文字起こし 🔗
WEBVTT
1
00:00:00.000 --> 00:00:48.000
I think one of the big headwinds in SaaS right now is like, that's like you putzing around in Claude code for a couple weeks. Like, that's not a product. It's not even really a feature hardly. It's like I talk to a lot of founders through Tiny Seed and like, outside, where they're just like an objection they're getting in sales right now or a churn cancellation. Reason is like, I just built this in Claude code. I think the more risky thing is that you build this thing and nobody cares. Hello. Welcome back to Rogue Startups. I'm joined once again by Colin Gray, co host of the season. And I think today we're talking about updates. What are we building? What are we working on? What's top of mind for us, I think, particularly around AI and how it fits into our business lives. Colin, how's it going?
2
00:00:48.000 --> 00:00:57.000
Yeah, good, good. Just reflecting on old age, making my memory completely rubbish, which is partly related to what we're talking about today, I guess.
3
00:00:57.000 --> 00:01:06.000
Yeah, Yeah. I can't relate at all. I'm 46 and not going insane, I'm sure. No, Tommy, you know, like, I've got
4
00:01:06.000 --> 00:01:10.000
a year on you, Craig. I've got a year. Okay.
5
00:01:10.000 --> 00:01:40.000
I mean, like, just to jump in, like, I think like something that we talk about both with our team and with like, my. My consulting clients, where I do. I do a bit of consulting with, like, companies to help them, like, make this AI transition that we're all going through is the goal is to take the cognitive load off so you can focus on other, hopefully more important things. And I think that's what we're talking about. Right. With some things that we're working on is how can I just not have to worry about remembering stuff? And that could be pretty powerful.
6
00:01:40.000 --> 00:02:36.000
Yeah. For me, there's that part which is just remembering. The thing that I was alluding to was we talked about two months ago about what I think is a really good idea for an event, actually, and we. I just never followed up. I just. I saw a note earlier on in looking at. In fact, you know what it was? I'm using granola right now for my note taken. And they seem to have just added a new feature which surfaces what you talked about previously. So when this meeting was coming up, it was granola, actually popped it up. So this is an element of what we're talking about actually in, you know, systems that surface previous information at the right time. Um, and the other part of it is, to me, it's our inconsistency So I think what I've been trying to build is something to help me make sure I'm more consistent with the content I'm trying to put out. So newsletters, podcast episodes, that kind of stuff. Is that like. That's kind of what was on my mind around it. Was that the same for you?
7
00:02:36.000 --> 00:03:04.000
Yeah, I think. I think that's. I think that's part of it. I can talk from, like, I use granola as well. I'm recording this in granola, even though we're recording this in Riverside. I can talk about, like, how I'm using some of those recordings for content later as well. But I'd love to hear. Yeah, because I know that you have a whole system that you're working on for this podcast to, like, repurpose content, actually start sending emails, all these kinds of things. So, like, I'd love to hear what you're working on, and I can kind of chime in with some things I'm doing too.
8
00:03:04.000 --> 00:05:06.000
Yeah. The last two weeks, for me, it's been those two systems. You're. You're spot on. It's like what the one I've been working on longer is. It's just kind of a code name in my Claude code setup called Video Assist. And it just really started as a way of trying to take videos from whatever call recording tool I was using. So it could have been Alitude, could have been Riverside, could have been Streamyard, could have been whatever it is. But trying to make it relatively agnostic and then being able to do some really light cleanup, some auto cleanup, like this is. It's kind of what I built. I was aiming to build an alitude for podcasting, but I'm just trying to do it for YouTube now. It's like to take away a lot of those automated sites like audio cleanup. Well, I mean, I suppose a lot of the tools do that now, but it's more the trimming, so trimming the start and the end, taking out some of the mistakes, you know, and learning the bits that I want to take out. Not just like crutch words and things like that, but the fluff, you know, the little tangents, the phrases that I use more than anyone else that really don't need to be there, that are wasting time. So that's. That's kind of how it started. But nowadays it's kind of a whole publishing workflow. So it takes the video, it does a lot of that cleanup, but it then chops the video up into. It takes a 60 minute podcast video and chops it up into seven or eight segments based on what we talked about through that. And that's long form segments. So it does do the clipping, but I just don't think the 60 second clipping type stuff actually works that effectively. This is more 5 to 10 minute segments that actually cover a topic much more in depth and is quite question specific focused. So it's actually really good for SEO in terms of YouTube particularly so YouTube discoverability because it's generally answering a really specific question at the very start. So that's the kind of core thing I've been working on and been able to then turn episodes like this or my golf podcast, both sides of par into that kind of YouTube output. That's one side of it for sure.
