16:56 Setting Up Multi-Agent Systems
16:57 files on my system. and pushed up to GitHub, right? >> But those are not necessarily connected.
17:02 So I'm constantly like passing back and forth like here's an artifact that I
17:05 talked about over here. Now let's code it over here. Here's the progress we
17:08 made on the code over there. Let me bring those files into cloud and talk
17:11 about it over here. Like there's a lot of manual like lifting back and forth.
17:16 And my thought is hopefully I can figure out some system where like all of that stuff gets logged
17:23 and stored into like markdown files that live in this memory system that all of
17:27 my agents can access at any given time. And then I can start to have these
17:31 different agents. If I need if I need agents to to work on new features or bug
17:36 fixes in in the background, I can delegate that. if I want them to help me
17:41 draft social media posts and maybe even post them or or schedule them, they have
17:45 access to all this content that they can draw from. >> Yeah.
17:47 >> Um, >> so this is really like the the the idea is that this could be
17:56 >> the true manifestation of this second brain stuff. Like my second brain really is
18:04 here. What I'm confused about is is the context window in OpenClaw actually
18:10 bigger and why is it bigger on an OpenClaw machine >> as opposed to like if you're using a
18:16 model doesn't it? >> No, the model context windows are the
18:18 same. >> Okay. >> Um so what what makes the memory better?
18:23 >> It's a little bit different because like just how the tool works, right? So in in
18:30 let's say cla code when you when you're in a session in cloud code and then you
18:35 get and then you fill up your context window and you hit clear to clear it out
18:38 or or if you close your computer and come back the next day >> and you don't resume the same session
18:44 then it's then it doesn't remember anything. >> Yeah. >> From yesterday. It has to relearn
18:48 everything you you give it. >> Yeah. with this like you still have
18:54 context window limits but everything is getting logged into files that live on
18:58 your file system. >> Got it. >> So, so the the system prompts and and
19:04 this openclaw uh system that that gets installed when you install like it's all
19:08 designed around like it has access to all these files on your file system. It
19:11 literally like like there's like a folder called memory. >> Yeah. Yeah.
19:15 >> Um, and there's a folder called sessions for each agent.
19:20 >> And those are JSON and markdown files that just log everything. And so agents
19:24 can like when you start up a new session, I think in the system prompt it
19:27 has like what you talked about today and yesterday just right off the bat like
19:31 it's always there and then it could it could pull from late, you know, further
19:34 back. >> Yeah. >> Um, >> could you could you replicate some of
19:38 this in claude code? Like I've tried doing this a little bit where I just
19:42 have at the end of a workday, my typical process is to say uh commit and push what we've done
19:50 and I want you to write project notes for today. Unfortunately, right now it's
19:54 it's storing those in claude.mmd, but I've I've wanted to take those out and
19:59 just say every day I just want a journal of what we've done. So could you
20:05 replicate some of this by just saying >> yeah like that that's what I'm sort of
20:09 trying to set up right now. I'm trying to figure out the best logistics for
20:12 that for me and and I've looked at it from so many different angles because
20:16 the challenge is that like >> most of those project conversations I
20:21 want to be using opus >> right which means I want to be using my
20:24 max plan. >> Yeah. >> Which means I have to be doing it in
20:30 claude or cla code. Um, up until now I have only used Claude
20:37 code for coding and I've only used Claude for writing and thinking. Mhm.
20:42 >> And um um it's it's I haven't really figured out a good easy way to like automate taking my
20:52 conversations and artifacts out of Claude and logging them or storing them
20:57 in a file system other than like manually clicking it or remembering to
21:00 do that every single time. >> Yeah. Um, and so I'm I'm kicking around like the
21:06 idea of like maybe I ditch using Claude and and move all of my writing and
21:10 creative and marketing stuff into Claude code. >> Um, which is a possibility, but that
21:16 interface is not it's it's great for coding, but it's not great for like
21:19 non-coding stuff. >> Um, so I don't know. I'm still that's
21:24 that's an open question, you know. Um >> yeah because I think the other thing in
21:30 my head is I wonder if we can bypass this whole open need for openclaw
21:37 with just a new process inside of cloud code which is uh in claude code right now I can give
21:44 it examp I can give it permissions to access other directories theoretically
21:50 and project files. So, I could say, for example, right now I'm in the app
21:54 directory, but I could say, "Hey, I want you to go over to the marketing
21:58 directory. I want us to link these projects." And I've got notes
22:04 >> um in a folder called memory on both project directories. I've got a folder
22:09 called memory and all of my journal entries are there. >> Yeah. Um, and then maybe you could move
22:18 to uh the cloud CLI on your iPhone and then just interact with it just that way
22:23 as well. >> Yeah. And I and I also have been thinking about that is like do I really
22:29 even need OpenClaw? And I've seen other people do this too, which is like
22:33 >> like, okay, OpenClaw launched and it and it is a really
22:37 >> interesting and fascinating idea for putting agents on like their own
22:42 standalone machine, their own box, >> or do you even need OpenClaw? Can you
22:46 just run cloud code or or build your own whip up your own much simpler version of
22:50 OpenClaw? I've seen some people doing that like Christian Jenko was was
22:54 tweeting about that. >> I started to I started to explore that
22:57 for like half a day last week. I was like, maybe I can just do something
23:02 custom spun up. And then I was like, ah, there's all these like nice features
23:05 that I don't want to rebuild, >> especially the the memory system is is
23:09 pretty well thought out. Um, I don't know like that that might be an
23:13 evolution that like I do and that maybe many people will do over the next year
23:18 is like OpenClaw was the first of of many different similar concepts, you
23:21 know. >> Yeah. In the chat right now, they're asking um what about co-work? So, how
23:27 does co-work >> Yeah, that's that's another interesting
23:31 one. Co-work is sort of similar. You can start to get to like co-work is like it
23:36 does work on your Mac file system. I haven't really played with co-work too
23:41 much myself, but my understanding is um >> um I don't know. I I yeah, like that that
23:48 could be a good option. Um, but it doesn't like I don't know that it has all the memory
23:55 stuff the way that OpenClaw is is set up. Like I think it's still like
23:59 taskbased and sort of self-contained. Although you might be able to customize
24:02 it with a plugin and get it to because some of this it just feels like the
24:08 >> it's a process related thing which is again when I'm working with cloud code
24:15 basically I work all day and I'm I'm committing changes throughout the day
24:19 but before I leave for the day I want to commit and push and then I want to
24:24 update cla MD if there's anything important and then I also want to just
24:29 have a separate journal entry three, four project notes. What did we do
24:32 today? >> That's my ideal, but that could be for me a text expander snippet that I just
24:39 use every day at the end of the day in the terminal. Um, so I'm wondering if
24:46 there's like if if for the for the uh lite normies out there like me if if uh
24:51 that's a better system, >> you know, you know, like but like what
24:55 you're talking about is like I spent a day or two o over the past week just
24:59 thinking through that like what's my dayto-day how am I going to change my
25:03 day-to-day workflow >> so that I have more of my conversations
25:09 and artifacts centralized in a file system because right now there's still
25:12 too separated. And like, you know, I've I've also been a longtime Obsidian notes
25:17 user. I have like years of notes in there. Um, which is a nice markdown
25:21 editor and it does save them as files in your file system. So, that's nice.
25:23 >> Yep. >> Except it's lacking Claude. >> Yeah.
25:29 >> Right. Like I can't have Claude help me write my my notes. It's not a
25:32 conversational thing. It's just a notetaker. So I so I'm still trying to
25:36 figure out like how can I and I and then I I was playing around with the idea of
25:41 having a claude skill built into my Claude account so that when I'm having a
25:47 conversation like trigger saving this conversation >> somewhere and
25:53 >> but then even that ran into some some limitation of like it can't like like it
25:58 could maybe save an artifact somewhere but it can't take the whole
26:01 conversation. >> Mhm. Um, I don't know. So, there's a lot
26:05 of like little logistical things, but like ultimately at the end of the day,
26:08 the vision is and so I I spent a lot of time figuring out the multi- aent
26:12 things. Like by default, >> um, Open Claw is a single agent, right?
26:17 Like you're going to have one one Telegram chat with one agent
26:22 >> and I wanted to have multiple. >> Okay. >> So, in that screenshot, um,
26:27 >> I have something like six work agents and like two personal agents, right? Um,
26:32 and the idea is that each of those has a different Telegram >> chat with me.
26:36 >> Okay. >> Each of them has a different I even built like I had Claude come up with
26:41 like personas for each of them. >> Um, you know, like like the grumpy old
26:46 developer, the the junior young guy, the you know >> Yeah. Yeah.
26:50 >> I I use Gemini to to create some like silly like robot like inspired by like
26:55 the gorillas band, you know. Yes. Um anyway, so um so I have I have those and
27:03 like the idea is that I can I can have like different chats like like I have
27:07 one called Vale that's the content marketer. Mhm. >> Um I want to be able to have like
27:13 different Telegram chats and I also connected them to my Slack workspace so
27:19 that I can like have marketing related chats and tasks with this Vale agent and
27:25 development related related tasks with Bernard and you know strategy with
27:29 Clover. Um, >> I mean, I want this for sure. This kind
27:38 um, uh, persona that I can chat with and that I can do work with.
27:46 >> I also think the other the other thing that seems to be resonating about
27:51 OpenClaw is just the fact that you can chat with it via Telegram.
27:57 That interface layer actually solves a bunch of problems. Like for example,
28:01 everything that you dictate into Telegram now gets saved and all of their
28:05 responses get saved. So now you already have a a place to store conversations
28:15 and refer back to them. And I even like the the problem right now I have with
28:21 all this stuff is it's Siri. It's like the the when I'm using my iPhone and I
28:27 got my AirPods in, I the only person I can talk to out of the box is Siri. And
28:34 I can't do anything with Siri except for compose a note. I compose a note with
28:39 Siri and I get it to set >> timers and alarms. That's basically it.
28:42 >> Um, >> totally. And so I think the interface layer is what people are starting to I
28:50 think they wanted a better interface that was better than Siri, better than
28:53 what they were getting on their iPhone. Well, Telegram's easy. I can interact
28:57 with Telegram all day. I can interact with it voice. I can interact. I think
29:02 part of the the attraction of this open claw thing is just the fact that uh out
29:07 of the box you can connect it to WhatsApp and Telegram and Slack.
29:11 >> For sure. Um you're right. It's the interface thing of course like chat GPT
29:16 and claude like the part of the breakthrough with those a couple years
29:20 ago was the chat interface to AI right like GPT2 and one didn't have that chat
29:26 GPT layer yet so like that's why it didn't break through into like GPT3
29:32 >> but um but like so I think like chat GPT and the and the clouds like like
29:37 unlocked like chat is the way to interact with AI and then and then the
29:42 models got a lot better. But I think now with OpenClaw, the next like yes, it's
29:46 the chat with Telegram or Slack or WhatsApp or whatever you want. Mhm.
29:52 >> But it but then it's also the the other connections to everything else you do
29:55 because it has access to your to whatever you can do on a file on a
30:00 computer system. Like like that's the big challenge for me in my in my
30:03 business is like there's so much like okay I finish a I finish a a newsletter
30:08 that I that Claude helped me write but then I copy and paste that into kit and
30:13 then I format it and schedule it and like or you know just a lot of little
30:17 clicking and carrying this piece of content over to there and like
30:21 >> and scheduling it and like >> you know >> I think okay this is I I know we're
30:25 talking long about this but I want to this is actually something I I think was
30:29 also so attractive about all the mythology and legends and stories that
30:34 came out of OpenClaw is what's the challenge with anything? The
30:38 challenge with hiring a personal assistant who's a human, the challenge
30:42 with hiring an employee, the challenge with anything, the challenge with doing
30:47 work yourself is that there's always you're going to run into things, right?
30:51 So, it's like uh like you said, a typical thing would be h what's a very
30:56 common one that AI just can't touch right now? every >> video editing.
31:01 >> Well, well, video editing, but here's a simple one that seems simple. Every
31:05 quarter, my accountant says, "Justin, can you log into your bank accounts and
31:10 export all the PDFs and then send them to me so I can reconcile your accounts?"
31:16 That >> is difficult to hire somebody for because as soon as I hire a personal
31:22 assistant for it, first of all, it's sensitive. But if you get over that
31:27 hurdle, then you have two factor off that often depends on me being close to
31:30 my phone. >> Mhm. >> Then it just goes problem after problem
31:35 after problem. And I think one thing that appealed to people about this open
31:40 claw thing is it's like, oh, this thing will keep going if you instruct it to it
31:45 will keep going in the Ralph Wiggum loop until it's done. Like it will just plow
31:50 through barriers. And I can see why that was attractive to people. And then yeah
31:55 eliminating all these points of friction. So so much of what we do in a
32:00 day is all this connective tissue between like I compose something in
32:05 Claude. >> I might even take it somewhere else and clean it up and write it and then and
32:10 then from there I got to copy and paste it into Kit. and and and also it's like
32:16 for me it also happens when I'm out like I I do a morning walk for 30 minutes and
32:21 I'm doing a voice recording of of my next YouTube video like just the idea
32:25 just brain dumping the idea right >> um >> but you know I do want to speak to
32:31 something else too and this is where it didn't click with me the the day that
32:36 openclaw started to get popular and part of it is because of like the marketing
32:39 copy on the openclaw website and then what everyone what everyone started to
32:43 talk about Yeah, >> everyone is looking at it like this is
32:47 an agent that you're going to give full unfettered access to your own email
32:51 account and your own calendar and it's going to be like as you and do things as
32:55 you and have access to all of you and your life. Like to me like no, like that
33:00 is not what I want. I'm not. And that's that's why at first I was like
33:04 >> I'm not interested in that. I have no desire to have an agent reading and
33:09 replying to my emails for me. I don't have the problem of wanting an agent to
33:13 book restaurant reservations for me. I'd rather pick the restaurant myself. Like
33:17 >> like that that isn't a problem that I have. I'm not interested. That was my
33:21 that was my initial reaction, right? >> Yeah. >> What made it click for me was like,
33:28 think about this exactly like an employee or maybe a team of employees.
33:32 >> Yeah. >> So, the way that I'm doing it is I I did
33:36 set up this Mac Mini. It's it's its own user. It has no it's like I didn't log
33:41 into my Apple account on it or my iCloud or even my Dropbox account. It has it's
33:45 not connected in any way from there that that standpoint. >> I actually gave it its own email
33:49 address. >> Mhm. >> So I can invite that email to like
33:55 >> was that just a Gmail address or like um >> I used a proton
33:58 >> by me. >> Um and uh and so I like I used that to
34:03 to set up like a GitHub account for it so that so I could see like commits from
34:06 it. M >> um I can I can grant and revoke access through that.
34:11 >> Um >> did you give it its own credit card for
34:14 spending? >> No. Like like that's I mean I don't know like maybe at some point I I but I
34:21 haven't done that like I haven't had the need to do that right. Um,
34:26 it's just like what? So, if you think about it, like it's not it's definitely not the same,
34:34 but when you hire an employee, you are inviting them to to do things and have
34:37 access to things. >> And if they're if they're an employee
34:40 like in your company, you're probably going to buy them a a computer to to
34:45 work on, you know? Yeah. Um, it's kind of simil if you start to think about it
34:49 like that. You start like so for for me I I think about it like like my company
34:55 is me and a part-time video editor works for me. >> Mhm.
35:00 >> Uh other than that like I don't have clarity flow is separate. I have a
35:03 full-time developer who manages that. >> Yeah. >> But
35:06 >> in builder methods it's me and a video editor >> and uh I don't have any developers other
35:13 than me and and coding agents. I don't have a virtual assistant.
