, it's still it still catches like 70. Ish to 80 ish percent. So again, it's just like thinking about little like even the micro things that you do Right, like even the micro processes, I guess I build system prompts, but that you know, the testing portion is the biggest Is the biggest side for me.

Play from 0:40:50 8:88:88 0:40:50 How do I trim that down? Where is it my act? It's like problem identification is almost as hard as building the solution uh It's actually harder now I think I think if we go back maybe like 15 podcasts there was a whole section on like predictions and we were saying like, you know, like the best thing you can do is just keep asking the person who has the right questions right now is the person that's gonna win. Because the answers most likely can be solved.

Play from 0:41:15 8:88:88 0:41:15 You just have to ask the right questions and find the right things. But again, for people who are looking for ways to keep optimizing on this stuff, that little edge case detector, that one is my favorite. It's just like. trims, so many, so much revision time, so much re-coding, rewriting, like, cause you know, after a certain amount of time on a session, you've revised something 200 times.

Play from 0:41:41 8:88:88 0:41:41 It's like, Claude, why are you being dumb? Well, just gone through, I've read this line of code like 1 million times and edited it 200 times, right? So it's definitely something to consider there, but. You know what's a, there's like another little cool thing I think, again, the memory portion of this is interesting too. I've started this, I've got some depth, not to the, like I've only got a few people on this, but I'd love to continue to go down this path of like these memory files on people that I work with, right?

0:41:55Memory files for people and projects

Play from 0:42:19 8:88:88 0:42:19 So it could be clients, it could be you, right? uh People that I interact with regularly. And then, know, Claude knows exactly, right, in what context I know you, how you, you know, like the things that we do together, right, your position, um you know, how you like to be communicated with, right, et cetera, any active projects we got going on, et cetera, right? And like, this is where like connecting things to your Slack or your email or also help.

Play from 0:42:51 8:88:88 0:42:51 But I think that's a really nice shortcut too, because in some situations, I've got a client memory profile on a person, I can just say, I'm going to write an email to so and so about this. And it already knows about the project, and it already knows about the person. And so again, it skips a lot of the context or having to go back and forth every time of how I might write that email, or the things that I should include, or the tone in which I write the email.

Play from 0:43:18 8:88:88 0:43:18 um And again, like, going back to the point we were making of like one just global way to write my emails versus how I might write them to a specific person that holds value in my career, right? And so you might take a different approach to somebody like that versus just like any regular email you might send. So, you know, just so many of these like little things like that that are just super interesting, you know, and are layers in this process.

Play from 0:43:46 8:88:88 0:43:46 Like the memory files is super interesting. I don't know if I mentioned this last week. This one might be fun for you. But I built a skill that's called my Creative Analogy Engine. I love it. It's just like a way to take like big things and turn them into like whimsical sort of analogies. So it's just, this can be redeployed across like what I just did with Claude skills and like the Matrix sort of analogy there where it was like a plug-in like I know Kung Fu, right?

0:43:48The creative analogy engine

Play from 0:44:18 8:88:88 0:44:18 Like it's like taking big grand things that we talk about and like isolating them down into like really digestible analogies, which is good for me when I'm speaking, giving presentations, writing. Yeah. So it's like sales. simple but smart and efficient use case. That's a great point, right? Because when you get on those podcasts or on the stage, these grand ideas live in our head.

Play from 0:44:44 8:88:88 0:44:44 And in a lot of ways, because we're using them, we can shortcut the process of trying to explain them properly and rationally and with all the details to really land with the other person, whereas a metaphor can do that. such a way, right? ah You know, just in SAS, know, look for a long period of time, we're the Uber of blank, right? We're the blah, blah, blah of this, right? And it's like just an easy way to almost like create that category that, you know, gives you the foundational knowledge of what We're visual, right?

Play from 0:45:16 8:88:88 0:45:16 Like people, like people need to sort of connect the idea in their head without me saying like, I have an agent structure that operates like this, you know, it's like, yeah. It's like, no, I have a foreman that goes to, you know, the job site or like a, or like a general contractor that goes and like, actually I just need a plumber and I need a carpenter. That's all I need today. And then like goes and does the job. Like that's how I operate my agents, right? Like I have a general contractor and he's the one that's like deciding what we need.

Play from 0:45:44 8:88:88 0:45:44 And he goes and gets them. He manages them. He sees the whole process through, you know, you know, reviews, whatever was done. Right. So it's like, I've always, I've always used analogies to my advantage in sales and everything else. Cause I think it's just like the easiest way to get a point across. So it's like, can I just build this? And then it's a redeployable thing for if I'm writing something like I'm writing an outline for, for presentation, or I'm doing some, some more expansive writing, or I want to like deploy that into sort of like one of the image generation things that I'm doing.