9
00:05:06.000 --> 00:06:14.000
It's really cool. And I can't wait to. You're working on some of this and we're working on some of this for this show and I can't wait to see like what it does because I think our YouTube channel is not amazing. Like the YouTube channel for Rogue startups is not. Is not amazing. I just haven't put much effort
Load full transcript
into it. But from running Castos, I know like YouTube is the growth engine for a lot of, a lot of people's shows and so like we probably should do more to optimize for YouTube with like an actual proper, good intro and stuff like that. But my comment about this is like, you're saying this and I'm like, whoa, like we could like license this or build this into Castos or this is a whole product. And then I'm like, that's exactly. I think one of the big headwinds in SaaS right now is like, that's like you putzing around in Claude code for a couple weeks. Like, that's not a product. It's not even really a feature hardly. It's like I talk to a lot of founders through Tiny Seed and like outside, where they're just like an objection they're getting in sales right now or a churn cancellation reason is like, I just built this in Claude code. I don't have an answer. And like there's no like path out of this. It's just an observation that like, you know, it's used to be like, I can do this, I can buy HubSpot or I can buy Salesforce and now I can just make my own.
10
00:06:14.000 --> 00:06:50.000
Anyone can build anything. You're right. I think it's almost a kind of weird, unique person who actually spends the time on that though. Like I've spent. I've spent quite a lot of time. I know you said futzing around, but it's like, it's a hassle. It's like, it's fun. Like, I find it fun. But I think for a lot of people it be a hassle. And there's a lot of iteration and there's a lot of. It's just personal as well. Like, actually, the only way I've made this video tool useful to me is by building in a whole lot of things that are really personal to exactly how I make a podcast and the subject as well. Like, actually it's really finely tuned to golf content in the way that I make golf content interesting.
11
00:06:50.000 --> 00:07:14.000
Say more about that. Cause I'm really. I did. I built one of these systems out with hyperframes and I thought, like, I spent a whole afternoon on it and that, like, for me, that's all like a pretty long time. And it was not good. Like, the result was not good. So I'd love to hear, like, how the topic changes, like the setup of the. Of the project, because that. That's super interesting. Like, what about golf makes your tool need to be unique or whatever?
12
00:07:14.000 --> 00:08:31.000
Well, I think it's almost a taste thing. I think it's almost. I have trained it to. So I've trained on the format of our show, which is a variety type show. We talk about four or five different subjects. I've told it. We do gear segments, we do skill segments, we do course reviews. And for each of those types, I'll say for course reviews, if I veer off into, you know, I don't know what another course that's related, which is what I find I often do just cut that out. Just stick to the one course. Because actually if you're putting that in the title of the YouTube video, that's what the people want to hear. They don't want to hear the other ones. And similar with gear reviews, I'm like, often in a gear review, I'll end up talking about something else as well. And often actually that's kind of extraneous. So cut that out. So there's a lot of, what would you call it, training in terms of like, you know, refining the prompt in a way. So here's a particular type of content, here's exactly how I want that to work, and here's common things that go wrong with that type of content. So probably in some ways you could generalize that. Maybe that's what a lot of people do. But I think a lot of it is Personal to the screw ups that I make when I know I'm making a podcast that I know work in a 60 minute podcast but don't work in a 5, 6, 7 minute concise clip. Does that make sense?