35:16 >> I probably would if if this were a couple of years ago, I would have hired
35:20 a virtual assistant by by now in my business >> for all those little tasks, but I'm
35:26 convinced that a lot of that stuff can and should be processed by agents and I
35:31 haven't figured out all of the processes, but it's absolutely possible
35:34 to to do that. >> Yeah. >> And then same thing with like content
35:38 and marketing. Obviously, like I'm do I'm the content person, but like I I
35:43 heavily use AI in my creative process, you know, like like the idea of like
35:47 outsourcing to writers is not a thing anymore, >> you know? Um,
35:51 >> and even like like I'm going to I'm going to do like ads funnels at some
35:54 point. Like >> I'm not going to hire I'm not going to hire like a pay-per-click expert for
36:00 that. I'm I'm going to like have like AI prepare a strategy and a game plan and
36:04 maybe I'll like do the initial setup of it, but it it'll be a process that can
36:07 be automated, you know? >> Yeah. I mean, this this is just such an
36:16 I mean, it feels like when it comes to hiring, there's a few different jobs to
36:22 be done. So the boss sometimes is hiring because it's like I need someone to take
36:29 this work off my plate and this is work that I'm is coming to me and being directed at me
36:37 or is could just come to an email and is being directed at email but somebody
36:40 needs to go there and go okay the accountant needs this so right now that
36:44 comes to me and then I could theoretically delegate that to somebody
36:48 else say hey this needs to be done they're basically just taking it off my
36:53 plate. Um, and eventually they could be the point person for that. Listen, I
36:58 could tell the accountant, you just email this person, they will take it
37:03 from there. So, that's one form of work. And that's still, I think, um, you know,
37:09 a big part of hiring, but I could see how uh I could see how AI, you might
37:14 hire it to do some of that work, right? It's just like, hey, I just need this
37:18 taken off my plate. the the one thing that you're going to miss still is hiring someone who can be
37:28 autonomously creative and proactive. Uh because it feels like the these AI
37:33 agents are still very reactive. They're always just taking direction, right?
37:37 They're not >> um they're not able to come up with their own ideas for things that you
37:43 might do or opportunities or or whatever. Um, so yeah, I I mean I don't
37:47 you don't need to respond. I'm just thinking that off the top of my head.
37:51 >> That's that's another big challenge that like I I run a creative business. It's a
37:55 creator business, right? Like it it >> it's mostly built on like me creating
38:00 ideas for for the YouTube channel and courses and and stuff. Um I'm also
38:05 thinking a lot about like how can I streamline what I do before I record a YouTube
38:12 video? like the ideiation process. I And I can't and I don't want to rely on AI
38:16 to come up with ideas. Like there are people out there who will just scrape
38:19 YouTube and just do like copycat videos. I'm not I'm not interested in that.
38:22 >> Yeah. >> Um >> but I do want to be able to take my raw
38:27 ideas and not spend a whole day sitting here like staring at the screen, but
38:32 have more of a process where I could have a chat with my creative agent
38:37 >> with with Opus. >> Yeah. >> Right. I mean that this is this is what
38:41 I do with Opus now is like I start with a raw idea, an observation, an insight
38:46 or something I want to talk about and I'll brain dump a voice memo in it and
38:52 and then I have a a long back and forth of of creative like fleshing it out and
38:57 then and then writing it and crafting it, you know? >> Yeah.
39:00 >> Um but like that's still like too messy for me. I still want to be able to like
39:05 like I'm not publishing fast enough because I spend too much time in that
39:08 creative loop, you know? >> Yeah. Um, I mean, this is this is
39:18 there's still just a part of me that I I think what's attractive about
39:21 OpenClaw is the same thing that attracts me to the idea of getting an executive
39:27 assistant is there is just a bunch of stuff in my life that it would be great
39:32 to have that. First of all, it'd be great to have someone else be able to
39:37 delegate the work. That's one piece. But even better, a step above that is
39:43 somebody who is taking um who is taking responsibility for that
39:47 thing and is just doing it in an ongoing way. And here's another example I can
39:54 think of. It's like like if if I put AI in in charge of spider control at my
39:59 house, how would it do that? I don't I don't know. Maybe it would detect like
40:04 number of screams per day or something like that, you know, like
40:07 >> my wife screams, "Oh, that's probably a spider." So, like
40:11 >> number of screams per day. >> Um, but the the truth is we hired a
40:18 spider control guy, right? And I had to like it was like my wife is like,
40:22 "There's way too many spiders. We saw a bunch of black widows. Justin, do
40:26 something about it." So then I take that I call up a spider control guy. I say,
40:29 "I want you to take control of this problem." >> So, I had to do that initial step. He
40:34 comes in, he scans the whole house, he does it, he sets up traps, and then he
40:39 says, and personally, this is like my love language. He says, "And now I will
40:44 come back and check every year and just see how we're doing." Perfect. Now, I'm
40:49 not responsible for spider control anymore. This guy's just going to show
40:53 up on a regular basis. He's going to check all the traps. He's gonna do
40:58 everything he needs to do and then he's gonna bill me. That's kind of ideal
41:04 that this is these are the real problems people are trying to solve and AI might
41:08 be part of the solution, but then I'm also thinking how the human element with
41:11 all this stuff. How does it all interact? you know, like >> Yeah. I mean, the way that I think about
41:19 it also is like um like I don't I'm not really looking to delegate the
41:29 the uh ideiation and creation of new projects. >> Mhm.
41:33 >> To AI and frankly even to people like I I am a solo founder of this business.
41:38 Uh, I'm the creator of it and anything that I create, a a new product, a new
41:46 video, a new course, a new anything, that's what I want to be spending my
41:50 time on. Like, like I chose to do this business because I like yesterday I
41:53 spent the whole day >> creating, designing this app to run my
42:00 agents. That that screenshot is is a whole app that that took me about it
42:03 took me one full day to design it and build it. Mhm. >> Um and uh you know it it tracks their
42:09 tasks, it tracks the token costs, it tracks activity, it it does a lot of
42:13 things in there and and it hooks into the file system. It's pretty cool. Um
42:19 but like I that there are people out there who are tweeting about like how
42:24 they just task their open their OpenClaw agent to go build build a similar
42:27 dashboard like that. Just go go do it, figure it out, do it. Like to me, I
42:31 can't I I I want to be in the nitty-gritty. Like every UI choice in
42:36 here was me directing it or crafting it in the way that I want it, right?
42:40 >> Um especially if it's a tool that I'm going to be using.
42:42 >> Yeah. >> Um so the way that I think about agents is
42:48 more like I want them to be the next phase. They they are the maintainers,
42:53 right? like like when I have a new feature for this or for builder methods
42:59 or for whatever like I should be able to voice note it into a chat
43:04 >> and it turns into a spec and if it's like s like small to mediumsized like
43:08 that's something that an agent can and should be just creating a PR for
43:13 >> um you know if it's a totally new redesign or a totally new concept then I
43:17 want to be involved in that then I'm going to do that myself in in cloud code
43:19 or whatever. >> Yeah. Um, but if but if it's like fixing
43:26 a bug or or doing this or or like if it's a marketing process, right? Like if
43:32 I have a like um like there there should be some idea like some creative stuff
43:35 where it's like it has access to all of my activity, all of my projects through
43:40 these through the file system >> and every week like it knows that I need
43:44 to be generating new ideas. So maybe it surfaces like a list of like here are
43:49 some interesting uh p like concepts that I just pulled out of your recent GitHub commit or your
43:56 recent podcast transcript. Like these could be tweets maybe >> bookmarks or like the these could be
44:01 like >> maybe something that I said at 42 minutes into the recent panel episode
44:08 could become >> could become the basis of my next YouTube video. I don't know. But like
44:13 there there's so like to me the the big one of the big drivers was like I there
44:18 is so much activity that happens >> Yeah. >> between me and this computer and these
44:23 and these podcasts and the code >> that >> and I'm I'm not I'm not tweeting all the
44:27 time. I'm not like >> Yeah. >> I'm not broadcasting everything that I'm
44:32 doing. I'm broadcasting maybe 5% of what I'm doing. >> Yeah. you know, um, and so I if I could
44:39 figure out some systems to like extract more publishable content because I feel
44:42 like there's a lot of good material. >> Yeah. >> And and and sometimes I've done that and
44:49 and it's like it's like um I start to do that and then I'm like, ah, this is so
44:53 raw now I want to spend an hour crafting it and refining it and now I'm not
44:56 actually getting any real work done. So if there's just some sort of process
44:59 that's happening in the background like watching and extracting ideas and
45:03 turning it into at least pretty good draft that I can approve,
45:09 you know. >> Yeah. I I mean I yeah I I'm I'm curious to see how this uh this experiment I I'm
45:16 kind of waiting for everyone else to figure out this stuff and then but the
45:21 parts that appeal to me are you know the the Christian Genko like um uh talk to
45:26 my phone and because right now you know I I I can't interact with Claude code on
45:32 my phone at all but theoretically it could be doing all sorts of work for me
45:37 while I'm just out doing other things. So, yeah, I'm I'm curious about all of
45:41 that. >> Um, let's uh let's switch it up. I want to talk a little bit about building fake
45:47 products. >> So, you asked me, hey, why why aren't we using that that podcast countdown timer?
45:56 >> And I I feel like this was always true, but now the cycle times are so much
46:00 shorter. The value of build first has just increased dramatically.
46:09 Build first allows you to create something in a couple of minutes and start using it and
46:17 start uh basically interacting with reality right away. So the podcast timer
46:25 countdown timer is a perfect example. In that app, we set up segments, we give it
46:31 a timer, it counts down. As soon as I introduced that with you and I, it it
46:36 added I noticed it added a little bit of anxiety. It's like, oh man, I got time
46:42 now counting down. So, you start to feel a real interaction with
46:49 the product right away. Um, one of the things I've been talking
46:56 about on this show forever is can we at Transistor figure out a way to do video
47:01 podcast hosting? And it's just such a big hairy problem. It it feels like too much. And then a
47:10 couple weeks ago, I said, I'm just going to start building it and see my beer
47:17 >> how it feels. And not not because I want to build something that's production
47:20 ready that we're going to just drop into the app, but because for this very
47:24 reason, I want to try start to experience both from the technical
47:29 implementation side, but even more as a user, what do I run into? What do I
47:36 want? What do how does this feel? And in the past, you just could not do this.
47:39 Just to get to the point I'm at right now, I built a full video podcasting app
47:47 in a couple days, the first version. And just to get to this point with pretty
47:52 talented engineers would have taken 6 months, a year, maybe even two years.
47:58 >> It would have taken a long time. But I got I've got a little app running.
48:05 >> I mean, I and uh sorry, go ahead. What does this do like like uh for those who
48:09 aren't watching they're listening you just started to show this thing that it
48:13 looks like it's called Ampcast. I don't know if that's a temporary name.
48:16 >> Yeah, that's just a temporary name. This is never I don't I don't think this will
48:20 ever be a real product. This is just a prototype I've built.
48:24 >> So we've got show a list of shows. You click into one.
48:27 >> You can connect it to Transistor and import an existing show. So let's say I
48:34 import restaurant success here and then I can connect it to YouTube
48:40 and it that works. I can I've got all the off and connection to YouTube and
48:46 everything working and it again I I >> what makes it a I'm curious like what
48:50 does it make what makes it a video podcast? Well, this is this is what I'm
48:57 this is what was so helpful is previously I thought the the way to do this is we
49:04 have to do everything. We have to allow people to upload the video to us. We
49:09 have to encode it for like use HLS encoding, host it, etc. As I got into
49:18 this, I started to realize that what people actually want is not that what
49:26 they want is let me see if I can if I can find this thing here. like turning
49:29 their YouTube videos into a >> basically they want to upload a video,
49:36 have it syndicated to YouTube and maybe Spotify video and then create an audio
49:41 only version that syndicates to all the audio platforms. >> And it wasn't until I um it wasn't until
49:52 I started working on this that this became very clear. Right. So, we've been
49:59 talking I think you know HLS maybe should exist for the nerds but when I
50:04 think about 80 90% of people I talk to who are doing podcasting what do they
50:07 want? >> They want to upload one video file they want it to go to that to get get
50:16 uploaded to Spotify and YouTube as a video episode and then they want it
50:21 transcoded and as an audio file and added to transistor. That's what 80% of
50:26 people want. >> Totally. >> And so I didn't experience that.
50:33 I didn't fully realize that. That was in my brain, but I didn't fully realize
50:39 that honestly until uh I'll see if I can find it. But actually when I was working
50:47 through uh Design OS, I had it create a uh breadboard for the different flows.
50:53 And once I saw the breadboard, I was like, "Wait a second." And once I'm
50:58 talking to Claude Code and it's like, yeah, like that HLS video component
51:04 really adds a lot of complexity to this app. >> Mhm. >> Um,
51:09 that's that's like very similar to my process with any new new app. It's like,
51:13 all right, like here's my concept. I think I want to do it this way. And then
51:16 and then you run into things and have back and forth with Claude and
51:18 >> Yes. >> Yeah. And then once you see the see the
51:21 views see and like yeah >> well and there's just so many things
51:25 that you run into that until you're actually using it usage has always been
51:32 oxygen for product realizations for actually understanding something.
51:37 >> So here's another example. So, >> does it >> I was wondering like does it push out to
51:42 like if do would you upload the file to transistor and then that actually
51:47 uploads to YouTube and Spotify and >> in my little prototype what this does is
51:52 you just go here I'll I'll do one right now. So, let's just make sure this is
51:58 connected to YouTube. So, I'm going to I have to keep re-offing it because I'm still in test
52:05 mode. But for the listener, all I've done is I've opened up a fake show
52:08 called Super Awesome. I'm going to click new episode. I'm going to choose a video
52:14 file and it's starting to upload. That uploads to uh R2. I'm going to say
52:24 sample app for Brian. And uh the other thing I like about this, by the way, is it's allowed me to
52:32 play with a simple interface. Again, I even hid a lot of the advanced settings
52:37 in this expand collapse UI here. >> I'm going to click publish episode.
52:42 >> This is now uploading to YouTube. That's completed.
52:48 Now it's extracting the audio. That is processed and completed. If we click
52:53 here and view it on YouTube, still processing the video, but that will be
52:56 live right away. And you click on transistor. Episode is live. The YouTube
53:02 embed has automatically embedded. Oh, you haven't. You can't see that. YouTube
53:06 embed has automatically embedded in the share page. The episode is playable. It
53:13 all happens in one step. >> Nice. >> So, >> that's that's beautiful.
53:16 >> That's >> I mean, >> and and here's the other benefit. I've
53:20 been playing with this. I've been using it. >> Especially if you could you could
53:23 schedule it. >> You can schedule it. I've already got that built in. the I'm now showing this
53:31 to people. So, my one buddy, he he does uploads all the sermons for his church
53:35 and he's like, "Dude," he's like, "I wish I could just do this in one step.
53:38 It's taking me so much time to first upload it to YouTube. Then I basically
53:42 copy and paste the same title, description, etc. over to a podcast host
53:47 and then do it there. I just want it one-stop shop." So, I I can
53:53 record a video for him. Eventually, he'll be able to use this. And now I'm
53:57 gonna have a real person using it and will experience how does this work.
54:03 >> Today for the first time I I had to we at Transistor right now we don't have an
54:07 uploader that can handle uh files bigger than a gig. So I had cloud code build a
54:13 file upload file uploader that can handle big files. today for the
54:20 marketing for developers podcast. >> This Lars Laughof interview was uh 15
54:25 gigabytes or something like that. It's like okay well let's let's see how this
54:31 system handles a 15 gigabyte video. I'm also going to be tracking the costs now
54:36 on you know the different like for hosting and for bandwidth and all that
54:41 stuff. I can test all this stuff out before we implement it in our production
54:47 uh product >> and I think building fake products is going to it people need to be doing
54:57 this. This is this is the way to explore and truly feel and shape the entirety of
55:04 the problem before you sit down and write a a you know a implementation plan
55:08 for your existing app. This is like let's figure out everything now. Even
55:14 even something as small as like you'll notice this one says private. This is a
55:18 private podcast. And then I as soon as I start using this cuz I wanted to publish this episode
55:25 today. I'm like wait a second. A private podcast is going to have to have a
55:30 different setup than a public. And I'm like what would I want personally as a user? Well, I might
55:37 just want it to be an unlisted playlist, an unlisted video on YouTube. Can I do
55:42 that? Yeah. So, now if it's a private podcast, it automatically creates a
55:46 unlisted playlist on YouTube, publishes an unlisted video. >> I would I wouldn't have thought of that
55:51 beforehand. >> And all of this reminded me of this great story from Fresh Books. This is a
55:59 story that Mike McDermott told me ages ago. And I don't know if a lot of people
56:03 have heard about this, but basically they decided they they they had this
56:08 problem back in who knows when this was 2017. They had this problem where they had
56:12 this mature product, but they wanted to do a full rewrite. And so they decided they went through a
56:20 bunch of ideas, but then Mike said, "What if we just created a new company
56:24 and competed with ourselves? A new company could have its own name, brand,
56:27 logos, website, articles of incorporation, user agreement, service
56:31 staff. we could use that to figure out if we could build a product that is
56:34 truly better than the one we're offering. They built for so FreshBooks
56:40 V2 or V3 or whatever it was. They created an entirely different company
56:46 and then they built the product ground up from there. Got real users using it,
56:51 had its own pipeline and everything. And then only when it had proven itself with
56:56 whatever it was, 100 users, 100 paying customers, did they switch it over and
57:03 have it uh have it as the main uh FreshBooks app. And then >> I didn't know that they did that. That's
57:06 cool. >> Um, >> yeah, I think >> really cool. I like this what you're
57:12 describing this whole this whole process of like >> you get bringing a new feature uh at
57:19 least in a in a concept or a prototype form like it it it dovetales exactly
57:22 with what we were talking about with with Jordan on the on the recent episode
57:27 over here a couple weeks ago. It's like I'm I'm curious to know for you with the
57:31 with your team at Transistor like what is how is this going to change the
57:36 process of bringing new features into Transistor? Like I I just uh this
57:39 morning I recorded my next YouTube video and it's about the Claude integration
57:44 with Slack. >> Um and I at first I was like that's okay. It's just a little connector with
57:50 Slack. That's kind of a nice to have. What's the big deal? And then I saw one
57:52 of the >> one of the cloud um team members post an article on on X about it and how they're
58:00 using it in inside cloud and I realized like oh there there are some real
58:04 workflow breakthroughs here and one what what you were just describing about like
58:09 okay transistor is its own thing. It's it's it's wellestablished codebase. It's
58:13 got this whole user base on it like >> um you can't just like change change
58:17 course or ship features like in a day like you can with a vibecoded thing.