Play from 0:46:15 8:88:88 0:46:15 where it's like I need to throw that, throw the creative analogy engine in there and see if it'll make, like dumb this down to make it like a little bit more visual or how can I explain, how can I do like for creative advertising, like how do I take big idea and then visualize it? Like, cause that's like a, that's like a great little connector. Cause a lot of what AI is missing, right? Like this is, talk about this a lot. Like I can tell it, like, like advertising is like more than just the product or whatever the service that you're advertising.

Play from 0:46:47 8:88:88 0:46:47 Like, yeah, like if you're just going to do a pot, like how are you going to advertise a pot that like, you know, you cook things in? Like, yes, like AI sees that and goes to pot, it cooks things. Right. But like we also go like, oh, you got to go like one layer deeper than that. It's like, oh, it's like communal. Like, you know, you have like family dinners that come from a pot or you can put stuff in it. Right. You can just put like, you know, this pot could be on the in your in your cabinet and you stuff other stuff in it as like a space saving tool.

Play from 0:47:16 8:88:88 0:47:16 There's a lot of different ways to connect the dots other than just like it's a pot that cooks. Like that's where AI stops most of the time, but we can go a little bit further. it's, you know, and that's, sort of what makes those connections, right? And that's why this is so much different than like data analysis, engineering for Claude. This is like, we're trying to tie human creativity with tools, which is hard, um, which gets me into something I'll talk about later on, which is the crazy stuff, but Like the real meta, like I don't know what I'm doing, but maybe I shouldn't do this, maybe I should.

Play from 0:47:50 8:88:88 0:47:50 It's a tease for you guys to hang out to the end. That's basically what that is. it's just I keep thinking like, how do we make this thing more human, which is weird. Like don't want to think like that, but it is kind of like, how do we make us more human? How to think a little bit differently because they already have the process down. Yeah, they can. rip through information. They just don't know how to sort of like deploy it the same way that we do.

Play from 0:48:17 8:88:88 0:48:17 Because it's not, they're not trying to connect with anyone other than to execute a job. Right? Yeah, we're, we're trying to connect with people. Like we have to, this has to make sense to another human. Right? Like that's, that's everything we do, basically they get put out into the world is not for us anymore. Unless you're just an artist. Everything's an email that goes to someone else that has to get digested by them. or a spreadsheet, or an idea, or a presentation, or like whatever.

Play from 0:48:48 8:88:88 0:48:48 So it's all for someone else to understand what you're saying and make sense of. you know, they're good at doing very, very straightforward things, but when it comes to creative, it's like, yes, that's why every, every time you prompt for certain things or like come up with a prompt for this. And it's like, yeah, the most generic thing possible, cause that's just executing the job. It's not. It's not going further than that.

Play from 0:49:13 8:88:88 0:49:13 It's not thinking deeper, saying maybe I shouldn't do that. Maybe what's cooler than that. you know, it's like one singular vision. But does that make sense what I'm saying, It's like a super deep hours right there. Give me the analogy for that. Oh God. Yeah, Claude, if you're listening, dispatch. Hit that. Hit that right now.

Play from 0:49:39 8:88:88 0:49:39 Because I haven't, do you talk to it a lot? Like I haven't done that enough. Like instead of like, yeah, yeah. Yeah, I mean... Yes, I do. I mean, I would say not as much as I was doing on GBT because I feel like the GBT talking instance was better. It would run longer. This one, I think, kind of cuts off after a while.

Play from 0:50:07 8:88:88 0:50:07 There's another tool I'll use, uh Whisper Flow, uh which is great. But again, same thing happens. There's sort of like a time limit there. I don't know. I mean, it just depends on what it is, probably. ah If it's pretty straightforward and not overly contextualized, it's probably, I'm typing it. If it's something that just requires a little bit more energy, probably just go voice, you know?

0:50:35Why examples beat vague prompting

Play from 0:50:35 8:88:88 0:50:35 Yeah, it's like I've talked to it when I'm not doing anything like I'm not doing anything of value. I'm just like spitting ideas. I'm working on a project. not. need to like again, because I'm always very hyper vigilant on what I input because I'm like, can't if I I miss explain this or I don't say it the right way, it's not going to get it. It's too smart now.

Play from 0:51:01 8:88:88 0:51:01 It'll take that logic and like run with it in the wrong way. example, subjective words. I think about a lot, best, give me the best. What does best mean? You like you have to outline what best means if you actually wanna utilize that, right? ah So it's just like when you get into like describing things, I think the subject, I'm always like checking for subjective words that I'm using.