13
00:08:31.000 --> 00:09:12.000
Yeah. And so if you, if you were to load this podcast into that project, it, it would probably be subpar. Subpar, Below average. But, but if you gave it some of that context around like things we do or things we don't do or ways to. Ways to. And you're talking about like within our like production service we always say like there's the obvious stuff that we take care of, the ums and the ahs and the misspeakings and weird pauses and stuff like that. And then there's the subjective stuff which is like, hey, that thing you just talked about is just not interesting, like we're just going to remove that. Or you kind of repeated what the guest said and that's not productive. So you're kind of creating rules for each of those scenarios.
14
00:09:12.000 --> 00:10:10.000
That's it. I think that's exactly it. I think that's probably what the more generalized SaaS tools will become though over the next few years in that that's the exact type of use case whereby, you know, Salesforce or whatever actually becomes more useful because it does start to actually learn much more about the user and it does start to, you know, there's the whole kind of idea that UI becomes much more fluid because, you know, I don't use this feature, that feature, that feature. So why does it even appear to me in a tool that's run essentially by AI? It should be able to redraw that UI for me. It should be able to recalibrate how it manages things. So like a more general video editing tool should be able to recalibrate itself like that for any user based on what their preferences are. So I don't. Yeah, I mean I think it definitely would be possible to generalize what I've done for anything, but I will, I'll report back. I'll find it interesting actually to start using these videos on the one I've built and see how well it works.
15
00:10:10.000 --> 00:10:20.000
Yeah, I'd love to. Are you using, in terms of like tooling using CLAUDE code? Are you using Hyperframes or Remotion or what?
16
00:10:20.000 --> 00:10:51.000
Remotion, yeah, so it's Remotion with an FFMPEG and CLAUDE code as the development platform it's all running on. I mean you're probably more deep in this than me, but it always suggests you use a SUPABASE for this thing and
Vercel for that thing. And I'm like, yeah, just do it, don't care, just make it work. And it's all just running on my computer. I've got little bat files, bat files to start up all the services. I've got Docker running on my computer just now, all that kind of stuff. And so it all just runs locally that way. Which I.
17
00:10:51.000 --> 00:11:30.000
Isn't it a trip out like I was talking to someone about this. Isn't it a trip how like the most important kind of like social network for so many of us is GitHub now, you know, and we all have Docker and we all know what ffmpeg. I mean you and I know what FFMPEG is because we ran podcasting companies, but not a lot of people do. I, it's crazy the fluency of tech these days. I talk to like even non tech, you know, my, some of my consulting work, very non technical folks and they're like, oh yeah, like I, you know, we created an MCP for this thing over here and built a center tool and it's run on a cron. And I'm like, it really is amazing how technical everyone can and should be.
18
00:11:30.000 --> 00:11:45.000
Yeah. If you're interested. I think there's plenty of people out there that aren't really interested, but I think I became interested. I'm sure you were the same in working with, being non technical founders, working with engineers and trying to figure out whether they were doing the right thing. Basically.
19
00:11:45.000 --> 00:11:47.000
Yeah, yeah.
20
00:11:47.000 --> 00:11:54.000
Like I learned the term Docker back then, I remember and I was like, I have no idea what that means and I'm not going to spend the time to find out, but at least I've heard the term.