58:21 >> Yeah. But but what they were doing what they're doing at Anthropic apparently is
58:27 in Slack you can like with the Slack connector to Claude code
58:32 >> anyone on the team marketers product people go to market sales
58:38 anyone can at message claude and kick off a claude code project and then cloud
58:44 code goes and and works on it and creates a PR to the actual codebase.
58:47 Mhm. >> So like so that's an interesting product process workflow for teams,
58:54 right? So it's like >> um instead of vibe coding a prototype
59:00 which is very different and new and shiny and and separate from what your
59:05 actual codebase is like anyone can pull off a branch and >> and vibe and essentially vibe code and
59:11 not even vibe code but like at least just start like like what what if we add
59:16 a video uploading feature to our existing codebase? What would that look
59:20 like? And and it and and could we achieve this? Is that even technically
59:23 possible? anyone like obviously it'll depend on your on your own team's
59:31 like strategy and workflow but but you or anyone on your team could
59:36 like essentially at messagecloud and at least start a PR this isn't going to
59:39 ship tomorrow but at least it's something that it's like oh this is
59:42 technically possible and and like it it could or or we will run into these
59:47 technical challenges of it like now now we understand that and then the
59:50 engineering team can take a look at it and it doesn't even necessarily have to
59:54 be like a code review challenge. That could just be the the V1
59:59 >> of this or or or like a draft version of this feature so that the engineer can
60:03 see like, oh, okay, >> I see what I see what the product team
60:06 is going for here. Yeah. >> Let me rebuild this properly with so
60:11 that it's ready for prime time. But like that's a much higher fidelity
60:16 >> vision than like a wireframe or even a vibecoded prototype that's separate.
60:19 like if it's a PR off of your real code base, you know, >> I think we're going to do both. I mean,
60:24 we're still stepping into this carefully. >> Yeah. >> Um I think we're going to do both. For
60:29 this particular project, it made sense to do it separate because we needed a
60:33 separate R2 bucket. I don't want to use what Transistor already has. We There's
60:37 a bunch of things that I just wanted to keep separate because this is involves a
60:42 lot more moving pieces. Um >> so you're so you're going to rebrand to
60:45 AMCast. Um, I mean it there is a world in which we go, you know what, we don't want to
60:52 add this to Transistor. We're going to rebuild it as a separate thing. Uh,
60:56 maybe a video first podcasting app. That could be true. Uh, I I think also the
61:06 being able to to have something that I can hand over to my friend who's working
61:11 at this church um, and have a limited subset of people actually using this
61:16 workflow and just running into everything that could happen with us
61:22 with usage, taking all those notes and then using that to implement that in the
61:27 main app. I think that's the other piece. Um, and then the other thing is
61:31 just being able to see um, think I've also connected it to the
61:38 YouTube analytics API and the transistor analytics API and so now I have a a a
61:45 view of uh, YouTube views versus audio downloads. >> Sweet. Right. And so, um, being able to
61:53 do that in its own space and show it to people, show it to different users, uh,
6:55 Exploring OpenClaw and Agent Management
6:56 an autonomous agent or agents to run with scissors on a machine often
7:02 accessing claw code but it could be any model and it basically takes away all
7:06 the guardrails and uh or you can take away a lot of the guardrails. I mean at the basic level
7:13 you know how like when you run claude code you say like you you you claude
7:18 code are operating in this project directory for my project.
7:22 >> Mhm. >> It's essentially like you open claw you're you're operating like that except
7:30 you're operating on my machine like at the machine level. So it there there
7:34 there are different configurations you can do but but out of the box with with
7:39 no guardrails you are basically giving it access to your entire machine. So
7:43 like for like rule number one definitely just don't just run it on your own
7:46 machine >> like the the one where you do your work dayto-day or has access to your files
7:53 like just don't do that right and I think everyone hopefully everyone knows
7:58 knows at least that so then the next the next thought is like okay let's give it
8:03 its own dedicated machine like that's that's what I did a lot of people went
8:07 out and got a a Mac mini of course even more people this is also a valid
8:11 approach it seems is to is to get like a VPS account like Hets Hner is a popular
8:15 provider. Five bucks a month you can get a server in the cloud. Yeah,
8:19 >> I think Cloudflare even like the next day started offering like a
8:22 >> digital ocean too. One click installs on all these things. Uh so essentially it's
8:28 a box in the cloud where you can run um cloudbot you you could also do things
8:32 like docker like put it in a docker container and >> um I so so what I did was I I I looked
8:40 into all those options and I was like >> just the the VPS thing like I know it's
8:45 a lot cheaper than a than a $600 Mac Mini but I don't know I like I that I I
8:53 I still just like the old school idea of being being able to like see my Mac Mini
8:57 and like being able to remote into it and and manage it that way rather than
9:01 like remote into a cloud server through the terminal. I don't know.
9:04 >> There's just a lot of little clunky things about that that I didn't like.
9:08 And for I don't know for me like 600 bucks didn't seem like a big deal to to
9:13 and also like if if I end up using it as extensively as I plan to although I'm
9:18 not yet but if but theoretic I do have a business strategy case for this that I'm
9:22 thinking through if if that comes to reality >> then I'm definitely going to use more
9:29 >> uh services than the $5 a month >> like that's right like like the memory
9:33 and storage is definitely going to increase to 50 bucks a month and then
9:36 all of a sudden it's like you might as well just get a Mac Mini. It would have
9:38 been cheaper with a Mac Mini, you know. >> Yeah. You want you want to give it some
9:42 power. So, just for folks who haven't heard any of these stories that are
9:44 ringing around, again, some of these are fantasy, some of these are myths, but
9:49 here's the one I heard Ben Ornstein tell, which is his he said his friend,
9:53 this is how I understand it. His friend set up OpenClaw on a Mac Mini, gave it
9:59 access to many things basically unrestricted, talks to it via Telegram or WhatsApp and
10:06 says, "I want you to book me a haircut." So, it goes out and tries to book him a
10:10 haircut on the web. It runs into a capture, which automatically it can't
10:16 continue, but this person had given it instructions to keep going until it
10:22 solved the problem. So according again this could be apocryphal but according
10:28 to the story openclaw then decided to sign up for or use an 11 labs account
10:34 synthesize a voice used the Twilio account that was already there or some
10:38 sort of way of calling called the barberh shop booked an appointment using
10:43 this synthesized voice and then reported back to the person. So a lot of the
10:48 stories are of of open >> I hung out with Ben at Big Snow last
10:51 week. Okay. >> Um, and he was one of the him and like another guy, uh, Charles
10:57 >> were like the open claw like hype guys. Like they were into it like
10:59 >> Okay. >> So, and so it was like hanging out with
11:02 them and coming home and I was like, "Okay, I got to actually look at this."
11:05 Because at first I was like very skeptical. >> At first I was not so skeptical of the
11:11 security stuff. At at first I was just like, "I don't I don't really get what
11:15 the use case is for this." Like >> like why is this more interesting than
11:19 just running everything that I can do with like cloud code? Yeah.
11:24 >> Um uh but it it it took me about a week to realize like oh this this for me
11:31 personally this could solve um some real challenges that I have in my business I
11:33 think. >> Yes. >> That are actually different from what I
11:36 do with cloud code. >> So what have you done with it so far? So
11:40 you set up >> so like I I don't know like I I I really
11:44 nerd out on it. So like, but I'm not actually you using it like for real work
11:50 just yet because I've been so consumed in >> setting it up and learning about it and
11:55 learning about the security configurations and how I and and
11:59 planning out my workflows and setting up this multi- aent setup which I could
12:02 talk about. >> Um that I've just been in like builder setup research learning mode.
12:08 >> Yeah. >> Uh I feel like I've turned a corner last
12:11 week and and that was the tweet this morning. Um, so I think I'm ready to
12:16 start to really fire it up. But another big open question we should talk about
12:22 is the costs of token usage and the >> because one one learning which was okay
12:32 when you initially set up OpenClaw, it's going to ask you do you want to put
12:37 in an API key for open AAI or anthropic or any of these other ones
12:40 >> or do you want to connect your Claude Max account? Mhm.
12:45 >> That that is available like it's very easy to just a authenticate your Claude
12:50 Max plan, your subscription plan. >> Yeah. >> But that is against Claude's terms.
12:56 >> This is what I've heard. This is what I've heard. >> And so, but and like you know the costs
13:02 of just running tokens, especially if you're using Claude Opus, um are just
13:09 extremely expensive, right? So, I was going deep over the last week just
13:13 figuring out like, okay, if I'm going to operate this the way that I want my team
13:17 of agents to operate, I really need to figure out a strategy around like how am
13:21 I going to optimize the token usage because literally just in the first like
13:24 two days of tinkering with this and playing around with Telegram chats, just
13:27 a few chats, like figuring out configurations, like I rung a a bill of
13:32 like a hundred bucks in like a few hours of chatting, you know. Um, so like that
13:38 like clearly that's just not going to be sustainable. Um, I I've landed on a
13:42 couple strategies. I've I've looked at it from so many different angles. I like
13:46 at first my my initial thought was like maybe I'll just cancel my max plan and
13:51 just go all API tokens everywhere. But that's clearly just going to be way too
13:56 expensive. Yeah. On ongoing. >> So I think I am going to stick with my
14:00 max plan. And I'm and I'm trying to not break the rules. Like I'm not going to
14:02 use my max plan. >> Yeah. with open claw there there definitely are stories of claw of
14:09 anthropic shutting down like banning accounts. >> Uh I think as of actually January they
14:16 rolled out some sort of detection of like if if you are running a bot to
14:20 operate a cloud code instance like they can detect it and they can they can shut
14:23 your account down. So I definitely don't want that to happen.
14:29 >> Yeah. Um so uh uh but you know I do heavily use Opus 4.5 uh for a lot of my
14:36 hard hardcore work like coding and writing and that's really important to
14:38 me. >> So a lot of that I'm still going to use do myself like I I have no plans to stop
14:46 creating on my machine with cloud code with Opus like uh that's going to
14:51 continue. Um, when it comes to chatting with my agents through OpenClaw, I have
14:57 them now all defaulting to a to a cheap model like defaulting to like haik coup.
15:01 Um, I I might do one of the other ones. >> Okay. >> Um, but then you can have them like
15:07 delegate to sub agents. Uh, so you can have their sub agent use opus. So it's
15:12 like I will spend so I I expect I'll be spending for both like API tokens and
15:19 cloud max plan and um like the important activities that I want the agents that I
15:23 want to delegate to agents they can use opus on like a selective basis.
15:28 >> Um but day-to-day like chatting or low-level like admin stuff doesn't need
15:33 to use opus that can use cheaper models. Um, but then I'm also trying to figure
15:37 out like how can I how can I like because the the big benefit the big
15:42 reason why this is different from a cloud code or cursor or is the memory
15:48 aspect. >> Mhm. So this is this is where I think it could be a really big unlock for me in
15:55 my business is that like the theoretically the idea is that I want
15:59 all of my agents to have access to the same brain, the same memory. And that means like every
16:08 project we ever work on, every every blog post, every YouTube I create, every
16:13 code project, every commit that I make, every um podcast that I record, every
16:18 transcript of those podcasts, I want all of that flowing into this file system
16:24 that all of my agents can read at any at any time. >> Yeah.
16:28 >> No, no matter what, right? And I and up until now, the past month, the past
16:33 year, I sort of have that, but there's still a lot of friction, especially
16:36 between Claude on the web, which I use a ton for writing, for thought partner, strategic
16:46 planning, creative thinking. I I constantly go to Cloud to help me think
16:49 through stuff. So, I have pages and pages of strategic conversations and
16:53 artifacts in cloud. >> And and then I have Cloud Code, which is
16:57 files on my system. and pushed up to GitHub, right? >> But those are not necessarily connected.
17:02 So I'm constantly like passing back and forth like here's an artifact that I
17:05 talked about over here. Now let's code it over here. Here's the progress we
17:08 made on the code over there. Let me bring those files into cloud and talk
17:11 about it over here. Like there's a lot of manual like lifting back and forth.
17:16 And my thought is hopefully I can figure out some system where like all of that stuff gets logged
17:23 and stored into like markdown files that live in this memory system that all of
17:27 my agents can access at any given time. And then I can start to have these
17:31 different agents. If I need if I need agents to to work on new features or bug
17:36 fixes in in the background, I can delegate that. if I want them to help me
17:41 draft social media posts and maybe even post them or or schedule them, they have
17:45 access to all this content that they can draw from. >> Yeah.
17:47 >> Um, >> so this is really like the the the idea is that this could be
17:56 >> the true manifestation of this second brain stuff. Like my second brain really is
18:04 here. What I'm confused about is is the context window in OpenClaw actually
18:10 bigger and why is it bigger on an OpenClaw machine >> as opposed to like if you're using a
18:16 model doesn't it? >> No, the model context windows are the
18:18 same. >> Okay. >> Um so what what makes the memory better?
18:23 >> It's a little bit different because like just how the tool works, right? So in in
18:30 let's say cla code when you when you're in a session in cloud code and then you
18:35 get and then you fill up your context window and you hit clear to clear it out
18:38 or or if you close your computer and come back the next day >> and you don't resume the same session
18:44 then it's then it doesn't remember anything. >> Yeah. >> From yesterday. It has to relearn
18:48 everything you you give it. >> Yeah. with this like you still have
18:54 context window limits but everything is getting logged into files that live on
18:58 your file system. >> Got it. >> So, so the the system prompts and and
19:04 this openclaw uh system that that gets installed when you install like it's all
19:08 designed around like it has access to all these files on your file system. It
19:11 literally like like there's like a folder called memory. >> Yeah. Yeah.
19:15 >> Um, and there's a folder called sessions for each agent.
19:20 >> And those are JSON and markdown files that just log everything. And so agents
19:24 can like when you start up a new session, I think in the system prompt it
19:27 has like what you talked about today and yesterday just right off the bat like
19:31 it's always there and then it could it could pull from late, you know, further
19:34 back. >> Yeah. >> Um, >> could you could you replicate some of
19:38 this in claude code? Like I've tried doing this a little bit where I just
19:42 have at the end of a workday, my typical process is to say uh commit and push what we've done
19:50 and I want you to write project notes for today. Unfortunately, right now it's
19:54 it's storing those in claude.mmd, but I've I've wanted to take those out and
19:59 just say every day I just want a journal of what we've done. So could you
20:05 replicate some of this by just saying >> yeah like that that's what I'm sort of
20:09 trying to set up right now. I'm trying to figure out the best logistics for
20:12 that for me and and I've looked at it from so many different angles because
20:16 the challenge is that like >> most of those project conversations I
20:21 want to be using opus >> right which means I want to be using my
20:24 max plan. >> Yeah. >> Which means I have to be doing it in
20:30 claude or cla code. Um, up until now I have only used Claude
20:37 code for coding and I've only used Claude for writing and thinking. Mhm.