Play from 0:51:26 8:88:88 0:51:26 ah You know, there's this long standing thing that I always go back to. It's just like examples are 10. times better than just trying to contextualize something, right? So if you can give it something to work off of that is an actual example or reference, that's great. And then, you know, just like with a lot of the image stuff and stuff we talked about in the past, I think just the negative or constraints that you have are probably more valuable than the positive portion of that, you know, like so.

Play from 0:51:58 8:88:88 0:51:58 you could ask for all this, but at the end of the day, what are the things it's trying to constrain within? Do you need this post to be under a certain amount of words? What are the constraints it has? What can't it do? What don't you want it to do? I just think you, again, spending time on those things are always important. And this is, again, why I think all of this conversation, really figuring out the way that you work, you almost want to...

Play from 0:52:25 8:88:88 0:52:25 put all of those global things into that Claude MD. What are the things that you know, you want it to know every single time to save you the legwork? And so if, even the subjective thing, where it's like, I want to be very careful with subjective language, if I ever say this and don't define this, I uh haven't done that, but I'm just saying that would be maybe a practical example of that. You would probably want to just go ahead and put that in, and then maybe you're not you know, so anxiety ridden like we probably are with every word we're typing, we're just like, oh shit, like one word can change this result.

Play from 0:53:02 8:88:88 0:53:02 And it's just weird, man, because like we right now are so like, there's so much fun. uh And we're so consumed with this that like even times I'm like, I'm just thinking in code now, like things are very, you know, binary, uh you know, choices now. You know, and I'm almost like just thinking about how am I explain things and I catch myself explaining things the same way I might with like with Claude, you know.

Play from 0:53:28 8:88:88 0:53:28 you like, do you like get nervous about your, your MD files now? I haven't backed up in like six places. Like this is like everything I operate on now. So like, I need like five different places to back this stuff up. don't know if I've thought of that, but it's, yeah, yeah, right? Right. Okay, so you may have just altered my afternoon uh because yeah, I'm.

0:53:38Back up your Claude brain immediately

Play from 0:53:57 8:88:88 0:53:57 or you can put it on, you know, Google Drive or you can put it in GitHub or you can just like, you know, copy it to a hard disk or whatever. Yeah. Yeah. sure they're fresh, updated. Yeah, that's a good point. It's like my IP. I, like this is another like little, little trick here. Not little trick. I'm sure everyone knows this. I'm just new to it. Like I just put things in my own private repo in GitHub.

Play from 0:54:22 8:88:88 0:54:22 Like I don't even know what a repo is. It's just like a way to store code. Right. But like, I just know that if I put it there. Yeah, yeah, you could do the private thing. Yeah, that's a good point. You could do it there. cool thing is, is that everything is redeployable on codecs. So like I can give this to when Claude is being weird, like it was the last two days. It's like, it's eating up all of my usage, and like, not listening to the skill files.

Play from 0:54:49 8:88:88 0:54:49 And I have stuff that's due and it's like taking forever. I'm just like, I'm popping right into codecs codecs like does the plugin I know kung fu, and then it's like, let's go. And it just does the same exact thing and it moves really fast. So like that's That to me is like a way again, like how do I just basically like transfer my consciousness? It doesn't matter. Like what happens if Claude, you know, got like taken out, like one of their data centers got taken out during one of these like situations, right?

Play from 0:55:15 8:88:88 0:55:15 Then it's like, uh, well, like what am I going to do today? You know, it's like one of those things, like when the internet goes down and we're all like, what, what's our, what's our job? This is it. Um, so that's, that's like, you know, Really, because this is, I might as well show this on this page. Like, organize this stuff pretty simple on here.

0:55:30Orchestrator and skill roster

Play from 0:55:41 8:88:88 0:55:41 Like I have my Claude MD with all my skills in it. Sorry, I have the folders with all the skills in it, but I my Claude MD, have my orchestrator, which is like basically my like... guy that knows where to go and find stuff. looks in the roster and then there's like the architecture, which basically is the operations, like how everything functions or could function together or has like tie backs together. The agents that have been built before. then basically the playbooks is anytime I run something, just downloads an MD file of like the review, which can then also go to the agents, which is like, okay, this is good.

Play from 0:56:13 8:88:88 0:56:13 Go build this or this is not good. Don't go build this. But this is just like the main layer that I operate from. And, you know, then it goes to, if I was going back to some of the other skills that I have in here, it's my, like my skills database that I just keep in here. And it's like all the different, like little pieces that I've built, like for this one, like basically one thing that I do, you know, for my jobs, cause it's, you know, it's, it's cool, right?