21
00:11:54.000 --> 00:13:11.000
I was talking to, I was talking to someone a couple weeks ago and the, the reason that they wanted to talk is they're like, I, we need to hire like a new developer. Like a new non technical founder, they need to hire a new developer. They don't know how to hire developers, which is understandable. Like it's a hard, it's not a thing you do often. It's, it's kind of a hard skill to learn. And I got thinking about it and I didn't really kind of like finish my thought because we got onto other things. But I kind of have this thought that like and the developers are going to be yelling at me in their car after this. Like the, I'll say, the value of a good developer is different than it was two years ago in that I think even for production level SaaS applications, you can just throw Claude code at it or like Codex or like now. Man, I. We had a whole chat in the Tiny Seed Slack. Everyone is using Grok. Everyone is using Grok. They're like, it's awesome. You have, you know, five times the number of tokens for the same, you know, subscription spend, whatever, that doesn't really matter. But like, you don't have to know how to do stuff. You can just throw AI ADD. I think that's true 90% of the way. Like maybe 90% of the times and 90% of the depth of a problem. Like, is it architected perfectly? Does it work?
22
00:13:11.000 --> 00:13:12.000
Yeah.
23
00:13:11.000 --> 00:13:25.000
Okay, then kind of. Who cares? Like, is the code perfect? Who cares? Like, you probably need to make some really, really smart decisions at the early phase. But like, if you're using SOPA base versus I don't even know what, like DigitalOcean or whatever, like, does it matter?
24
00:13:25.000 --> 00:14:33.000
This is the question I have actually about whether these tools will be released. Like my. I think that we probably haven't. Having run products in the past, we probably have an advantage in that we know a lot of the gotchas to look for. So I would immediately start thinking about, like, what data am I holding? You know, how am I protecting that data? Thinking about penetration testing, security testing, like, making sure it's running these things once a day. Like, are there any vulnerabilities? Check the news, all that kind of stuff. And then you come to like checkouts and worrying about like onboarding and like product tours and all that kind of stuff. Like, that's the kind of thing I think that an enthusiastic builder doesn't really think or know about. And actually probably, you know, maybe that's another thing. They could get coached by AI, but it's the stuff that turns a good idea for a product and even a technically proficient product into a successful product. I think because it's the thing that makes real users be able to use it and for it to stay up robustly and not completely screw you over because it leaks all the data that the users put in.
25
00:14:33.000 --> 00:15:13.000
Yeah, yeah. I mean, I guess I'll break this into two scenarios. One would be, you know, taking over Kastos or Alitu as like a person who's never developed a product anymore and you're a solo dev and you're just throwing codecs at it all day. Probably not ideal. I mean, like, our application is quite complex and I'm sure yours was too. But if you're just starting out, I do Think it's sufficient because like, you don't know a lot of this stuff, but it kind of doesn't matter because really the important thing is like getting customers and marketing and all that kind of crap and then you can figure security out from there. Like, I think the more risky thing is that you build this thing and nobody cares. Then you build this thing and it's not secure or performant or beautiful or efficient or whatever.
26
00:15:13.000 --> 00:15:19.000
It's a good point. Yeah, yeah, for sure. Yeah. The market, the code has never been the hard part, has it? It's always been distribution.
27
00:15:19.000 --> 00:16:38.000
Yeah, yeah. I want to share something I'm working on because I think it's related. So mentioned like cursor and like, yeah, again, like I'll probably get canceled for this. Like I'm a big Elon fan. Like I'm a big Elon fan for like 80% of kind of who he is. Like the political stuff, I can kind of leave. But you know, as a inventor and business person, I think he's kind of the whatever da Vinci, like of
our time, I think. Or Edison. I'm using cursor since Grokbot came out. So I'm on the $200 a month plan and I have like a whole group, I won't call them army, a group of bots. And so this is a replacement for like openclaw or Hermes. And the big difference, I don't know if you played around with the big difference is like as opposed to openclaw and Hermes, which are open source, you gotta host em on a server and security and all that crap. Like this is platform, so it's like using Claude code or Codex, like in the app, like in their desktop and mobile apps. And it's awesome, man, it works, it just works. The memory is really good. The integrations to like MCPS and native toolings are all there. It's cool. I'm doing stuff that I never got to with openclaw and Hermes in three days that I've been using it.