20:42 >> And um um it's it's I haven't really figured out a good easy way to like automate taking my
20:52 conversations and artifacts out of Claude and logging them or storing them
20:57 in a file system other than like manually clicking it or remembering to
21:00 do that every single time. >> Yeah. Um, and so I'm I'm kicking around like the
21:06 idea of like maybe I ditch using Claude and and move all of my writing and
21:10 creative and marketing stuff into Claude code. >> Um, which is a possibility, but that
21:16 interface is not it's it's great for coding, but it's not great for like
21:19 non-coding stuff. >> Um, so I don't know. I'm still that's
21:24 that's an open question, you know. Um >> yeah because I think the other thing in
21:30 my head is I wonder if we can bypass this whole open need for openclaw
21:37 with just a new process inside of cloud code which is uh in claude code right now I can give
21:44 it examp I can give it permissions to access other directories theoretically
21:50 and project files. So, I could say, for example, right now I'm in the app
21:54 directory, but I could say, "Hey, I want you to go over to the marketing
21:58 directory. I want us to link these projects." And I've got notes
22:04 >> um in a folder called memory on both project directories. I've got a folder
22:09 called memory and all of my journal entries are there. >> Yeah. Um, and then maybe you could move
22:18 to uh the cloud CLI on your iPhone and then just interact with it just that way
22:23 as well. >> Yeah. And I and I also have been thinking about that is like do I really
22:29 even need OpenClaw? And I've seen other people do this too, which is like
22:33 >> like, okay, OpenClaw launched and it and it is a really
22:37 >> interesting and fascinating idea for putting agents on like their own
22:42 standalone machine, their own box, >> or do you even need OpenClaw? Can you
22:46 just run cloud code or or build your own whip up your own much simpler version of
22:50 OpenClaw? I've seen some people doing that like Christian Jenko was was
22:54 tweeting about that. >> I started to I started to explore that
22:57 for like half a day last week. I was like, maybe I can just do something
23:02 custom spun up. And then I was like, ah, there's all these like nice features
23:05 that I don't want to rebuild, >> especially the the memory system is is
23:09 pretty well thought out. Um, I don't know like that that might be an
23:13 evolution that like I do and that maybe many people will do over the next year
23:18 is like OpenClaw was the first of of many different similar concepts, you
23:21 know. >> Yeah. In the chat right now, they're asking um what about co-work? So, how
23:27 does co-work >> Yeah, that's that's another interesting
23:31 one. Co-work is sort of similar. You can start to get to like co-work is like it
23:36 does work on your Mac file system. I haven't really played with co-work too
23:41 much myself, but my understanding is um >> um I don't know. I I yeah, like that that
23:48 could be a good option. Um, but it doesn't like I don't know that it has all the memory
23:55 stuff the way that OpenClaw is is set up. Like I think it's still like
23:59 taskbased and sort of self-contained. Although you might be able to customize
24:02 it with a plugin and get it to because some of this it just feels like the
24:08 >> it's a process related thing which is again when I'm working with cloud code
24:15 basically I work all day and I'm I'm committing changes throughout the day
24:19 but before I leave for the day I want to commit and push and then I want to
24:24 update cla MD if there's anything important and then I also want to just
24:29 have a separate journal entry three, four project notes. What did we do
24:32 today? >> That's my ideal, but that could be for me a text expander snippet that I just
24:39 use every day at the end of the day in the terminal. Um, so I'm wondering if
24:46 there's like if if for the for the uh lite normies out there like me if if uh
24:51 that's a better system, >> you know, you know, like but like what
24:55 you're talking about is like I spent a day or two o over the past week just
24:59 thinking through that like what's my dayto-day how am I going to change my
25:03 day-to-day workflow >> so that I have more of my conversations
25:09 and artifacts centralized in a file system because right now there's still
25:12 too separated. And like, you know, I've I've also been a longtime Obsidian notes
25:17 user. I have like years of notes in there. Um, which is a nice markdown
25:21 editor and it does save them as files in your file system. So, that's nice.
25:23 >> Yep. >> Except it's lacking Claude. >> Yeah.
25:29 >> Right. Like I can't have Claude help me write my my notes. It's not a
25:32 conversational thing. It's just a notetaker. So I so I'm still trying to
25:36 figure out like how can I and I and then I I was playing around with the idea of
25:41 having a claude skill built into my Claude account so that when I'm having a
25:47 conversation like trigger saving this conversation >> somewhere and
25:53 >> but then even that ran into some some limitation of like it can't like like it
25:58 could maybe save an artifact somewhere but it can't take the whole
26:01 conversation. >> Mhm. Um, I don't know. So, there's a lot
26:05 of like little logistical things, but like ultimately at the end of the day,
26:08 the vision is and so I I spent a lot of time figuring out the multi- aent
26:12 things. Like by default, >> um, Open Claw is a single agent, right?
26:17 Like you're going to have one one Telegram chat with one agent
26:22 >> and I wanted to have multiple. >> Okay. >> So, in that screenshot, um,
26:27 >> I have something like six work agents and like two personal agents, right? Um,
26:32 and the idea is that each of those has a different Telegram >> chat with me.
26:36 >> Okay. >> Each of them has a different I even built like I had Claude come up with
26:41 like personas for each of them. >> Um, you know, like like the grumpy old
26:46 developer, the the junior young guy, the you know >> Yeah. Yeah.
26:50 >> I I use Gemini to to create some like silly like robot like inspired by like
26:55 the gorillas band, you know. Yes. Um anyway, so um so I have I have those and
27:03 like the idea is that I can I can have like different chats like like I have
27:07 one called Vale that's the content marketer. Mhm. >> Um I want to be able to have like
27:13 different Telegram chats and I also connected them to my Slack workspace so
27:19 that I can like have marketing related chats and tasks with this Vale agent and
27:25 development related related tasks with Bernard and you know strategy with
27:29 Clover. Um, >> I mean, I want this for sure. This kind
27:38 um, uh, persona that I can chat with and that I can do work with.
27:46 >> I also think the other the other thing that seems to be resonating about
27:51 OpenClaw is just the fact that you can chat with it via Telegram.
27:57 That interface layer actually solves a bunch of problems. Like for example,
28:01 everything that you dictate into Telegram now gets saved and all of their
28:05 responses get saved. So now you already have a a place to store conversations
28:15 and refer back to them. And I even like the the problem right now I have with
28:21 all this stuff is it's Siri. It's like the the when I'm using my iPhone and I
28:27 got my AirPods in, I the only person I can talk to out of the box is Siri. And
28:34 I can't do anything with Siri except for compose a note. I compose a note with
28:39 Siri and I get it to set >> timers and alarms. That's basically it.
28:42 >> Um, >> totally. And so I think the interface layer is what people are starting to I
28:50 think they wanted a better interface that was better than Siri, better than
28:53 what they were getting on their iPhone. Well, Telegram's easy. I can interact
28:57 with Telegram all day. I can interact with it voice. I can interact. I think
29:02 part of the the attraction of this open claw thing is just the fact that uh out
29:07 of the box you can connect it to WhatsApp and Telegram and Slack.
29:11 >> For sure. Um you're right. It's the interface thing of course like chat GPT
29:16 and claude like the part of the breakthrough with those a couple years
29:20 ago was the chat interface to AI right like GPT2 and one didn't have that chat
29:26 GPT layer yet so like that's why it didn't break through into like GPT3
29:32 >> but um but like so I think like chat GPT and the and the clouds like like
29:37 unlocked like chat is the way to interact with AI and then and then the
29:42 models got a lot better. But I think now with OpenClaw, the next like yes, it's
29:46 the chat with Telegram or Slack or WhatsApp or whatever you want. Mhm.
29:52 >> But it but then it's also the the other connections to everything else you do
29:55 because it has access to your to whatever you can do on a file on a
30:00 computer system. Like like that's the big challenge for me in my in my
30:03 business is like there's so much like okay I finish a I finish a a newsletter
30:08 that I that Claude helped me write but then I copy and paste that into kit and
30:13 then I format it and schedule it and like or you know just a lot of little
30:17 clicking and carrying this piece of content over to there and like
30:21 >> and scheduling it and like >> you know >> I think okay this is I I know we're
30:25 talking long about this but I want to this is actually something I I think was
30:29 also so attractive about all the mythology and legends and stories that
30:34 came out of OpenClaw is what's the challenge with anything? The
30:38 challenge with hiring a personal assistant who's a human, the challenge
30:42 with hiring an employee, the challenge with anything, the challenge with doing
30:47 work yourself is that there's always you're going to run into things, right?
30:51 So, it's like uh like you said, a typical thing would be h what's a very
30:56 common one that AI just can't touch right now? every >> video editing.
31:01 >> Well, well, video editing, but here's a simple one that seems simple. Every
31:05 quarter, my accountant says, "Justin, can you log into your bank accounts and
31:10 export all the PDFs and then send them to me so I can reconcile your accounts?"
31:16 That >> is difficult to hire somebody for because as soon as I hire a personal
31:22 assistant for it, first of all, it's sensitive. But if you get over that
31:27 hurdle, then you have two factor off that often depends on me being close to
31:30 my phone. >> Mhm. >> Then it just goes problem after problem
31:35 after problem. And I think one thing that appealed to people about this open
31:40 claw thing is it's like, oh, this thing will keep going if you instruct it to it
31:45 will keep going in the Ralph Wiggum loop until it's done. Like it will just plow
31:50 through barriers. And I can see why that was attractive to people. And then yeah
31:55 eliminating all these points of friction. So so much of what we do in a
32:00 day is all this connective tissue between like I compose something in
32:05 Claude. >> I might even take it somewhere else and clean it up and write it and then and
32:10 then from there I got to copy and paste it into Kit. and and and also it's like
32:16 for me it also happens when I'm out like I I do a morning walk for 30 minutes and
32:21 I'm doing a voice recording of of my next YouTube video like just the idea
32:25 just brain dumping the idea right >> um >> but you know I do want to speak to
32:31 something else too and this is where it didn't click with me the the day that
32:36 openclaw started to get popular and part of it is because of like the marketing
32:39 copy on the openclaw website and then what everyone what everyone started to
32:43 talk about Yeah, >> everyone is looking at it like this is
32:47 an agent that you're going to give full unfettered access to your own email
32:51 account and your own calendar and it's going to be like as you and do things as
32:55 you and have access to all of you and your life. Like to me like no, like that
33:00 is not what I want. I'm not. And that's that's why at first I was like
33:04 >> I'm not interested in that. I have no desire to have an agent reading and
33:09 replying to my emails for me. I don't have the problem of wanting an agent to
33:13 book restaurant reservations for me. I'd rather pick the restaurant myself. Like
33:17 >> like that that isn't a problem that I have. I'm not interested. That was my
33:21 that was my initial reaction, right? >> Yeah. >> What made it click for me was like,
33:28 think about this exactly like an employee or maybe a team of employees.
33:32 >> Yeah. >> So, the way that I'm doing it is I I did
33:36 set up this Mac Mini. It's it's its own user. It has no it's like I didn't log
33:41 into my Apple account on it or my iCloud or even my Dropbox account. It has it's
33:45 not connected in any way from there that that standpoint. >> I actually gave it its own email
33:49 address. >> Mhm. >> So I can invite that email to like
33:55 >> was that just a Gmail address or like um >> I used a proton
33:58 >> by me. >> Um and uh and so I like I used that to
34:03 to set up like a GitHub account for it so that so I could see like commits from
34:06 it. M >> um I can I can grant and revoke access through that.
34:11 >> Um >> did you give it its own credit card for
34:14 spending? >> No. Like like that's I mean I don't know like maybe at some point I I but I
34:21 haven't done that like I haven't had the need to do that right. Um,
34:26 it's just like what? So, if you think about it, like it's not it's definitely not the same,
34:34 but when you hire an employee, you are inviting them to to do things and have
34:37 access to things. >> And if they're if they're an employee
34:40 like in your company, you're probably going to buy them a a computer to to
34:45 work on, you know? Yeah. Um, it's kind of simil if you start to think about it
34:49 like that. You start like so for for me I I think about it like like my company
34:55 is me and a part-time video editor works for me. >> Mhm.
35:00 >> Uh other than that like I don't have clarity flow is separate. I have a
35:03 full-time developer who manages that. >> Yeah. >> But
35:06 >> in builder methods it's me and a video editor >> and uh I don't have any developers other
35:13 than me and and coding agents. I don't have a virtual assistant.
35:16 >> I probably would if if this were a couple of years ago, I would have hired
35:20 a virtual assistant by by now in my business >> for all those little tasks, but I'm
35:26 convinced that a lot of that stuff can and should be processed by agents and I
35:31 haven't figured out all of the processes, but it's absolutely possible
35:34 to to do that. >> Yeah. >> And then same thing with like content
35:38 and marketing. Obviously, like I'm do I'm the content person, but like I I
35:43 heavily use AI in my creative process, you know, like like the idea of like
35:47 outsourcing to writers is not a thing anymore, >> you know? Um,
35:51 >> and even like like I'm going to I'm going to do like ads funnels at some
35:54 point. Like >> I'm not going to hire I'm not going to hire like a pay-per-click expert for
36:00 that. I'm I'm going to like have like AI prepare a strategy and a game plan and
36:04 maybe I'll like do the initial setup of it, but it it'll be a process that can
36:07 be automated, you know? >> Yeah. I mean, this this is just such an
36:16 I mean, it feels like when it comes to hiring, there's a few different jobs to
36:22 be done. So the boss sometimes is hiring because it's like I need someone to take
36:29 this work off my plate and this is work that I'm is coming to me and being directed at me
36:37 or is could just come to an email and is being directed at email but somebody
36:40 needs to go there and go okay the accountant needs this so right now that
36:44 comes to me and then I could theoretically delegate that to somebody
36:48 else say hey this needs to be done they're basically just taking it off my
36:53 plate. Um, and eventually they could be the point person for that. Listen, I
36:58 could tell the accountant, you just email this person, they will take it
37:03 from there. So, that's one form of work. And that's still, I think, um, you know,
37:09 a big part of hiring, but I could see how uh I could see how AI, you might
37:14 hire it to do some of that work, right? It's just like, hey, I just need this
37:18 taken off my plate. the the one thing that you're going to miss still is hiring someone who can be
37:28 autonomously creative and proactive. Uh because it feels like the these AI
37:33 agents are still very reactive. They're always just taking direction, right?
37:37 They're not >> um they're not able to come up with their own ideas for things that you
37:43 might do or opportunities or or whatever. Um, so yeah, I I mean I don't
37:47 you don't need to respond. I'm just thinking that off the top of my head.
37:51 >> That's that's another big challenge that like I I run a creative business. It's a
37:55 creator business, right? Like it it >> it's mostly built on like me creating
38:00 ideas for for the YouTube channel and courses and and stuff. Um I'm also
38:05 thinking a lot about like how can I streamline what I do before I record a YouTube
38:12 video? like the ideiation process. I And I can't and I don't want to rely on AI
38:16 to come up with ideas. Like there are people out there who will just scrape
38:19 YouTube and just do like copycat videos. I'm not I'm not interested in that.
38:22 >> Yeah. >> Um >> but I do want to be able to take my raw
38:27 ideas and not spend a whole day sitting here like staring at the screen, but
38:32 have more of a process where I could have a chat with my creative agent
38:37 >> with with Opus. >> Yeah. >> Right. I mean that this is this is what
38:41 I do with Opus now is like I start with a raw idea, an observation, an insight
38:46 or something I want to talk about and I'll brain dump a voice memo in it and
38:52 and then I have a a long back and forth of of creative like fleshing it out and
38:57 then and then writing it and crafting it, you know? >> Yeah.