Play from 0:56:43 8:88:88 0:56:43 Like once you get to that certain point of like, I've just built a bunch of skills. It's like, I want to be able to use them, but use them like appropriately. You know one just like you were saying like don't over index on skills and then like they sort of like screw it up Right. So like this is now how like I function right like it's Basically all these little skills, but there's a process to it. It's like I call this the this one I like a lot This was one I built is just like basically I get briefs from different people and different projects from different people and they're all nothing structured Yeah that you're supposed to read and digest 100%.

Play from 0:57:21 8:88:88 0:57:21 Yes. But nothing structured either. It's not like this is how we do this. It's like basically it takes noisy data and turns it into structured data for whatever's incoming. It's just like, yeah, we want to change this. We don't want to do this. We might want to change this. I don't know about this timing. It's like when you get one of those, it's like, all right, take that, pass it off so that it can be structured so the rest of the people know what we're doing, the rest of the agents know what we're ah Asset, this one also, great.

Play from 0:57:49 8:88:88 0:57:49 If stuff comes in messy, like naming conventions are wrong, like no easy way to catalog and rip through stuff. Like it preps that before it basically sends it. So it's like a, it preps the info for the rest of it. So it doesn't have to be going through. like I said, think last episode where I'm like, don't give it the, don't try to give Claude the, know, the Charlie Day conspiracy board.

Play from 0:58:19 8:88:88 0:58:19 Don't give it this. There's too much for it to handle. It spends more time decoding this than actually reading the stuff there. Because then it can operate and function at a much higher level, where then the orchestrator goes, we have our specialist, which are the skills, we have our output that'll package everything up the way it should be. I review stuff, I can have it sent directly to the clients, I can have my playbooks written and then sent to the...

Play from 0:58:48 8:88:88 0:58:48 cross-leaning or the agent that goes and builds other agents. So it's cool, but it's all because of this roster building approach where it's like I just build a roster and then depending on whatever project comes in, it just deploys different things and then goes for review. But it comes down to that data prep stage, I would say, which I don't think a lot of people are focusing on enough.

Play from 0:59:14 8:88:88 0:59:14 It's just like you got to make stuff ingestible. Like it's just like getting bad info on a breach. Right? It's like, I'm like, I don't know how do this. uh the everything else downstream. Yeah, that then that goes into the edge case sort of skill where I'm like, most of the of the briefs you get from clients or if you're working with, you know, anyone a boss, whatever, they don't tell you all the roadblocks are going to hit you figure that out along the way.

Play from 0:59:45 8:88:88 0:59:45 So let's sort of like, foresee that ahead of time. That's, that's where I'm trying to at least like have a good plan in place for majority of them, he's probably won't get all of them. But that's that to me has been like, I look at that. I like pat I did like a little do like one of these when it was done and it worked you know I was like this is good job Rory like you thought of something that that's helping you I don't have anyone else to tell that around here you know my wife be like what nerd nerd I'm in an edge case detector think that's right though.

Play from 1:00:21 8:88:88 1:00:21 Like this has gone from, like there's still value in a good single prompt, but it's the connection of connected files, I would say. that almost is a... workflow or pre-flight checklist in itself, right? Like you have to sort of like connect those things is gonna do a lot more work than spending a bunch of time on a singular prompt that is super deep that you hope nails it completely.

Play from 1:00:51 8:88:88 1:00:51 I mean, there's still a lot of value in that, but I'm just saying for some of the things that probably require a lot more context and then super specificity in some cases, right? Like they probably should be connected files and assets that it goes through and maybe even a sequential order. right in order to get you there. And again, it's just like, you don't have to worry about perfecting the prompt because it's run this play before, right? It knows what to do.

Play from 1:01:17 8:88:88 1:01:17 And so creating those five, six files or whatever it is to get you the result is great because it's front loaded work that is gonna help you out. But. wish I could have like a more integrated, a more integrated like memory across chat, which is why I'm having it right on basically every single one. Just like, here's an MD file for what we did and like store it somewhere just in case I need to find it. Cause I know it does it locally, but like I still want like a copy of it just to have like specific context.

Play from 1:01:45 8:88:88 1:01:45 Cause sometimes like Claude, Claude, I think it was cowork a couple months ago, just crapped out on me. Just wouldn't work. had to like, reset, like factory reset it I lost all the conversations and it was I don't know what happened. But it was like, you know, virtual machine timeout and I'm like, what does that even mean? I had to do like a lot of research, like, fix code on my computer or something was going on. But, you know, there was some there's some new problem every day with it too, though, because it gets you in that mode where it's like, Oh, you could do more.

Play from 1:02:18 8:88:88 1:02:18 Don't ask me to do more, because I will because I will. And now the problem is, right, we used to operate out of several tabs in the browser, now we're operating out of several tabs in coworker code, you know, at any given time, you know, you're waiting for one to finish, you start another one, here we go, bouncing around from three to four, you know, concurrent sort of sessions at a given time, uh folder guy to begin with? Like a very clean, organizationally, like on your on your desktop and your downloads folder?