28
00:16:38.000 --> 00:16:51.000
I've heard of Grok, obviously I've never used it actually. Never actually come to use Grok at all. But Grok, what's the difference between Grok and Grokbot then? Is it just a kind of layer on top that helps it run in a more automated fashion?
29
00:16:51.000 --> 00:17:41.000
So I think what Grok bot is probably cursor cloud agents just in a chat interface. I think that's probably what it is under the hood. And cursor cloud agents think of like, I don't know if you Guys ever like, had something like had a job and then you would put it in like a GitHub action or something to run every morning or whatever. That's all it is. It's like a thing that can run like on a cron. I think they're working on accepting webhooks like from other events or other services now. Okay. It's based in the cloud though, which like, as opposed to openclaw and Hermes, which like you could run on your computer, but then if your laptop is closed, it's not running via freaking Mac Mini or whatever or hosted on a VPS. Like this is on the, you know, the XAI or SpaceX servers. It's running on Linux.
30
00:17:41.000 --> 00:17:54.000
So Cursor is just essentially the IDE you're running it in, is that right? It's just the app that you're linking in the model and using to interface with the model or is it more tight?
31
00:17:54.000 --> 00:18:11.000
That's what Cursor is. Yeah. So Cursor is a fork of VS code and then it has a bunch of automations and like agent orchestration. That's Cursor. Grokbot is basically just a chat. It looks like WhatsApp is the model.
32
00:18:11.000 --> 00:18:12.000
Yeah.
33
00:18:11.000 --> 00:18:26.000
And then you can do plugins and interactions and it can put, you know, like you need to drop in an API key. It drops like a little input thing right in the chat so you don't have to go to like settings and you don't have to worry about like a dot ENV file and all that kind of crap.
34
00:18:26.000 --> 00:18:29.000
So it can do that for you without having to put it into the chat as you're not supposed to do.
35
00:18:29.000 --> 00:19:11.000
Right, Right. So, yeah, and this is one of my, this is one of my kind of issues with it
is like there's like Cursor on your computer. There's Cursor Cloud, which like runs these, you know, cursor agents in the cloud. And now there's Grokbot, which is like doing other things like, and then like, is it x AI or SpaceX or Grok or Cursor or what the shit? Like, I mean, I think all these companies have so many products that like, it's confusing. It gets super confusing. Like I want to set up a thing like where do I do it? How do I do it in the Cursor, Grok and you know, ecosystem. It's, it's the most complicated. I think Codex and GPT actually have done a good job of like reasonably keeping things easy to understand.
36
00:19:11.000 --> 00:19:19.000
I read just the other day that apparently There are Now something like 50 different versions of Copilot. So Microsoft have done that, but like 10 times.
37
00:19:19.000 --> 00:19:22.000
Not surprising, right? Like everywhere.
38
00:19:22.000 --> 00:19:50.000
And there's even like non AI related ones. It's just all over the place. They've not done a good job. But yeah, I know what you mean. It's like, I suppose it's because it's all moving so fast and because people don't really know where it's going, it's really tricky to. I suppose maybe even a worse thing would be just to invent a new name for all the different variations of this one thing in like one in the cloud, one on your computer. I don't know whether that be better or worse. Yeah. So what are you. So are you still using Hermes or openclaw?
39
00:19:50.000 --> 00:20:41.000
No. Yeah, no. And truth be told, like, I haven't used them as like primary things in a while. Like I had a Hermes, I still have a Hermes agent that runs and does a couple specific things. But like, I think if you want to get the most out of it, the goal with these is like to use them for everything. All of your kind of knowledge and coordination work. So like my, my bots now like have access to my email. They like triage my email and they tag stuff and they put it in folders and they archive stuff that doesn't need to be archived. Give me like an update every morning. Like, hey, you have this thing, it conflicts with this other thing you're talking to Colin at 9 o', clock, docs, Calendar, email, Notion, Slack. Like it has visibility into everything. And so I can be like, hey, create an issue for this thing and pull context from this email over here, like from my phone, you know.