39:00 >> Um but like that's still like too messy for me. I still want to be able to like
39:05 like I'm not publishing fast enough because I spend too much time in that
39:08 creative loop, you know? >> Yeah. Um, I mean, this is this is
39:18 there's still just a part of me that I I think what's attractive about
39:21 OpenClaw is the same thing that attracts me to the idea of getting an executive
39:27 assistant is there is just a bunch of stuff in my life that it would be great
39:32 to have that. First of all, it'd be great to have someone else be able to
39:37 delegate the work. That's one piece. But even better, a step above that is
39:43 somebody who is taking um who is taking responsibility for that
39:47 thing and is just doing it in an ongoing way. And here's another example I can
39:54 think of. It's like like if if I put AI in in charge of spider control at my
39:59 house, how would it do that? I don't I don't know. Maybe it would detect like
40:04 number of screams per day or something like that, you know, like
40:07 >> my wife screams, "Oh, that's probably a spider." So, like
40:11 >> number of screams per day. >> Um, but the the truth is we hired a
40:18 spider control guy, right? And I had to like it was like my wife is like,
40:22 "There's way too many spiders. We saw a bunch of black widows. Justin, do
40:26 something about it." So then I take that I call up a spider control guy. I say,
40:29 "I want you to take control of this problem." >> So, I had to do that initial step. He
40:34 comes in, he scans the whole house, he does it, he sets up traps, and then he
40:39 says, and personally, this is like my love language. He says, "And now I will
40:44 come back and check every year and just see how we're doing." Perfect. Now, I'm
40:49 not responsible for spider control anymore. This guy's just going to show
40:53 up on a regular basis. He's going to check all the traps. He's gonna do
40:58 everything he needs to do and then he's gonna bill me. That's kind of ideal
41:04 that this is these are the real problems people are trying to solve and AI might
41:08 be part of the solution, but then I'm also thinking how the human element with
41:11 all this stuff. How does it all interact? you know, like >> Yeah. I mean, the way that I think about
41:19 it also is like um like I don't I'm not really looking to delegate the
41:29 the uh ideiation and creation of new projects. >> Mhm.
41:33 >> To AI and frankly even to people like I I am a solo founder of this business.
41:38 Uh, I'm the creator of it and anything that I create, a a new product, a new
41:46 video, a new course, a new anything, that's what I want to be spending my
41:50 time on. Like, like I chose to do this business because I like yesterday I
41:53 spent the whole day >> creating, designing this app to run my
42:00 agents. That that screenshot is is a whole app that that took me about it
42:03 took me one full day to design it and build it. Mhm. >> Um and uh you know it it tracks their
42:09 tasks, it tracks the token costs, it tracks activity, it it does a lot of
42:13 things in there and and it hooks into the file system. It's pretty cool. Um
42:19 but like I that there are people out there who are tweeting about like how
42:24 they just task their open their OpenClaw agent to go build build a similar
42:27 dashboard like that. Just go go do it, figure it out, do it. Like to me, I
42:31 can't I I I want to be in the nitty-gritty. Like every UI choice in
42:36 here was me directing it or crafting it in the way that I want it, right?
42:40 >> Um especially if it's a tool that I'm going to be using.
42:42 >> Yeah. >> Um so the way that I think about agents is
42:48 more like I want them to be the next phase. They they are the maintainers,
42:53 right? like like when I have a new feature for this or for builder methods
42:59 or for whatever like I should be able to voice note it into a chat
43:04 >> and it turns into a spec and if it's like s like small to mediumsized like
43:08 that's something that an agent can and should be just creating a PR for
43:13 >> um you know if it's a totally new redesign or a totally new concept then I
43:17 want to be involved in that then I'm going to do that myself in in cloud code
43:19 or whatever. >> Yeah. Um, but if but if it's like fixing
43:26 a bug or or doing this or or like if it's a marketing process, right? Like if
43:32 I have a like um like there there should be some idea like some creative stuff
43:35 where it's like it has access to all of my activity, all of my projects through
43:40 these through the file system >> and every week like it knows that I need
43:44 to be generating new ideas. So maybe it surfaces like a list of like here are
43:49 some interesting uh p like concepts that I just pulled out of your recent GitHub commit or your
43:56 recent podcast transcript. Like these could be tweets maybe >> bookmarks or like the these could be
44:01 like >> maybe something that I said at 42 minutes into the recent panel episode
44:08 could become >> could become the basis of my next YouTube video. I don't know. But like
44:13 there there's so like to me the the big one of the big drivers was like I there
44:18 is so much activity that happens >> Yeah. >> between me and this computer and these
44:23 and these podcasts and the code >> that >> and I'm I'm not I'm not tweeting all the
44:27 time. I'm not like >> Yeah. >> I'm not broadcasting everything that I'm
44:32 doing. I'm broadcasting maybe 5% of what I'm doing. >> Yeah. you know, um, and so I if I could
44:39 figure out some systems to like extract more publishable content because I feel
44:42 like there's a lot of good material. >> Yeah. >> And and and sometimes I've done that and
44:49 and it's like it's like um I start to do that and then I'm like, ah, this is so
44:53 raw now I want to spend an hour crafting it and refining it and now I'm not
44:56 actually getting any real work done. So if there's just some sort of process
44:59 that's happening in the background like watching and extracting ideas and
45:03 turning it into at least pretty good draft that I can approve,
45:09 you know. >> Yeah. I I mean I yeah I I'm I'm curious to see how this uh this experiment I I'm
45:16 kind of waiting for everyone else to figure out this stuff and then but the
45:21 parts that appeal to me are you know the the Christian Genko like um uh talk to
45:26 my phone and because right now you know I I I can't interact with Claude code on
45:32 my phone at all but theoretically it could be doing all sorts of work for me
45:37 while I'm just out doing other things. So, yeah, I'm I'm curious about all of
45:41 that. >> Um, let's uh let's switch it up. I want to talk a little bit about building fake
45:47 products. >> So, you asked me, hey, why why aren't we using that that podcast countdown timer?
45:56 >> And I I feel like this was always true, but now the cycle times are so much
46:00 shorter. The value of build first has just increased dramatically.
46:09 Build first allows you to create something in a couple of minutes and start using it and
46:17 start uh basically interacting with reality right away. So the podcast timer
46:25 countdown timer is a perfect example. In that app, we set up segments, we give it
46:31 a timer, it counts down. As soon as I introduced that with you and I, it it
46:36 added I noticed it added a little bit of anxiety. It's like, oh man, I got time
46:42 now counting down. So, you start to feel a real interaction with
46:49 the product right away. Um, one of the things I've been talking
46:56 about on this show forever is can we at Transistor figure out a way to do video
47:01 podcast hosting? And it's just such a big hairy problem. It it feels like too much. And then a
47:10 couple weeks ago, I said, I'm just going to start building it and see my beer
47:17 >> how it feels. And not not because I want to build something that's production
47:20 ready that we're going to just drop into the app, but because for this very
47:24 reason, I want to try start to experience both from the technical
47:29 implementation side, but even more as a user, what do I run into? What do I
47:36 want? What do how does this feel? And in the past, you just could not do this.
47:39 Just to get to the point I'm at right now, I built a full video podcasting app
47:47 in a couple days, the first version. And just to get to this point with pretty
47:52 talented engineers would have taken 6 months, a year, maybe even two years.
47:58 >> It would have taken a long time. But I got I've got a little app running.
48:05 >> I mean, I and uh sorry, go ahead. What does this do like like uh for those who
48:09 aren't watching they're listening you just started to show this thing that it
48:13 looks like it's called Ampcast. I don't know if that's a temporary name.
48:16 >> Yeah, that's just a temporary name. This is never I don't I don't think this will
48:20 ever be a real product. This is just a prototype I've built.
48:24 >> So we've got show a list of shows. You click into one.
48:27 >> You can connect it to Transistor and import an existing show. So let's say I
48:34 import restaurant success here and then I can connect it to YouTube
48:40 and it that works. I can I've got all the off and connection to YouTube and
48:46 everything working and it again I I >> what makes it a I'm curious like what
48:50 does it make what makes it a video podcast? Well, this is this is what I'm
48:57 this is what was so helpful is previously I thought the the way to do this is we
49:04 have to do everything. We have to allow people to upload the video to us. We
49:09 have to encode it for like use HLS encoding, host it, etc. As I got into
49:18 this, I started to realize that what people actually want is not that what
49:26 they want is let me see if I can if I can find this thing here. like turning
49:29 their YouTube videos into a >> basically they want to upload a video,
49:36 have it syndicated to YouTube and maybe Spotify video and then create an audio
49:41 only version that syndicates to all the audio platforms. >> And it wasn't until I um it wasn't until
49:52 I started working on this that this became very clear. Right. So, we've been
49:59 talking I think you know HLS maybe should exist for the nerds but when I
50:04 think about 80 90% of people I talk to who are doing podcasting what do they
50:07 want? >> They want to upload one video file they want it to go to that to get get
50:16 uploaded to Spotify and YouTube as a video episode and then they want it
50:21 transcoded and as an audio file and added to transistor. That's what 80% of
50:26 people want. >> Totally. >> And so I didn't experience that.
50:33 I didn't fully realize that. That was in my brain, but I didn't fully realize
50:39 that honestly until uh I'll see if I can find it. But actually when I was working
50:47 through uh Design OS, I had it create a uh breadboard for the different flows.
50:53 And once I saw the breadboard, I was like, "Wait a second." And once I'm
50:58 talking to Claude Code and it's like, yeah, like that HLS video component
51:04 really adds a lot of complexity to this app. >> Mhm. >> Um,
51:09 that's that's like very similar to my process with any new new app. It's like,
51:13 all right, like here's my concept. I think I want to do it this way. And then
51:16 and then you run into things and have back and forth with Claude and
51:18 >> Yes. >> Yeah. And then once you see the see the
51:21 views see and like yeah >> well and there's just so many things
51:25 that you run into that until you're actually using it usage has always been
51:32 oxygen for product realizations for actually understanding something.
51:37 >> So here's another example. So, >> does it >> I was wondering like does it push out to
51:42 like if do would you upload the file to transistor and then that actually
51:47 uploads to YouTube and Spotify and >> in my little prototype what this does is
51:52 you just go here I'll I'll do one right now. So, let's just make sure this is
51:58 connected to YouTube. So, I'm going to I have to keep re-offing it because I'm still in test
52:05 mode. But for the listener, all I've done is I've opened up a fake show
52:08 called Super Awesome. I'm going to click new episode. I'm going to choose a video
52:14 file and it's starting to upload. That uploads to uh R2. I'm going to say
52:24 sample app for Brian. And uh the other thing I like about this, by the way, is it's allowed me to
52:32 play with a simple interface. Again, I even hid a lot of the advanced settings
52:37 in this expand collapse UI here. >> I'm going to click publish episode.
52:42 >> This is now uploading to YouTube. That's completed.
52:48 Now it's extracting the audio. That is processed and completed. If we click
52:53 here and view it on YouTube, still processing the video, but that will be
52:56 live right away. And you click on transistor. Episode is live. The YouTube
53:02 embed has automatically embedded. Oh, you haven't. You can't see that. YouTube
53:06 embed has automatically embedded in the share page. The episode is playable. It
53:13 all happens in one step. >> Nice. >> So, >> that's that's beautiful.
53:16 >> That's >> I mean, >> and and here's the other benefit. I've
53:20 been playing with this. I've been using it. >> Especially if you could you could
53:23 schedule it. >> You can schedule it. I've already got that built in. the I'm now showing this
53:31 to people. So, my one buddy, he he does uploads all the sermons for his church
53:35 and he's like, "Dude," he's like, "I wish I could just do this in one step.
53:38 It's taking me so much time to first upload it to YouTube. Then I basically
53:42 copy and paste the same title, description, etc. over to a podcast host
53:47 and then do it there. I just want it one-stop shop." So, I I can
53:53 record a video for him. Eventually, he'll be able to use this. And now I'm
53:57 gonna have a real person using it and will experience how does this work.
54:03 >> Today for the first time I I had to we at Transistor right now we don't have an
54:07 uploader that can handle uh files bigger than a gig. So I had cloud code build a
54:13 file upload file uploader that can handle big files. today for the
54:20 marketing for developers podcast. >> This Lars Laughof interview was uh 15
54:25 gigabytes or something like that. It's like okay well let's let's see how this
54:31 system handles a 15 gigabyte video. I'm also going to be tracking the costs now
54:36 on you know the different like for hosting and for bandwidth and all that
54:41 stuff. I can test all this stuff out before we implement it in our production
54:47 uh product >> and I think building fake products is going to it people need to be doing
54:57 this. This is this is the way to explore and truly feel and shape the entirety of
55:04 the problem before you sit down and write a a you know a implementation plan
55:08 for your existing app. This is like let's figure out everything now. Even
55:14 even something as small as like you'll notice this one says private. This is a
55:18 private podcast. And then I as soon as I start using this cuz I wanted to publish this episode
55:25 today. I'm like wait a second. A private podcast is going to have to have a
55:30 different setup than a public. And I'm like what would I want personally as a user? Well, I might
55:37 just want it to be an unlisted playlist, an unlisted video on YouTube. Can I do
55:42 that? Yeah. So, now if it's a private podcast, it automatically creates a
55:46 unlisted playlist on YouTube, publishes an unlisted video. >> I would I wouldn't have thought of that
55:51 beforehand. >> And all of this reminded me of this great story from Fresh Books. This is a
55:59 story that Mike McDermott told me ages ago. And I don't know if a lot of people
56:03 have heard about this, but basically they decided they they they had this
56:08 problem back in who knows when this was 2017. They had this problem where they had
56:12 this mature product, but they wanted to do a full rewrite. And so they decided they went through a
56:20 bunch of ideas, but then Mike said, "What if we just created a new company
56:24 and competed with ourselves? A new company could have its own name, brand,
56:27 logos, website, articles of incorporation, user agreement, service
56:31 staff. we could use that to figure out if we could build a product that is
56:34 truly better than the one we're offering. They built for so FreshBooks
56:40 V2 or V3 or whatever it was. They created an entirely different company
56:46 and then they built the product ground up from there. Got real users using it,
56:51 had its own pipeline and everything. And then only when it had proven itself with
56:56 whatever it was, 100 users, 100 paying customers, did they switch it over and
57:03 have it uh have it as the main uh FreshBooks app. And then >> I didn't know that they did that. That's
57:06 cool. >> Um, >> yeah, I think >> really cool. I like this what you're
57:12 describing this whole this whole process of like >> you get bringing a new feature uh at
57:19 least in a in a concept or a prototype form like it it it dovetales exactly
57:22 with what we were talking about with with Jordan on the on the recent episode
57:27 over here a couple weeks ago. It's like I'm I'm curious to know for you with the
57:31 with your team at Transistor like what is how is this going to change the
57:36 process of bringing new features into Transistor? Like I I just uh this
57:39 morning I recorded my next YouTube video and it's about the Claude integration
57:44 with Slack. >> Um and I at first I was like that's okay. It's just a little connector with
57:50 Slack. That's kind of a nice to have. What's the big deal? And then I saw one
57:52 of the >> one of the cloud um team members post an article on on X about it and how they're
58:00 using it in inside cloud and I realized like oh there there are some real
58:04 workflow breakthroughs here and one what what you were just describing about like
58:09 okay transistor is its own thing. It's it's it's wellestablished codebase. It's
58:13 got this whole user base on it like >> um you can't just like change change
58:17 course or ship features like in a day like you can with a vibecoded thing.
58:21 >> Yeah. But but what they were doing what they're doing at Anthropic apparently is
58:27 in Slack you can like with the Slack connector to Claude code
58:32 >> anyone on the team marketers product people go to market sales
58:38 anyone can at message claude and kick off a claude code project and then cloud
58:44 code goes and and works on it and creates a PR to the actual codebase.
58:47 Mhm. >> So like so that's an interesting product process workflow for teams,
58:54 right? So it's like >> um instead of vibe coding a prototype
59:00 which is very different and new and shiny and and separate from what your
59:05 actual codebase is like anyone can pull off a branch and >> and vibe and essentially vibe code and
59:11 not even vibe code but like at least just start like like what what if we add
59:16 a video uploading feature to our existing codebase? What would that look
59:20 like? And and it and and could we achieve this? Is that even technically
59:23 possible? anyone like obviously it'll depend on your on your own team's
59:31 like strategy and workflow but but you or anyone on your team could
59:36 like essentially at messagecloud and at least start a PR this isn't going to
59:39 ship tomorrow but at least it's something that it's like oh this is
59:42 technically possible and and like it it could or or we will run into these
59:47 technical challenges of it like now now we understand that and then the
59:50 engineering team can take a look at it and it doesn't even necessarily have to
59:54 be like a code review challenge. That could just be the the V1
59:59 >> of this or or or like a draft version of this feature so that the engineer can
60:03 see like, oh, okay, >> I see what I see what the product team
60:06 is going for here. Yeah. >> Let me rebuild this properly with so
60:11 that it's ready for prime time. But like that's a much higher fidelity
60:16 >> vision than like a wireframe or even a vibecoded prototype that's separate.