Play from 1:02:47 8:88:88 1:02:47 Me either. And now I'm like naming conventions, I got like, you know, I'm like, this is so much easier for me to to keep up with now that someone else has like started it for me. Versus like me trying to figure out a system for myself. yeah. Yeah, which it's like after day one of having your computer, you're already daft, you know, and you never want to do it again. So, you know, okay, one last thing. I was just going to say that, you know, I did kind of touch on this before, but maybe this is also a thing with, you know, you see a lot of the creators hyping up other skills or plugins or this and that, right?

Play from 1:03:21 8:88:88 1:03:21 Like just going back to that for a quick second, because this is a real problem. I get enamored with, is this something I should be using? Right? Before you go and download it and install it, right? I hope everybody's doing it, but I'm just saying like, you know, something gets to sound really good. Some of it is a little bit mystifying whether it's relevant to me or not. So like anything like I'll save that is maybe pointing to something, right?

Play from 1:03:48 8:88:88 1:03:48 Which has the GitHub link and all this stuff. I just feed that to Claude, right? Is this... Is this relevant for me? How good is this tool? But asking it those specific questions of like, carefully read through each document in here, compare it to the way that I'm currently using you. Is this something that would be of value to me? If it is, do I already have something for this? Does it contradict or overlap? Like blah, blah, blah. Sus out anything before you add it, I think, at this point is the main takeaway because it is kind of fun to collect these things.

Play from 1:04:19 8:88:88 1:04:19 You see something new and shiny and you're like, yeah, I want to do that or that sounds cool or. It's got 50,000 stars on GitHub, must be great. Maybe it's not even relevant. I did four of those yesterday. 0 for 4, whether or not Claude thought I should even mess with it. I was just like, nah, skip it, skip it. I have many skills in Claude. They're all housed in folders, so I can use them when I want to most of the time.

Play from 1:04:45 8:88:88 1:04:45 mean, the other stuff, like the agent stuff and what I'm doing there, that gets deployed automatically based on what work tree I'm working off of on my GitHub. But those types of things, I'll stash those for later. Maybe there comes a time when I need it, but it's just downloaded on my... Instead of installing it and then being like, it's conflicting with the other skills, you don't need to have it running. You can just... You know, hoard them. You can hoard them for another time.

Play from 1:05:12 8:88:88 1:05:12 Because who knows? I mean, also lot of this stuff is going to be outdated. It's gonna be outdated. That's what I'm saying. Even if you download them and don't use them though, there'd probably be better version of it in two months or a month. It's like the standard stuff you see. I saw one guy that was like, put out this whole entire agency. It was like, here's the GitHub, here's your whole agency.

Play from 1:05:37 8:88:88 1:05:37 And I... much time you spent on that, right? And I'm sure, you know, best case scenario, you make a lot of money off that, but yeah, it's either not specific enough for people to have ultimate value, and or it's gonna be outdated soon, right? Like there's just gonna be a better way to do these things, just like, you know, any of the space. Six months from now, it's probably a better way to do it, you know?

1:06:04Hard-coded flows vs flexible AI judgment

Play from 1:06:04 8:88:88 1:06:04 I like the structure. When I see something that looks like it's structured well, that's what I'm looking for versus what the content is. If the structure and the thought pattern, how it connects the dots and how it orchestrates everything, that's where the gold is. Because then it's just fine tuning. Then it's like, OK, I don't have to rethink the entire structure. Which is interesting. I saw the other day that I don't know how true it is, again, to take the stuff with a grain of salt, the stuff that I see on X, that was saying that the Like luckily I'm not tooting my own horn and this was just like probably maybe it resonated with me because of bias, but it was like, yeah, the, when, when tested against each other, like hard linear workflows versus like letting the, letting AI decide which way it wants to go with something versus like you do this or you have the freedom.

Play from 1:06:50 8:88:88 1:06:50 Like it ends up being like 14 % better if like it has the freedom versus like these hard coded workflows, which I was like, huh, again, I don't know how true that is, but it makes sense to me because that's how I operate. Yeah. think about it. We may still know things that we want, but we're unable to express them either because we forget or we don't know how to communicate it correctly versus like, and it's a concrete thing that we're given it versus something, an entity like this that has that quote unquote freedom to find the optimized way that speaks.

Play from 1:07:27 8:88:88 1:07:27 for however you might use it, right? Like sometimes we don't know what's best for us. Exactly. You have to go into little self-discovery, which... I mean, that to me is not surprising at all. In fact, I think the 14 % might sound a little low to me. Yeah, so that leads me into the meta thing, I guess, which is like sort of not meta the company, meta me, which I've been building, which is like, I've deployed it in a few different ways.