40
00:20:41.000 --> 00:20:57.000
Yeah, yeah, that's cool. Does it feel more safe to do that with all of your platforms? With something like Grokbot, when it's built into one single cloud platform that you relatively trust than an open source system that you've put on your own server?
41
00:20:57.000 --> 00:21:09.000
Yes and no. Yeah, right, like yes and no. If somebody hacks, you know, the SpaceX or cursor servers, then we're really, everyone's really screwed. Do I trust them more than I trust myself? Like, yeah.
42
00:21:09.000 --> 00:21:35.000
Yes. Yeah. Because I mean, that's exactly the barrier I'm coming up against just now. I can talk about what I've been trying to build in openclaw in a minute, but it's a lot around. Just being that I think that's the thing you said a minute ago is spot on. It needs to be the one place for everything. It needs to be where you go for everything. It needs to be where your transcripts end up, where your ideas end up. You're out for a run and you have someone pop into your head. You need to just type it into that one single place.
43
00:21:35.000 --> 00:21:36.000
And.
44
00:21:36.000 --> 00:21:49.000
And then the output as well, I think needs to be the same place. But I'm just so reluctant to link all of my platforms into openclaw right now. Yeah, just. Yeah, no, I mean, for all the obvious reasons. Yeah.
45
00:21:49.000 --> 00:22:50.000
And just in terms of like, everything living in one place. Like, I, you know, I'm on the $200 a month cursor plan. I've moved Codex and. Or ChatGPT and Claude down to $20 a month just to keep them going. A month ago it was the opposite, basically. Like, I wasn't doing cursor at all. I was on the $20 a month ChatGPT plan and the $200 anthropic plan. Because I can get both now. I can get like the, you know, always on agent in Grokbot and cursor as a, you know, when I want to sit down and like, code, you know, make an application or, you know, contribute to. To Castos. So, and just specifically what, like, what I'm doing. So, like, I'm recording this call. I record a lot of my calls. Like, I get permission to record them, but I record all my calls. And then it pulls in every Sunday, all of my calls and gives me, like, social content ideas. It doesn't write it for me, but it's like, hey, you talked to Colin about this thing. That would be an interesting tweet. And so then I can just take those seven ideas, bang out like social posts for them, and then we're good.
46
00:22:50.000 --> 00:23:58.000
I've built a kernel of that in that whenever I. The kind of initial idea was to take the granola transcript from any coaching call that I do or a call like this could be as well. And to mine, I've got a kind of content mine Chat, I think I called it chat mine skill. So it's got kind of a set of specs around the types of content I want to be publishing, the audience, ideal audience profile, all that kind of stuff. And so it comes up with five or so ideas for every single one. The trouble. So that worked quite nicely, and it does work quite nicely. The trouble I had, and what I've been trying to solve more recently is you're scheduling one that you seem to have cracked, which is I then have to still remember to sit down and do that. I Want something to be my executive assistant, my personal assistant, who then on a Tuesday morning every single week, sends me a message that says, oh, look, here's four tweet ideas from these three conversations. Here's four LinkedIn posts from these three conversations. And then Wednesday morning it says, right, it's time to publish a newsletter.
Here's a couple of things worth writing about from last week as well. That's the part I haven't cracked. So is that kind of where you've got to.
47
00:23:58.000 --> 00:24:44.000
Yeah. And with the Grokbot, you can just say, like, run this every Sunday at 7 o'. Clock. And it just does it, it sets it up. But I have it. I have it in Century, I have it in GitHub, I have it in Slack, I have it in Help Scout, you know, so not like surveilling the team, but instead of me, like, logging in and looking at those places, like, I got my update this morning and I basically said, like, hey, everything is good. You know, like, hey, Dev team's working on this. There's a couple tickets in Help Scout. I know that Kelly and Jimmy will work on those. And I can come do this podcast knowing that, like, everything is good. I don't need to. I don't need to worry about it. I have a couple emails I need to check out. But it's not like the end of the world. I think it's that the big difference is that, like, push versus pull. Right. Like, if it's pushing to you, you don't have to go pull that information.