60:19 like if it's a PR off of your real code base, you know, >> I think we're going to do both. I mean,
60:24 we're still stepping into this carefully. >> Yeah. >> Um I think we're going to do both. For
60:29 this particular project, it made sense to do it separate because we needed a
60:33 separate R2 bucket. I don't want to use what Transistor already has. We There's
60:37 a bunch of things that I just wanted to keep separate because this is involves a
60:42 lot more moving pieces. Um >> so you're so you're going to rebrand to
60:45 AMCast. Um, I mean it there is a world in which we go, you know what, we don't want to
60:52 add this to Transistor. We're going to rebuild it as a separate thing. Uh,
60:56 maybe a video first podcasting app. That could be true. Uh, I I think also the
61:06 being able to to have something that I can hand over to my friend who's working
61:11 at this church um, and have a limited subset of people actually using this
61:16 workflow and just running into everything that could happen with us
61:22 with usage, taking all those notes and then using that to implement that in the
61:27 main app. I think that's the other piece. Um, and then the other thing is
61:31 just being able to see um, think I've also connected it to the
61:38 YouTube analytics API and the transistor analytics API and so now I have a a a
61:45 view of uh, YouTube views versus audio downloads. >> Sweet. Right. And so, um, being able to
61:53 do that in its own space and show it to people, show it to different users, uh,
61:58 friends of mine that I know are actual real podcasters and eventually getting them to use it. I
62:06 think, um, there's just a real benefit to that. And, yeah, then using
62:11 everything we learn to inform what we're doing with Transistor. The other reason
62:17 to do it outside of your regular app is this also led me to to it's still a
62:21 Rails app on the back end, but I said on the front end I don't care what you use.
62:26 I you you can use whatever you think is best cla >> and you know we we had uh uh when we
62:33 built the transistor app, we had some strong opinions about how much
62:36 JavaScript we were going to use, what kind of JavaScript libraries we're going
62:39 to use. I I wanted to go and run with scissors away a little bit.
62:45 >> Fresh start. I I mean, yeah, like I'm I really see a night and day
62:52 experience between literally apps that I started the codebase 6 months ago or
62:56 earlier >> and apps that began within the past 3 to 6 months.
63:02 >> Yeah. like um the ones that I started more recently are just so much easier to
63:08 get into and iterate on because AI helped build from the from day one and
63:14 like and you know going into legacy code bases even builder methods which started
63:18 earlier in the year last year >> there's a lot of my old opinions baked
63:23 in there which which does make it harder for agents to be able to pick up from
63:29 from that point right because like it it has my handcoded opinions from day one
63:33 that it has to work around that it has to try to align itself to and yeah like
63:38 my agent OS sort of helps with that but like it's still a challenge but like it
63:42 but something like like the app I built yesterday for the agent tracker like the
63:45 HQ thing it >> like like I just designed it and I used
63:51 design OS for the design but it but it it created React components that I'm
63:56 that I you know brought into a Rails and Inertia and >> well here's the other thing I'm doing
64:01 this is this is the Other thing is again I thought okay I I built this ampcast
64:05 thing and I that was just me working with cloud code here's the vision here's
64:09 the plan we went through plan mode we did all that stuff but now in a separate
64:15 cloud code session I'm using design OS to re design and architect the same app
64:20 again >> but using a totally different approach and um and now so design OS is helping
64:27 me to really think through onboarding helping me to really think through all
64:32 these edge cases and oh well you if you if you don't handle this at the
64:37 beginning even something like at what point do you connect to YouTube at first
64:41 I had it on the account level right when you get in and then as I'm using the
64:46 first version of my app I'm like >> actually you want this to be you want to
64:50 be able to connect to different accounts and channels per show
64:54 >> okay >> well I go into design OS and I'm like
64:59 writing the spec from and we'll See, maybe the first version I make is
65:03 actually the better version, but when the cost of building a prototype is
65:09 essentially just your time, >> why not do it multiple times? And then
65:14 maybe the second version I build, I just get more realizations. It's like, oh my
65:20 god, like now, and I'm building all these asset. One thing I really like
65:23 about Design OS, if you're not using it, I I highly recommend it, is I didn't
65:27 realize that you you create all these like interactive prototypes. So, after a
65:34 section's shaped, Design OS creates those like you can click through on
65:38 those screens, but they're basically interactive prototypes. You know what
65:40 I'm talking about, right? >> Well, you can Yeah, you can view the the
65:44 designs that it creates. >> Yeah. I I didn't realize that Design OS
65:48 did that. >> Oh, yeah. I'm going to bring I'm going to bring all of those assets
65:55 >> I when we actually if and when we decide because that's the other nice thing
65:57 about this is I can build all of this get it out of my system
66:01 >> actually like since we're talking talking about I can pull up the design
66:04 OS that I did for that thing yesterday. >> Yeah, I want to see that.
66:06 >> Let me see if I can um >> and maybe I'll pull up the one
66:09 >> I actually tried to record this as a video for for members but then the the
66:13 recording crapped out and I couldn't use it. Um, >> here I'll show you. Um, let me show you.
66:17 >> I think >> let me see. Oh. Oh, mine's not loading
66:30 >> Um, I'm I'm definitely going to use all of those screens.
66:34 >> Um, how do I share this? >> Should be share button right at the
66:42 >> I've got like two screens here. Screen. >> by the way, Riverside, can we get some
66:50 keyboard shortcuts? Can we Can we get some Is there a keyboard shortcut for
66:52 just sharing your screen? >> There it is. >> Okay. Do you see this?
66:59 >> Yeah. >> So, this is the design OS. Um, this is
67:05 for what a a little product that I call BMHQ. It's like builder methods HQ. It's
67:10 it's only for me to use with my agents. Um, and I started with like the the
67:14 overview of it. And then I ended up building out these sections. It's like I
67:20 like it's going to have dashboard. It's going to have the agents, the activity,
67:25 a usage tracker, uh, the the crown jobs that it's running, um, the tasks with
67:29 canban board and skills, right? So then I went went through and um got into the
67:34 design stuff. I didn't I didn't spend a ton of time on that, but then in into
67:38 the sections. So, like here's the dashboard and then you can view the
67:41 actual dashboard. >> Yes. >> Um, and it's got like a like a, you
67:49 know, like a responsive thing there. And then it also has like dark mode that you
67:53 can turn on. >> Um, >> like this was a game changer just to be
67:58 able to see things beforehand. >> Yeah. So like like this is like the the
68:02 dashboard view, but then I sort of need to go back into um so like if I go to
68:07 agents and then agents has like a list view and an agent detail view. So then if I go to
68:14 the list view um I ended up actually redesigning this a bit in my in my real
68:19 one. So like the real one looks like this and I I mapped a domain to it. um
68:25 agents. I I ended up making them like sort of like Polaroid um things.
68:31 Uh but in Design OS, it sort of started out like this. Um and then
68:38 so yeah, like I I I sort of like went through the process. One thing like
68:41 going through it yesterday, it's been a few weeks since I actually used Design
68:44 OS on anything. >> Um here's the tasks one. So like if I go
68:51 to the board view, I've got this like canban board where I can, you know, um
68:56 work with tasks. Um I can >> It's crazy that design OS just creates
69:00 this >> just from your just from the interview. >> Yeah. Yeah. Exactly. I mean the to get
69:10 to this first of all like I did the the shell, right? to get to this the the key
69:15 insight I think in any design process with AI now is yeah the interview gets
69:20 me to this like little mini spec right here which is like user flows and UI
69:25 requirements but then I think the key thing >> is this sample data
69:30 >> the the data models are less important what's what's a little bit more
69:35 important is generating this sample data >> first before you design it
69:40 >> so have AI like get a get a really raw a rough concept of like, okay, we're going
69:44 to do this canband board, it's going to need to have tasks and assign them to
69:50 agents. And if we were to generate some JSON data off of that, here's what that
69:53 data would look like. Here's the shape of it. And then once it has that, it's
70:00 able to like create a view that takes all of that into account and puts it
70:03 into an interface. because if you if you skip that step, then it's going to sort
70:08 of like dream up all these details, >> you know, and then it's not going to be
70:12 quite right, you know. Um, >> so yeah, but that like going through it
70:18 yesterday, I I did find like like now I'm thinking like at some point I want
70:21 to do an update to Design OS because it it still took too long.
70:26 >> It does take a long time that I I I did have a feeling of like ah like
70:31 >> I this took a while. >> Yeah. And like there's a lot of these
70:34 steps where I'm just like even going through the spec and the sample data.
70:38 Like it's important, but it's not even important that I review it. I just want
70:41 it to be done. Yeah. >> And get right into the design part.
70:45 >> Um and then the other part that's a little bit too heavy right now is is the
70:51 export. So like um when I get to the final export where I export out these
70:55 React components and the views, >> it's a little bit too prescriptive
71:00 >> in terms of like like yes, it gives you all the design
71:04 react components ready to drop into your real app. >> Yeah.
71:08 >> But it's also including in that like all this like prescriptive like here's how
71:12 you should model the database and here's how the backend logic should work and I
71:17 and I don't want it to do any of that. I want the I want cloud code in the final
71:21 app to be able to take these React components and just the highle feature
71:28 requirements and then that agent to decide and help you help you figure out
71:32 the actual data model. >> Yeah. >> So there's I think like the there is
71:36 going to be another version of design OS at some point that strips out just like
71:40 I did with agent OS version three. I stripped out a lot of the bloat. Um,
71:44 >> I think going through it helped me realize like there's a lot of extra
71:48 bloat that can be removed. >> Yeah. Yeah. I'm a big fan. I think even
71:56 just the just even having the uh just even having the process is really
71:59 helpful. >> Yeah. Um, I got a quick thing, but I do want to hear about Lars.
72:06 >> Let's do a quick thing. Yeah. Um, well, this is just like a I don't
72:09 know, like a mindset and like a challenge thing that I that I've been
72:13 feeling a lot lately. And part of it is what led to wanting to do this agents
72:19 thing with OpenClaw, but like I'm a solo founder, solo creator. Uh,
72:24 the business was going really well, you know? So, I used to feel like last year
72:27 it was more of the urgency around like I got to keep hustling and pushing to to
72:33 make this business happen so that I can get revenue going and then I feel a
72:37 little bit more secure. Like that's I've moved on from that. I feel I feel
72:41 pretty secure with this business now and it and it's pretty smooth sailing and
72:44 definitely growing. >> But now I still find myself working all
72:54 the time and I still have this urgency. Mhm. >> Um and now right now the the challenge I
73:01 think is this bal like there if if you split my whole work life into two
73:04 halves. >> Mhm. >> One half is the public facing you can
73:10 call it marketing like all all of my work on YouTube and email like the
73:16 weekly email newsletter and this podcast. Yeah. And my and my new builder
73:20 methods podcast and like all of that is like public facing stuff.
73:24 >> Yep. And especially with the wave of competition that is rolling into my
73:30 space here in January, February. >> Yeah. >> Like I constantly feel a pressure to
73:35 keep keep that like pedal to the to the metal like keep pushing on growth and exposure
73:44 and audience and lead flow and funnel. >> And I I've got a funnel that's working
73:48 really well. Mhm. >> But every time, like right now, like
73:53 since I had the big snow trip, >> I couldn't record and publish that week
73:57 >> and then the week after I I still didn't have enough time to record a new one.
73:59 >> Yeah. >> So, there's going to be like a two week
74:03 gap in like a new video on my YouTube channel. That's like way too long
74:07 >> for me in my space right now. >> At the same time, the other half of what
74:13 I do is delivering a Cloud Code course, delivering uh private content for my
74:17 members. Yeah, >> interacting with my members every day,
74:21 you know, and then actually building stuff using my own tools, design OS,
74:26 improving design OS, improving agent OS. >> Yeah. >> Like that is so incredibly timeconuming.
74:33 And then not to mention I need days per week just to tinker and research and
74:37 learn. Like I'm going to do a big YouTube video on OpenClaw at some point
74:41 based on all this learning, >> but I need the time to do to do that
74:46 learning, right? And so it it it's really challenging right now. Um
74:50 >> I think I mean I think you should do a video just on this topic because that
74:56 pressure you just described. I mean your pressure is unique in a sense because of
74:59 the business you're in. But I think everybody's just feeling this pressure
75:04 like this is the land grab moment. This is the the the train's taking off and if
75:09 I'm not on the train I'm going to miss out. Everybody's just thinking I got to
75:13 build all the time. It's like we're all addicted to drugs right now and um that
75:22 >> it's exhilarating. It's fun and a lot of people on Twitter are
75:26 talking about the fun part of it, but we're starting to also get some of the
75:31 reflection like Caleb Porzio just had on his podcast notes on work just had. He's
75:35 like, "Is this better? Am I still having fun?" >> I heard that. I actually have some
75:40 separate thoughts about that. Um >> but uh but the Yeah. Like
75:46 it's um like like one of one of the things that has this tension that that I that I felt for
75:51 a while now is that like yeah, there's a lot of guys out there who are tweeting
75:56 non-stop every day about >> like claiming to to be the the AI guy
76:01 for this and that, right? And but I think that my whole angle on the on the
76:05 YouTube channel is that I'm a little bit more thoughtful. I take a lot more time
76:08 to create and and I only create a video when I actually have something to say
76:11 about something. I'm not just going to copy any popular topic that's out there.
76:14 >> Yeah. >> Um but at the same time, like it's that
76:18 Jerry Seinfeld thing, right? It's like it's like, yeah, you you I don't know if
76:23 it's Seinfeld or some other person said this, but it's like yeah, you got to you
76:26 got to like create your your best like creative ideas, but you have to have
76:29 those creative ideas at 8 a.m. every single day. >> Yeah.
76:32 >> You know, like like you still have to have a schedule uh to to make sure that
76:37 this thing is actually still producing content essentially, you know?
76:41 >> Yeah. So, so you're feeling the pressure of just how do how do you stay on top of it?
76:47 Things are changing so fast. >> It's it's really even more like I what I
76:54 actually want to do is deliver twice, five times, 10x more value to my
76:59 paying to the paying members. >> Like if I could spend all my time or
77:04 even just most of my time >> inside Builder Methods Pro
77:08 >> Yeah. creating and giving them my full attention and creating tools and
77:13 creating private videos that show really interesting projects. Yeah.
77:17 >> Because what's fun about those types of videos is that I don't have to polish
77:20 them for YouTube >> and they don't have to be so scripted.
77:23 They can be just like watch me build this really cool thing.
77:27 >> Yeah. you know, or teach a lesson on claude code like um like
77:34 that's actually more val I I want to be delivering even more value to my members
77:39 and yeah like there's a lot of stuff but but but like this it's it's still this
77:45 like I I feel an obligation to the funnel like I I can't just
77:50 >> I can't like lay off the funnel too, you know? It's uh it's just really really
77:53 difficult. >> I mean I think Like not to mention the sponsorship stuff. Like I I get I I
78:01 could probably triple my revenue next month if I just open the doors to
78:03 sponsors. >> Yeah. You know. >> Um Yeah. And is there a way of doing
78:08 that that won't kill you, that won't overwhelm you, that won't um
78:16 >> I was talking to a sponsorship coach, um Justin Moore the other day about about
78:21 this and uh >> yeah, like he he he was really helping me out with some strategy on how this
78:27 could work and and what kind of value is or like untapped value there is for for
78:34 my channel and what I'm doing and >> and I plan to do it. I I should be doing
78:39 it now, but it's just a it's I don't have the bandwidth to insert the sales
78:44 process and and and the and the operation for doing sponsors just yet.
78:48 >> Yeah. >> This is this is I mean this is the challenge in any
78:57 market, but a a market that's a rocket ship that feels like it's taking off.
79:00 That's my favorite pely quote is he he said he launched Balsamic the wireframe
79:05 app and there was so much demand in the market he felt he said it was like
79:09 holding on to a rocket ship by my fingernails like that's how it felt and
79:14 it sounds like you're feel like you know the rocket ship's there
79:18 often the problem is the rocket ship's not there right you're just like trying
79:22 to trying to make something happen and it's like this thing's not taking off
79:26 but then the reverse of that problem is like this thing takes off and you're
79:30 just holding holding on hoping that you can keep holding on, right?
79:35 >> Yeah. Yeah. Well, I'm curious if Lars Laughrren has any uh wisdom that could
79:39 that could help with that problem. >> I mean, some of this is going to uh some
79:45 of this will uh probably make your life worse. >> Probably.