1:07:43Building a creative intelligence layer

Play from 1:07:55 8:88:88 1:07:55 It's like a skill that sort of distills all of my life experience and creativity into like a operating system, if that makes sense. Not sure how to deploy it other than for when I'm using creative projects, but it's interesting because I was thinking if there's a future. where is where we get Midjourney deep hours because we're thinking about this. If I if like, I wanted to work on a project with Drew, but I'm not in the same place.

Play from 1:08:23 8:88:88 1:08:23 And we're not going to sit on the phone all day. If I sent my little consciousness, Claude bot with Drew's and then we worked in a space together and then they came back and like that's you know, walked into a room almost temporarily. Yeah. My brain, like the whole open call thing, sort of set my brain when they all like got released into the wild. I'm like, that's cool. Because they've all been so siloed to like our little desktop spaces. Like that's where they stay.

Play from 1:08:48 8:88:88 1:08:48 Right. And they're in the cloud, but they come to us. But then it was like, okay, now they go out in the world. They're interacting with other ones. Like, that's cool. I would like to have my, like the full power of up here, not just like my MD files, like going out into the world. So I've been doing, it's called like this. creative intelligence layer where I've having it interview me for a month and it just asks very broad questions that sort of elicit responses, it's really good, it's kind of like a therapist where it's like, how do you feel about...

Play from 1:09:18 8:88:88 1:09:18 idea. Yeah, that's a neat idea. So it keeps building little micro skills and little micro things from that. There's one thing was the only child layer, or only child plugin, if you want to call it. They asked me why I thought I was creative. I'm like, I don't know. That's a really broad-ass question. And I was like, maybe because I was an only child. when I was playing, I just had to make stuff up.

Play from 1:09:44 8:88:88 1:09:44 it was like, that's like, That's like a very introspective thing to say. Do you think it relates to this? I was like, yeah, dude, thanks for like, you know, hyping me up and gaslighting me. But then it like wrote a whole like operating system based on that thought process. And I was like, cool. And then, you know, I was asking me about like the first time I felt like amped about something. And I was like, I don't know, like when I maybe when I got my car for the first time, because I wanted it for so long, I one of my license.

Play from 1:10:12 8:88:88 1:10:12 But the only thing I ever really wanted as a kid is I just want to drive and get out, like just be free. Right. And it was, like did this whole like dissertation on that and like basically built like little, little skills and like little thought patterns to like feed into these agents. So if I'm like, you know, if there's certain things that he can pull from that, it's cool. And I don't know how else to deploy it right now. I just feel like I keep going. Yeah. it is good though.

Play from 1:10:38 8:88:88 1:10:38 It's pulling the bartender skill out, essentially. like, it gets that out of there. Yeah. like the invisible things that it just, you know, maybe there are some breadcrumbs of observations that it's picked up, sure, but you're kind of providing more background, know, and tangibility to some of those things, which is, and maybe to some degree it's analyzing how you're thinking about it.

Play from 1:11:04 8:88:88 1:11:04 You know, like there's this layer too. I always struggle with these things where, So, you know, like, you know, there's all these tests out there, like, and of course, can't think of any of them, but there's like Meyers, like these personality things, which is good or, you know, as used as those things are. The thing that always stuck with me as a huge problem is this bias of what we're aware of and self aware to answer those questions.

Play from 1:11:31 8:88:88 1:11:31 I mean, how much are those are? aspirational selections that we're making versus actually true objective responses. To me, something like that never values, those tests don't value that. That doesn't come into play at all in most of those, unless you get into maybe more of a situational play. But it's just...

Play from 1:11:58 8:88:88 1:11:58 It's, yeah, and I guess maybe where I'm getting at is like, maybe there's like this element too, that's interesting, that at some point, you know, this gets to right where it can recognize like how fully you're able to communicate something that could be true or not true, right? Or. been like seven of those personality types throughout my life though. Like when I was in sales, I'm different than when I'm doing creative. versus give me the test two years ago.

Play from 1:12:25 8:88:88 1:12:25 Totally, you're gonna get, yeah. Highly like sales, I'm like way more aggressive and like way more like highly logical and like organized or whatever. Then we get to creative, it's like the opposite. It's like I have to be like insular and sort of like in my head and like not out. So it's like, I could be both. It could be today or tomorrow. I could be totally different. think it's a reactive thing. I don't know. I'm just trying to make Claude like me, which is just or at least Claude or Chad GPT or Codex or whoever.