48
00:24:44.000 --> 00:25:15.000
Yeah. Particularly when you're working on more than one thing as well. That's what I'm finding just now, because my work's a bit more irregular right now, and I'm working on a few different. It's just. I mean, it was always hard to stick to a schedule of content publishing, wasn't it? But, like, it's even worse now. So. Yeah. And I basically need a boss. That's what I'm trying to build for myself. It's a boss. It's like someone's come along and tell me, here's the four things you said you were going to do last week. Here's the thing you told Craig you would do the week before that. Here's how you need to do it this week. If you get this done on Monday, this done on Tuesday, then we're all good.
49
00:25:15.000 --> 00:25:28.000
How do you try to keep yourself organized like that? Because I find it really hard, you know, like, you're a solo founder, you don't have a Co founder or a boss. How do you keep yourself accountable maybe is the right word.
50
00:25:28.000 --> 00:26:57.000
I have always had a kind of a system whereby I plan a chunk of time and then break it down and deliver that. So in my early days, like before I had much of a team, it was a month at a time. So I would have a month plan and I would sit down and I treated it like a proper company. And I, I had a board pack. So as if I was actually running a board. I had a document which had commercial goals, technical goals, tasks that those break down to, blah blah, blah. So I'd have the kind of overall monthly strategy in this document and I'd look at that at the start of the month, every single month. And then I would take that and I would break it down into the weeks. So every Monday I would then look back at that doc from the start of the month and I would pick out four or five tasks and I literally had a spreadsheet where. Which was color coded. So like Monday was Monday morning was planning. So that had a yellow on it for admin. Monday afternoon though was commercial. So that had to be something growth related. And so I would slot in a couple of tasks from that, related to the commercial goals from my board pack into that slot in the week and then the rest of the week panned out like that too. And I found that was the only way I could keep on top of that was how I like published two or three blog posts a week and a podcast episode a week and a video a week for years to grow the podcast host because it was on the calendar, it just had to be done. And I kept referring back to that monthly plan every single, every single month. And I moved towards. It was more. I think I've heard you talk about this, Craig, like using the EOS system. Is that right? Am I right?
51
00:26:57.000 --> 00:26:59.000
We have, yeah, we have, yeah, yeah.
52
00:26:59.000 --> 00:27:42.000
In the past I. So like I moved more towards their format which was a wee bit so you. When I had a team essentially it moved out to a two month. We tried quarterly, tried three months, found that was way too long to create kind of the larger goals. And so we ended up on a two month cycle. Almost like base camp's shape up process where you've got like six, six weeks working, two weeks recovery. Yeah, so that was where we ended up and. But it's still the same thing really. But I just planned it with my team leads instead I would take my head of growth and we'd say here's the growth strategy for the two Months and it's like the top level stuff. And then every week we'd come back to that plan and say, right, what are we bringing off this, this week to do? Yeah, so that was kind of the rough idea was that how do you manage that stuff?
53
00:27:42.000 --> 00:28:55.000
So I'll say for, like, for myself and for our team, we don't, we don't have any. We have done, we've done it all. We've done what, we've done what you're talking about and tried it all. And we've been super big on EOs and KPIs and rocks and all this kind of stuff. I guess we still very much have KPIs. Like, you know, everyone reports things like uptime and tickets and response time and marketing metrics and stuff like that. We don't have fixed schedules because what we found is from like a product in development, but also like marketing stuff takes different amounts of time. And so like why do we need to force the development team to fit this feature into a six week bucket? And then they're stressed and like, why are they stressed? Just because it's week five, you know that they have to get something out for next week. So we don't, we don't do that. Like other folks I know and like some folks that I coach are very big into. Like we have two week sprints and everything fits in a two week sprint and you figure out what you can fit in the sprint and then you ship it. I just like, I don't know, maybe I'm too chill, but like, I don't see the point in that. But then we, we do have this like ramp up work,
ship and cooldown phase. Like, like the shape up process, it just is over different periods of time.