79:51 >> Um Yeah. So, let me actually >> He's such a good comm I I can't wait to
79:54 hear that episode, but he is such a good speaker and communicator, you know.
79:59 >> Yeah. Yeah. So, let me let me see if I can play a little clip from that because
80:02 I think this is one of the big takeaways I had. All let's see if this works here.
80:07 >> Okay, we'll make this live. >> Google isn't just looking at links in
80:13 order to figure out entities. >> They're looking at the entire web.
80:15 >> Yeah. >> So, they're looking at Reddit, they're
80:20 looking at YouTube, they're looking at social, they're looking at everything
80:25 else. If you go build a website and you're just relying on SEO and links,
80:29 like you don't have that footprint everywhere else, >> right? You don't you don't have reviews,
80:35 you're not on Trust Pilot, you're not on G2, you don't have a Google business
80:39 profile, you're not on Reddit, no one's talking about you because you have a
80:42 brand new business. Like, you haven't built up your brand, but more
80:44 importantly, you haven't built up that like entity across the web. And if
80:50 Google's looking at your and says like, "Wait, who is this entity?" They
80:54 won't even look at the content. >> Yeah. >> They'll just say, "Hey, we don't know
80:58 who they are. We don't know this thing, so we're not going to rank them."
81:02 >> So, uh, >> be everywhere. >> The This is This was
81:08 one of the key insights and and I'll today. The challenge with LLMs, and this is for
81:19 personal brands like yours, this is for companies now LLM and Google aren't just looking
81:29 at uh oh, Brian Castle, let's go over to briancastle.com and see what's there.
81:32 No, they're not just looking at transistor.fm. No, they're not just looking at the
81:37 backlinks to transistor.fm, with them which was in the past all you needed was
81:42 a bunch of authoritative links >> linking to your site.
81:49 >> Today they are looking at your overall reputation as an entity. Meaning if all
81:54 you have is a website they're going to look at you suspiciously. But if you
81:59 have a website plus 100 reviews on Trustpilot plus a bunch of real people
82:05 discussing your brand product, etc. on Reddit, plus a bunch of people doing the
82:09 same on YouTube, plus a bunch of people doing the same on LinkedIn, plus it just
82:14 keeps going and going going. It's the footprint that matters. And this also
82:20 has to do with personal brands. So, one of my favorite things to do these days
82:25 is to go into one of the LLMs. I I actually like Grock for this because
82:29 it's connected to Twitter. And I'll say, "What is Brian Castle known for?
82:35 What is Justin Jackson known for? What is Ian Lansman known for? And
82:43 you'll see how sticky some of this stuff is. So if all the if if the primary
82:52 thing that LLMs and Gemini have on Brian Castle is that he's the process guy.
82:58 He's the productized service guy. That's what they're going to list out. So if
83:02 all your reviews, if people talking about you on Reddit, if people talking
83:05 about on you in Twitter, etc., >> that's what they're going to talk about.
83:10 If your brand, if your product has done a pivot, uh like Transistor did from
83:16 being like the podcast hosting for brands and professionals to just podcast
83:22 hosting for proumers and creators. Um you that needs to be everywhere and your
83:30 brand needs to be everywhere. And he and Lars is just like this is this is
83:35 foundationally changing everything. It's the reason he's way more active on
83:39 LinkedIn now as his personal brand because he needs people interacting with
83:45 him and uh more information that now when the LLM's scrape him or their their
83:51 sense of who he is as an entity, it won't say Lars is the CMO at
83:58 Kissmetrics. He wants he wants them to know something different and the way you
84:03 do that now is you have to manage a very wide footprint across multiple
84:07 platforms. >> I get the concept. The question I think everyone is probably thinking is like
84:14 how what is the how do we do that realistically even if you're a company
84:18 with a with a lot of resources that's still super hard right?
84:21 >> I mean some of this is going to depend on personality. So, the thing that Lars
84:26 and I discuss is he's like, basically, this is system's perfect for Justin
84:31 because I've always taken this kind of gardening approach to marketing where
84:34 I'm just walking through the garden. It's like, oh, there's my Reddit part of
84:38 the garden. I'm just going to go and help that to grow and I'll go over here.
84:44 >> Like, you're obviously very out there, especially on like podcasts and and
84:49 YouTube, but you do jump around to platforms a lot more than people might
84:53 realize like new platform that came up >> channel and then you show up on blue sky
84:57 and then in Reddit and it's like dude >> I'm just I'm just walking around the
85:03 garden and I got lucky in this case that this approach just works awesome and
85:08 then stuff I used to garden in the past like Kora is now a ranking factor again
85:15 it was a ranking factor like 10 years ago now it's not as much of a ranking
85:19 factor but I invested so much of it in it well not 10 years ago but
85:22 >> I When you say walking through the garden, is that like literally
85:27 throughout the day, throughout the week, >> you do is there any process to this or
85:32 is it just random like I'm going to log into my Reddit and see my notifications.
85:35 I'm going to log into my Twitter. >> If you're really into structure, this is
85:39 going to maybe I mean, you could probably do it in a structured way.
85:44 >> I'm I look at Reddit every day. I look at LinkedIn every day. Uh, one of the
85:50 reasons I'm trying to do more podcasts, and this is Lars's approach, too. The
85:55 reason, by the way, the reason that all of these AI CEOs are still doing
86:00 podcasts is exactly this reason. >> They know that the more stuff like that
86:05 they do like this, the more it feeds in to the models. They're doing this on
86:12 purpose. And so it is for me it is just like I have a rhythm
86:17 to my day and my weeks which is just yeah I'm going to go around and just
86:23 keep and then every even to the point where it's like every quarter, every
86:31 half every six months I'm going to go back I don't spend very much time on
86:34 Indie Hackers. I'm going to go back to Indie Hackers just update some of our
86:38 stuff. I'm going to go back to Product Hunt, our initial product hunt launch.
86:42 I'm just going to update a bunch of our stuff in our in our next newsletter. I'm
86:48 going to ask I'm going to put a PS and say, "Hey, we're a small company. We
86:50 really need some reviews on Trust Pilot." I'm this is just all the
86:56 approach of just moving things forward in different areas throughout my day and
86:59 my week. >> I'll tell you how my vision for how I want to approach this.
87:06 I'm totally failing at this >> right now. But the what I really want to
87:12 happen in 2026 sooner rather than later is and I think that this is like a general strategy
87:18 that maybe other people have been using and and can use which is like focus on
87:23 your primary one or two things where not only is it natural and easy and and plays to your
87:31 strengths as a creator like for me it's like video and podcasting right
87:35 >> like YouTube and podcasting like the >> but also like it's also where you like
87:39 to hang out. Like I like to watch YouTube, you know. >> Um Twitter, I I like to scroll it and I
87:45 tweet from time to time, but I I'm not a high volume tweeter.
87:47 >> Yeah. >> Um >> LinkedIn I never touch. Facebook I never
87:52 touch. >> You know, so I'm naturally terrible on those platforms from a strategy
87:57 standpoint. >> Um >> so I I do like it's the age-old like
88:05 repurposing concept, right? Um, and I I've been doing a little bit of that
88:08 like where we cut my my editor will cut shorts out of my long forms to to but
88:12 I'm only publishing them as YouTube shorts. I should be publishing them on
88:15 other places. >> I should be So my my thought now >> is again I I I do I really do want to
88:26 use AI agents for this. This might be like the number one use case for me for
88:32 for use for for OpenClaw is like >> I'm still going to create content the
88:36 way that I I've always naturally created it. YouTube videos and podcasts and I
88:40 don't want to change that. I like doing that. That's that's always where my best
88:43 ideas are going to come out. >> Mh. >> Um and I want those to come from me, the
88:47 human Brian. >> Yeah. you know. Um, and so like I was saying before, like if
88:57 >> even if it ends up being me or even if I create social accounts for my agents,
89:01 which I plan to do, you know, like >> um like >> I still want them to be able to like
89:06 what you talked about like like you you jump around and log in and tend to the
89:10 garden on Reddit and tend to the garden on Kora and on Blue Sky and on YouTube.
89:15 >> Like I don't do a lot of that. I just stick to my one or two channels that I
89:19 that I just like to consume anyway. But if but if the agent can prompt me to say
89:24 like, "Hey, you haven't posted to indie hackers in 6 months." Y
89:28 >> um but here's something that might fit for you or here's something that you
89:31 should reply to. Y >> you know, um like that's something that
89:35 like a a good team member would be able to watch and prompt me to go do.
89:38 >> I mean, and I would use that too. I think that would be helpful for me as
89:42 well. I think an easy way for anyone to start is to just like I just said on
89:48 Google, what is Brian Castle known for? And >> product I service.
89:52 >> The interesting thing is at least on my screen the the number one kind of feature
90:02 snippet is from uh tutsplus.com in vat. >> So there there's a
90:07 >> I mean that's like 15 years ago. Th this is why you need to start googling
90:12 yourself asking Gemini. >> What was it? Like a like an article that
90:16 I that I guest wrote in 2011. >> Yeah. Yeah. Yeah. Here's a here's a nice
90:21 uh fresh face of >> how to use WordPress or something like
90:23 that. Like >> it's uh writing a and designing a killer
90:28 headline and how to scale and grow your online business by systematizing.
90:35 >> Um so that's one thing they pulled. then your website, then your LinkedIn
90:39 profile. Brian Castle Builder Method. So, at least you've updated your
90:42 >> I've updated the company that I work for. >> I'm noticing a lot of people uh I won't
90:46 mention names, but there's a lot of people who have not updated their
90:49 LinkedIn profile. >> Even just to update your job and your
90:55 company will help a lot with this. >> Yeah. Like you're you're right. Cuz even
90:59 just people that you meet in the world, like they're going to Google you and and
91:02 and also like professionally like people are going to just check you out on
91:05 LinkedIn like what what company is that person from? Like >> Yeah.
91:10 >> And I I also think it's it'll bring up some other things like your Mixer G
91:14 interview from 2017 is on here. >> Um it I think >> that's like eight years ago.
91:19 >> That's a good prompt though to say, "Oh, you know what? I should just reach out
91:21 to Andrew and just see if he wants to do a follow-up episode." Um because then we
91:27 have an updated version of this thing that's already getting pulled.
91:31 >> Uh so yeah, and even like some of these past interviews, like there's one from
91:34 2022, um where it says you're the founder of Zip Message. If you know that person,
91:40 you can just write a quick email or get your agent to write a quick email that
91:43 says >> it's so tiring to go out and get and like change all those names and links
91:47 and like >> but even just to write the email to at least say, "Hey, I just wanted to know
91:52 if you could update your your the interview with something and 10% of the
91:56 people will do it." But >> I've received those emails from, you
91:59 know, because we have random links on our Clarity Flow blog >> and like I'm too lazy to go update it.
92:05 So I'm I'm going to ignore that. I mean, this is part of the work right now. And
92:10 so, you know, there's part of gardening that sucks. Pulling weeds is not fun.
92:14 So, another ranking factor that's just showed up in the last couple months is
92:19 now Facebook discussions and groups are are ranking factors in Google again. And
92:24 I don't like Facebook, but that's just part of the garden. I got to go on there
92:29 if I want to win in this space. And the way people are finding Transistor is by
92:34 googling podcast hosting and seeing what and they're like, "Oh, a Facebook
92:38 discussion." I got to be there. That's just the job. I think you
92:42 >> But I also think that like fundamentally like like like yes to all these things
92:46 that were saying that you got to go technically go to these websites and be
92:50 active or respond to stuff, but that's not enough to me. The the way more
92:53 important than any of that >> is just be interesting. >> Yes. like you got to be doing some and
93:00 and like people hate to hear that kind of thing. >> Yeah.
93:05 >> But but the re but the thing is your product has to be interesting now. It
93:09 can't just be another another product that does the same thing as all your
93:13 competitors. You have to know the reason why it is interesting to your to the
93:17 customers who love your product the most. Right? like that. Like, yes, that
93:22 sounds like salesy and markety and and and you're self-promoting, but at the
93:26 core of that, you're understanding why why your best customers are paying you
93:30 the money that they are paying you. It's cuz they really resonate with some idea
93:35 and you you peel that back 10 layers deeper. Then you get into like content
93:41 that is not salesy, but it's it actually means something to people, you know,
93:45 like that's that's where that's what people actually connect with. I mean,
93:51 you know, like it that you have you have to just really like care on that level.
93:56 But once you do, then then you can start to find the opportunities like where
93:59 where does it make sense for me to like actually be helpful.
94:03 >> Yes. You know, this is another we we probably want to wind down here, but the
94:07 other key part so interesting from that Lars Lafrren interview, I think this will be a a an
94:14 earthquake for some people and some people have already realized this
94:17 intuitively. He basically said if you're going to do organic now
94:24 really one of the only approaches that works is founderled marketing.
94:29 He said that was that's always been a superpower but now it is the interesting founders
94:38 who are being themselves on Twitter, LinkedIn, YouTube, whatever who are
94:46 generating kind of the the the best organic footprint to get found on Google
94:53 but also via LLMs. and his the the big line that uh that I shared I think it's
94:56 at the beginning of the episode is he's like you either have to be the
95:00 influencer or you have to hire the influencer. He's like that's those are
95:04 the two options for organic these days. And he said if I had a SAS company with
95:10 a marketing team basically every person I hire on the marketing team would have
95:15 to agree to posting personally on some social network. So not he said
95:21 branded channels are dead like your branded YouTube channel like your
95:24 company YouTube channel >> that's stuck and we've seen this like go
95:29 >> like there are some branded stuff that comes up where it's still a person
95:31 >> that's right >> like >> uh clearly they hired somebody and and
95:36 like their video is answering a question that I happen to be googling.
95:41 >> Sure. Like Lar is basically Jeffrey we go to the Tailwind channel versus Adam
95:44 Wan's >> YouTube channel >> there's a big difference. uh in in terms
95:50 of engagement as well. Uh >> to that point though, the um I
95:54 definitely I I could see how like the founder thing is is the thing that works
95:59 best, but I I don't know that I totally >> like yes, it's technically possible for
96:05 a company to like hire a YouTube influencer type person to be the your
96:11 sole job is to be the person who creates YouTube video. I know like
96:16 companies hire like developer education educators like I guess that's a thing
96:20 that sort of works but >> Anthropic I think is killing it on this
96:24 front where >> they have their whole real team >> Twitter accounts
96:32 >> and they have super interesting >> Twitter accounts and you can even tell
96:36 that like members of their team are not longtime Twitter people
96:40 >> like Boris Churnney the creator of Claude Code like he just joined joined
96:43 Twitter I think like 3 months ago. >> Yep. >> You know, um but he's not the only one.
96:49 There's like 10 more more than 10 anthropic employees. Yeah. On Twitter
96:54 and and yeah, they're they're promoting cloud product updates, but they're
96:59 talking about what's interesting about them. And a lot of a lot of them are
97:03 like the person who is responsible for this new feature is the one who writes
97:06 the tweet about what what was in interesting about it. >> Yeah. And I mean like I started creating
97:11 a list of on on Twitter to follow just those people. Um the cursor team is also
97:16 pretty good at this. I think they have a a fewer number of people on their team
97:21 doing it. But this is like real content talking about what's actually
97:25 interesting about their product, the real problems that they solve.
97:29 >> It's like real. So you don't have to be the CEO. You you can have your team
97:35 >> Yeah. that it goes back to this thing >> does that like the the the person that
97:38 everyone sees representing HRES I can't remember his name but that's not the CEO
97:43 that's that's their CMO I think >> I didn't even know that until recently
97:47 right like I I like for many years I thought he was the CEO you know um
97:52 >> yes I another company that's been good at this forever
97:57 >> Base Camp you got Jason Freed you got DHH you got Ryan Singer you got Jamus
98:01 Buck there's all these old names of people um that were representing the
98:05 company >> more the some of the some of the guys like recent like with their recent stuff
98:12 I feel like there Jason and David are seem to be a little bit more open to
98:16 having their team be more visible like sharing like the designer behind
98:20 >> fizzy or whatever shared a thing and like >> um
98:24 >> it's be and and and by the way the reason it works is the thing that you
98:28 and I have been talking about basically since the beginning of this podcast
98:34 which is in especially in an age of AI and synthesized content, human contact,
98:42 human connection, uh being interesting as a human online is going to win. And so if you have the
98:50 whole anthropic team posting on Twitter all these interesting things about their
98:54 work, about what they're discovering, little hacks, oh here's how we work with
98:59 Claude, and here's a nice little list. Of course, that's going to do well and
99:04 way better than if someone sees like the Claude Code account. It's like, okay,
99:08 some corporate shill is just shilling out, you know, whatever. No, people want
99:13 real human beings. They want to hear from Richard Branson. They don't want to
99:18 hear from Virgin Mobile uh Twitter account, right? So >> yeah, this is
99:22 >> and clearly there are different channels like like their their Twitter game is
99:25 really aimed at the developers using clog code, but then they're going to do
99:29 this Super Bowl commercial that where they're which is fantastic by the way
99:33 like the hits that they're they're doing on Open AI for doing ads.