Play from 1:12:54 8:88:88 1:12:54 using just like I wanted to be able to ping ideas around in my neurons the way that I do where it's yeah this is what happened but also let's pull from something like 10 years ago that might be useful and then come back you know some random skill that might be useful right now it's like a you know like a video game right like you're adding XP points or you just like have things in your bag that you might not ever use but when the time comes and need that shovel are you running the analogy skill right now?

Play from 1:13:25 8:88:88 1:13:25 This guy, this guy. to build it. It was like, oh, this is just I do this all the time. I just need to have it Yeah Yeah, I've that probably came from one of these interviews where it was like, know, how do you you know? How do you how are you with persuasion? And I'm like, oh, you know, probably pretty good. Okay, they used some tactics like always analogies. It was like Analogy, it's like I'm like dude.

Play from 1:13:51 8:88:88 1:13:51 I know you're I know you're hyping me because that's what you do as an AI but like Let's build something with that. What can we build? It's like probably the best therapy I've ever had too. I've never had therapy, it was like, I'm having fun doing this, like getting asked this stuff. It's like the best podcast interview of your life kind of thing. talk about it. to me, that's where this is just unbelievable.

Play from 1:14:20 8:88:88 1:14:20 Like the customization of this quote unquote operating system, right? Like we all have the same tool, but it's this blank canvas of what we do with it and to make it ours. And in a lot of ways there is no right or wrong in some ways. And so that's what's really fun about this right now. I mean, I like that creative intelligence layer. It's super interesting. It's a way to differentiate, I would say.

Play from 1:14:47 8:88:88 1:14:47 If everyone's clawed is gonna be really good, and everyone's gonna be operating on the same kinds of things, and it's gonna be producing the same kind of stuff, it's like, what makes us different again? It's us. So like, how I bake myself into a lot of the responses. That's it. Even if it's different in a bad way, right? Like, still different. I'm trying to think of like a world where everything doesn't become homogenized. Right? And it's all the same. Like, it's not this, it's that. This is gonna be a game changer.

Play from 1:15:13 8:88:88 1:15:13 don't even get me started. this, you know what mean? It's like that. But you see it and it's like that's that's that's just one piece of this. there, I was gonna go there, but I didn't wanna add like another 15 minutes as convo because it's like, that's kind of what I'm doing with the writing system. Like to your point, you need the customizable pieces because if you're just, you could say all day, change the tone or this and that, but it's still gonna use LLM writing patterns, still gonna use things that feel like everyone else and feel like nobody at the same time, right?

Play from 1:15:45 8:88:88 1:15:45 It's just like. in order to get to break that, really, it really needs to understand you, your writing style, you know, your examples, your ticks, you know, your slang, the words you use versus you don't use, right? Like all that. And that's where I went down this whole rabbit hole with that because it was just like, as soon as I thought I was in a good spot, I just discovered like there's more and more context I can provide to elevate the consistency here.

Play from 1:16:13 8:88:88 1:16:13 And, you know, I realized that this is a big problem for a reason. because it's just like, requires this really just in depth cleanse of like how it would put these things together. And yeah, just, it's fascinating, right? Because the more you lean into the customization is kind of like the unlock to getting the real value out of this. Because then it's operating with specifics constraints to you, which is ultimately the point that matters the most.

Play from 1:16:44 8:88:88 1:16:44 Well said. I think maybe we just cut it there like that. Or does Mike drop it at that one? I'm good. I mean, we didn't get to see dance, but I think we can revisit that next week when we get into, you know, maybe this other model, mid-journey is gonna drop, we'll see, and then kind of go from there. But hopefully, you know, this was super valuable for you guys. We tried to like go in depth a little bit more on kind of like our thought process behind it.

Play from 1:17:10 8:88:88 1:17:10 What did you think? I mean, I think we kind of like really, hopefully gave some people at least a few things to think about or check to see if they're kind of. thinking about the same way. We're not perfect either. So maybe there's something that we didn't talk about or you noticed that we're not doing right. Dude, point that out. I mean, we're all learning together, so. comment section. The comment section is the reason we have this episode and you know we probably will improve upon the Claude episode as we move along.

Play from 1:17:37 8:88:88 1:17:37 This is just like how do we get stuff out because I think the cool the way that I'm thinking about now is everything is like in relation to Claude so it's like alright that's cool I just heard that how do I do that like how do I how do I like engineer something for that? um like the technology itself of like that customizability operating system that now can be carried over to virtually everything that you're doing digitally, right?

1:18:01William Shatner, Babe Ruth, and better analogies

Play from 1:18:02 8:88:88 1:18:02 Yeah. Like there was there was one last example, cut it off. Someone said this said this out loud the other day. They're like, I heard this great stat. And it was like, you know, William Shatner, the late, you know, William Shatner, and I think he was 95 years old. So I think he's dead. Don't quote me on that. But the stat was like a it was like William Shatner was born. It was like, I think William Shatner was born when Babe Ruth had 147 home runs or something like that.

Play from 1:18:32 8:88:88 1:18:32 was like, like, forget how many home runs Babe Ruth had, but it was like, he was born like, just like it tied that, instead of saying he's born in like 1927, right? When you say like, oh, Babe Ruth had 100 and you know, 100 home runs at that point or something. It's like, okay, there's like a, so I was like, how do I, how do I build that into a skill? Like a cool like way to, right? It's like a very, witty way of saying something or tying an idea together to make it a really quick connection.

Play from 1:19:00 8:88:88 1:19:00 um like that you've got a couple like anchoring things. Like that's an anchoring theme beyond that's sort of like connected across a lot of the things that you brought up today is digging deeper into your personal persona and pulling those things out. And then like this communication layer on top of that, em which super interesting, man. Yeah, by the way, he is still alive.

Play from 1:19:27 8:88:88 1:19:27 He's 95. Okay, he is 95. in 1931. Now here's a hard truth. Says he was married from 2001 to 2020, which means he got a divorce at the age of 89. What? What? What are we doing? Just write it out, baby. Just write it out.

Play from 1:19:53 8:88:88 1:19:53 Let me see if they actually hold on one second. How many home runs did Babe Ruth have when William Shatner was born? Let's see. Babe Ruth had hit, on the birth of what? Oh, sorry, the stat was Babe Ruth hit 149 more home runs after William Shatner was born. That was the stat. like that, like when you say something like that, it's like, that was a really long time ago, you know?

Play from 1:20:22 8:88:88 1:20:22 % of his homers. Yeah. So it's like it's just a very creative way of framing something and I'm like that is really good Like I want to add that like I know kung-fu like I want that to be at all times, So to figure out how to how to sort of like structure that so it can make complex things or like random things More interesting. It's basically all that we do right? That's kind of marketing in a nutshell, but Dude, let's let's uh, what's you know ask everyone If they're still here to like and subscribe, damn it.

1:20:53Outro and comment bait

Play from 1:20:58 8:88:88 1:20:58 And then also, what else? What else? Who else can we tell you to tell? You tell your, you your kids teacher, maybe, uh, maybe the, uh, sanitation department. Um, did I think we did? I think we did Natty and babysitter. Maybe you're, maybe your local, uh, your local coffee shop, you know, worker or did we?

Play from 1:21:24 8:88:88 1:21:24 um You know, it's a waiter bartender Stu, you know air What are the airplane attendants called? I don't even Flight attendants. Yeah them maybe a pilot if you see one they got a lot Yeah, they got a lot of time on their hands when they're in the air That's a good, hey, that is a good, our TAM just expanded.

Play from 1:21:49 8:88:88 1:21:49 Like, oh, that's good. Okay. episodes for you guys to binge. That's like a lot of hours. That's a couple transatlantic flights. You're good. We could say this for off air too, but I just want to bring this up. I was on a call with somebody the other day that somebody had introduced me to and it was just like one of those like you guys should meet kind of thing. You guys are operating in same space. First thing I brought up, we've been watching you and Rory's podcast for a long time, It's just like one of those things, you tell me all the time and I was just like, we love to hear that because we sometimes don't even understand who or how many people listen to this stuff and it's always fun.

Play from 1:22:28 8:88:88 1:22:28 when you just kind of hear that in the wild and you're just like, that's awesome, It is cool, it feels like, you know, it just feels like again, it's like people tell me it feels like they're sitting in, you know, like a bar conversation most of the time, or we're just ripping through stuff. And hopefully we're providing just ideas. Like, I think that's like at a minimum, just like take this and run, do something with it. You know, not going to 37 prompts to make you a great marketer. It's like just take the ideas and go apply it.

Play from 1:22:54 8:88:88 1:22:54 If you're listening for 30 episodes, 40 episodes now, you already know how to do this stuff. So yeah. hopefully we're just providing you with the breadcrumb or the nugget that is the jumping off point for a new idea that you execute on, right? Hell yes. Again. Alright, we'll cut it off there. That was a way to end. to see you always glad we were able to string this together two in a row. We gotta hey one more. They call that a winning streak.

Play from 1:23:22 8:88:88 1:23:22 Do you know that reference Major League? Alright, let's go. Yeah man, this is fun. Glad we got into it guys. Let me let us know in the comments what was the most interesting part. If you had a favorite. If you can add something that maybe we didn't think of or didn't cover that we're doing wrong. That's also high likely. So yeah, let us know and we'll see you guys in the next one. Peace.