99:35 >> Yeah. >> Um so that's obviously for the masses,
99:39 for the consumers, you know. I um yeah, I think this is going to be a very
99:43 interesting time and again it's disruptive because >> asking your whole team to post on their
99:52 personal channels uh is a big ask for a lot of teams. >> Um
99:57 >> I know I know we're going long. Um do you have a do you have like
99:59 >> Yeah, I think I don't think I have a sauna appointment or anything. Let me
100:01 just check my calendar. >> No, no sauna. No sauna emergencies.
100:05 >> No sauna emergency. >> I got one more thing you mentioned. You
100:10 mentioned Caleb Porzio. >> Oh yeah. Yeah. This episode. Yeah.
100:15 >> I I want to speak a little bit to to what he was saying, but this is more
100:17 general because I hear the sentiment come up a lot, right? >> Which is, okay, all these people keep
100:23 talking about AI every day, but where are the products? Who's actually
100:27 building products with like real products with AI? Real SAS tools with
100:30 real customers. Like where where are all the products? Right? Like I hear that
100:34 all the time. Um, number one, there are real products that are 99 100%
100:46 coded with AI today. I mean, I I started a new uh show called Builder Stories,
100:51 interviewed Arvid Khal. >> 99% of his code is is AI. He's he's
100:59 doing serious MR with his pod scan. Talked to Brennan Dunn. Same thing.
101:01 Write message like, "Yeah, it started years ago." Handcoded. It's Brennan Dunn
101:06 solo, him and agents doing all the work on this is a real product with tens of
101:10 thousands of MR. >> Yeah. >> Um I I I have weekly episodes coming out
101:15 showing real builders how they are really I talked to Colleen Schnetler
101:20 last week. That one just dropped. Um >> real builders doing real products. It is
101:25 really happening. I I just did a private workshop with a team that who who's
101:30 doing serious MR for for a niche industry. Mhm. >> This is a team of eight engineers who
101:36 are all using cloud code. Um it like it it might not be the loudest thing you
101:41 hear on on Twitter, but like it's real now. Like >> uh you know and and there isn't going to
101:46 be this line in the sand and saying like that product is AI and that one is not.
101:51 All the products, all the engineers are using cloud code now. It's just a a fact
101:55 of life. That that's >> that's the first thing I I just wanted
102:00 to to say. But I was hearing Caleb I'm I'm such a fan of of first of
102:06 all how not only Caleb's work and and and the products that he does, but like
102:09 the way that the way that he does this solo podcast and he's so open and and
102:13 honest and and it puts so much personality into it. I really respect
102:18 that and he's really good at it. Um and I've been hearing his episodes, I
102:22 especially tune in when he's talk when he talks about his journey with AI
102:25 lately. Mhm. >> I think Adam Wan's journey also plays into this.
102:29 >> Mhm. >> You know, Caleb on his on his recent moment was talking about is is AI really
102:36 actually speeding me up? >> Yeah. >> Um it makes a lot of things easier, but
102:41 but what I've actually built and shipped this even faster. Um I could sort of see how that would be
102:50 the experience that someone like Caleb would have with it. And I'm not saying
102:54 that he's using it at any like rudimentary level. He's probably using
102:58 it super at a super deep level. >> Yeah. >> But the difference between what he is
103:05 doing and what most other product teams are doing is that Caleb's work his
103:11 Caleb's product is code. Livewire like his product is Livewire is code or or
103:17 Alpine or you know like >> Flux like he is creating code products
103:24 >> which means he has to be super hands-on with exactly how his products are
103:28 crafted at the code level. >> Yeah. but product people who are
103:34 creating job tobone >> interfaces that real customers not that
103:40 his customers are not real but like business you know businesses have use
103:45 cases to use SAS tools that you know like >> your customers on transistor don't care
103:50 about how the code is crafted underneath the inter interface y
103:54 >> you know um and that's obviously not to say that like it's bad code or people
103:59 building with clawed code don't care about the architecture. They obviously
104:04 do and it's gotten much better now, but >> but there's just no question that teams
104:11 can ship and build real products for real customers like a thousand times
104:16 faster now. Yeah. You know, um >> it's just that it's just the nature of
104:19 the product. Like if you're using AI to like to to refactor a service class that
104:26 that that that like powers one component of a code related product.
104:32 >> Yeah. >> The you're only touching a tiny thing.
104:35 You're you're not you're not actually shipping an end toend solution for a job
104:42 to be done product like those it it just really comes down to the actual
104:46 >> use case around around the product like so. Yeah, like if if you're really just
104:52 using it to as like a at at the code level, >> like I could see how that would that
104:58 would slow things down. Yeah, I think I mean the thing I appreciate about uh
105:03 what Caleb was saying, which I think we do need, we we need a counterbalance to
105:07 all the hype and excitement and put this into my veins and like let's go is is
105:13 some thoughtful reflection of like >> how do I feel about this? And and and I
105:19 mean the challenge and there's still lots of unanswered questions like
105:27 what we don't know like for sure there's going to be you know uh 6 months 12
105:32 months 36 months from now there will be a a whole bunch of these apps that have
105:36 been uh built with cloud code not even opening an IDE nobody really knows what
105:42 the code looks like that will start to have problems that all of a sudden cloud
105:48 code can't fix. And >> I would push back on that though.
105:50 >> Yeah. >> Because I do think that these engineers
105:56 do know how the code is architected. They're just not writing the lines of
105:58 code. >> Yeah. >> There there's a new there's a new type
106:05 of engineer. There's the there's the vibe coder who has no idea what their
106:09 what their codebase does. There's the hand coder who's resistant to AI and
106:14 they are they are behind the times now. It's 2026. You are not doing it the
106:18 professional way if you're not using AI. >> Yeah. >> But there but the the more the now it's
106:25 becoming mainstream where you are a professional full stack engineer
106:31 designer developer product person. >> You know how your product is
106:35 architected. You know how database modeling works and interfaces and user
106:38 experience. You know, you know how all those dots connect. You you you are putting that knowledge
106:48 into specs and into prompts and into and and and using cloud code to bring those
106:53 ideas to life like so much faster. Like and like that's the thing like they they
106:58 can write code. They can read every line of code and a lot of times they don't
107:02 have to read every line of code to to trust that they've crafted their cloud
107:07 code systems in a way with the right training, the right prompts, the right
107:11 systems, the right specs. Yeah. they they they they're now at a level that
107:15 that they trust that it's actually writing and and architecting things the
107:19 way that they expect without even needing to to of course they have like
107:23 code review and test suites and and >> you know QA processes but like
107:27 >> yeah there's of course there's a lot of code that I don't review that I still
107:31 ship you know but I know how it's built you know >> yeah I I I I can understand though
107:36 people's feeling of like where's this going because there is this so let's see
107:41 even put the code issue suicide. There's still there's so many like even
107:47 me I'm just like like >> it is scary especially the the open
107:50 class >> you know I've I've always said listen people you can vibe code an app all you
107:57 want but the true product people will still stand up maybe that won't be true in six months
108:05 maybe there won't even be we won't even need apps because everything will be an
108:09 API that is just talking to your everyone's personal cloudbot and we have
108:16 no websites, no apps, no SAS, no anything. Like that's all theoretically
108:23 possible, right? Like why why have uh podcast hosting when you can just get
108:27 OpenClad to just like, "Hey, build me this thing and I'm going to record the
108:31 first episode." Now, >> I think it's like yes to all really. I
108:35 think that I think I think it's like on the one hand, I I think it's overhyped
108:40 that SAS is going to go away so quickly. I don't think it's going to go away so
108:44 quickly, especially for wellestablished companies. Yeah. And especially for
108:48 companies that serve businesses um that and and businesses, >> it's almost like the larger the company,
108:56 the more behind the times they are with with adopting uh internal AI AI
109:00 solutions, right? Yeah. >> Um, but at the same time, I I do I know we're rehashing stuff that
109:08 we've talked about before, but I do think that there's going to be more
109:15 there already is more of a any new idea that comes out. Everyone has has the option of like go
109:24 use that new product or just vibe code a similar thing that's perfect.
109:28 >> Mhm. like and and I think that pattern is going to play out again and you're
109:31 already definitely seeing it in open source stuff like things like agent OS
109:36 there is like a thousand other agent OSS yeah you know and it's literally like
109:40 every developer's like personal opinion on how spectriven development should
109:45 essentially work it's like and you know um and that's that's part of it is like
109:50 that's great like like I don't like >> um I was talking to Bruno Borenstein on
109:54 on the builder stories uh podcast yesterday like he he just created
109:58 an open source thing that that creates a cananban view in in the CLI. It's like a
110:03 CLI canban coder thing and it's just like like cuz that's how his brain
110:08 works, you know? Um and and it's like >> um >> but and so here here's this is
110:13 >> so like like Fizzy or even like even the video podcasting thing, it's like yeah,
110:18 I look at that I'm like I could buy that. >> Mhm. >> Or I could vibe code it
110:22 >> or I could build Yeah. Um >> and and here's my thinking.
110:28 In the short term, I think I I have a fair amount of confidence that a distribution is going
110:37 to matter more than anything. So the reason Agent OS is more important than
110:41 all the other ones is agent OS has more distribution. More people know about it.
110:46 That is going to be an advantage. But the other thing that kind of worries
110:49 me, and I'm sure other people are feeling this intuitively, and I think
110:53 it's one reason everyone's so panicky and like feeling all this pressure right
110:58 now, is I'll say the quiet part out loud. Part of me is feeling like I need
111:03 to make as much money as I can right now >> before it all comes crumbling down
111:06 >> because it feels like this is the this is the window and my worry
111:12 >> is that in 12 months, 24 months, 36 months all of a sudden
111:18 >> the only people making money will be digital ocean, AWS and open AI
111:26 anthropic etc. there will be no other businesses except for you know these
111:30 basically hosting and infrastructure companies and LLM companies and I I I
111:39 that just saying that is just like I part of me is joking but there's this
111:42 other part of me that's like feeling this pressure like I need to make as
111:47 much money as I can right now because I might need to like provide for my kids
111:53 and their future for a long time. You know, I was hanging out with a with a
111:58 SAS founder at Big Snow last week. Um, and he runs a very successful SAS. It's
112:02 been it's been well established for a very long time. Has a great customer
112:08 base and it is still growing in 2026. >> And he was saying how and and they sell
112:12 to like small business owners, uh, a lot of like non-technical customers.
112:17 >> And, um, he was saying that he's he's he was sort of expressing
112:21 the same type of thing that you were just saying like like kind of kind of
112:26 like I just don't want anything to change because it's been so good for our
112:29 business for the last 10 years. Like why does this AI thing have to shake
112:33 everything up? I'm I'm kind of scared for the future. A lot of uncertainty.
112:37 And I was saying to him and I and I believe this. I'm like, you know, of all
112:43 the people that I know, I feel like he's one of the most safe.
112:47 >> And I think that about you, too. podcasting space has its own unique
112:52 competitive challenges I would say but the fact that you have an established
112:58 SAS with with a wellestablished customer base means >> means that like
113:04 >> and a footprint that currently LLMs are highlighting
113:09 >> yes and exactly same thing with him too and it's like
113:11 >> um >> I just wonder I think the anxiety it's much more to me to me more of the
113:18 anxiety is This is why I'm not starting a SAS business in 2026. Like, if you were
113:22 trying to start one, >> Yeah. >> Uh-uh. Like, like, I know that there's a
113:27 lot of people who are and and you can be successful. There's a lot of different
113:29 ways to be successful, >> but I'm not I'm not playing that game
113:35 anymore. It's just too hard to do that in my opinion for me. for me, you know,
113:38 um, >> and and like to me it's like I could see why it would be so difficult because of
113:46 AI to try to break through as a new player. But if you are established, you
113:51 have thousands of customers like >> I mean customers are not are not all
113:56 going to vibe code your app tomorrow. That's just not going to happen. The
113:59 stress is is like first they came for the sites that ran on documentation like
114:04 Tailwind, then they came for the established course brands like Larcast,
114:10 you know, like now we're seeing it was like before it's like no, this you're
114:14 going to be fine. Like Jeffrey Way like you're you'll be fine. You know, Larcast
114:18 is an established brand. It's loved. Everybody who starts in Laravel starts
114:24 with Lariccast and he's just posted this video on his YouTube. I'm done. It's
114:31 like he it it's a he's he had to fire 40% of his staff. He's he's like this is
114:36 having a real repercussion. >> I got to check out that uh that video. I
114:39 didn't see it. I I did hear about the layoffs though. the um you know I also
114:43 think that there's a spectrum of product type and it goes back to this is an
114:47 age-old like product marketing thing which is like the more essential your
114:53 tool is to to the the way that revenue comes in or the way that you like like
114:58 podcasting any sort of hosting based business is re to me has always been
115:03 fantastic because it's like >> your your podcast is hosted on it like
115:06 to to can like the the pain of cancelling is very disruptive even if
115:10 you stop podcast and you still want your podcast to be online.
115:11 >> Yeah. >> Um and you can say the same thing about
115:15 any other tool that is like really part of the way that you sell your product or
115:19 the way that you deliver your services to customers. The ones that are much
115:23 more that I would say are much more at risk now are the nice to haves, the the
115:30 to-do list apps, the notetakers, the >> the little project management stuff, the
115:36 you know the the little super niche tool that does this one utility task over
115:40 there. Like yeah, I'm still subscribed to a lot of them, but like at some point
115:43 there could be a little paper cut that I'm like, huh, I'll just I'll just vibe
115:47 code a better version. You know, >> I I do think there's something about
115:50 this pressure that a lot of people are feeling, which is, you know, for Adam and
115:57 Tailwind, part of him was like, listen, he was making a lot of money for a long
116:01 time and that window was open for a while and then he saw that window start
116:05 to close. And I think a lot of people can look at that and go, "Fuck, are we
116:12 in another like is this the window? Is it open now?" And and then
116:16 >> but also like what Adam is doing with UI, is it UI.sh?
116:20 >> Yeah. Yeah. I mean, he's >> what a domain. I mean, but like this is
116:24 the kind of thing like I'm not saying everyone needs to go out there and start
116:27 a new brand or start a new product, but like he he was obviously very public
116:32 about the business challenges he had and >> and he's like taking them headon. Him
116:36 and Steve are learning how to design with AI and launching a whole new
116:40 product and brand line out of it. Yeah. and like like just like sort of like I
116:46 hate this term but like lean in and start to be creative and like yeah the
116:51 especially for him and where his product is code related and and and clearly it
116:57 has had an impact on his business. It's like okay time to time to get creative
117:01 and and actually figure something out. Yeah. >> Yeah. Time to pivot pivot into something
117:06 related. I think I think in the short term that will work. I just I want to
117:10 recognize for a lot of people out there, I think there is a lot of people that
117:14 are feeling like I got to work super hard right now. I got to code my brains
117:17 out right now. I got to figure this out right now because eventually
117:25 the only thing left to do is going to be plumbers, you know? And that's part of
117:29 what Caleb was talking about is like, man, it's like eventually is the only
117:33 thing left is going to be plumbing. >> Start start an HVAC company.
117:37 >> I mean, yeah. All right, we should probably leave it
117:41 there. Thanks everyone for joining us here in the chat. We had all sorts of
117:45 great people. Keith and Dave Jones and Ryan Hefner. >> Great. Bo is here as well.
117:50 >> Great to see you all. >> Good times. Good to catch up.
117:53 >> And uh I think I'm here next week. Are you here next week?
117:57 >> Yeah, I'm here uh all of February and then I I'm supposed to have a trip out
118:01 to Park City in March, >> but there is no snow in Park City. We're
118:04 trying to figure >> No, a lot of people go. So, we'll be
118:09 back February 12th. Good to see you again. Yeah. See you next week.
118:11 >> Later, folks.