54
00:28:55.000 --> 00:29:00.000
Yeah. So you're not strictly stuck to six weeks. You just choose a length of time based on what the task is.
55
00:29:00.000 --> 00:30:15.000
Yeah, like, like at Castus, we're building, we're building. I haven't talked about it. Here we're building this thing called Castus Studio. Right. It's like Riverside. Right. Recording, editing, all in the dashboard. Don't have to go anywhere else. That's a big feature. And so that's months of work. Why would we bucket that into, you know, two six week cycles? I don't see the point in that. For myself, I use the calendar a lot to block my schedule for things that are important. And so I'll have a lot of stuff in the calendar that's not meetings but is like important time. So like I'll block You know, Tuesday morning or a Thursday afternoon or something. And I usually the. The best thing for me is like going somewhere else, like the physical different spot, especially for like writing or thinking. I'll go to the coffee shop. I have an office, a little like office in town. I'll go to the office if I need to do certain things. But actually, but actually the best, the place I get the best work done is in the chair downstairs with the dog sitting next to me. Early in the morning on the weekend usually is when I get my best work done. Like six o' clock in the morning with a coffee on a Saturday is when I get my best work done. Or 10 o' clock at night outside
56
00:30:15.000 --> 00:30:22.000
of the normal schedule when the team's not working away and requiring some attention.
57
00:30:22.000 --> 00:30:47.000
But I think that not at the desk and like literally in like a recliner, you're just like physically in a different mode. You're like my desk. I'm like, I got the fucking lights and I got the microphone and I got all the big monitor and stuff. But when it's just me and the laptop, like, I think I'm more calm, I'm more introspective. I just think differently. I don't know. Do you always work at the desk or do you work other places?
58
00:30:47.000 --> 00:32:17.000
I do, actually. Yeah, but I get what you mean about the time. Like I, I actually, I work terribly if I'm away from my normal task because I just have everything here that suits me and I can, you know, and really simple example is I use a lot of voice transcription nowadays to, instead of typing. So whether it's like, whether it's writing a little note for myself or update my calendar or blah, blah, blah, anytime I need to type something in, more often than not I will hit my keyboard shortcut and use whisper flow to type in that way. And if I'm away from my desk suddenly the mic's not in the right configuration or there's people around me, so I can't really do that. And you know, it's just the little niggles like that where when you work in the computer like eight, ten hours a day, I think actually make a big difference over time. So I love being at my desk and I've set it up the way I want it. But on the other hand, I do get that about time. Like, I do love it when you're, when you're outside of the normal working hours. It just feels like there's something about it feeling like bonus time. And even if it's weird, I maybe should have done this more. But if I had taken an afternoon off to go and do something, whether it was taking the kids to do something at school or whether it's just someone fun, myself going for a cycle. So I end up doing a bit of work in the evening. So it's not bonus time at all, it's catching up. But it still felt like it because it was outside of the normal hours and it just feels. I know what you mean. It's like there's a subconscious, feels a bit more relaxed. It's funny.
59
00:32:17.000 --> 00:32:19.000
Yeah.
60
00:32:19.000 --> 00:34:27.000
Something I want to come back to actually that you mentioned before, like the timelines thing. I totally get what you mean about like the six week cycles, two week sprints. Why try and stress yourself out by fitting things into these? I think I tried a few different ways. I think I ended up, I ended up coming back to actually doing these time bound activities mainly because actually I found it was a way of chunking work that forced you to actually define it a bit more strongly. I think that was the benefit I got from it. And we never got too stressed about missing deadlines. Actually in the end it was more the thought, right, we've got a two week sprint coming up or we've got a six week cycle. What can we build in six weeks? And if we did actually do a full six weeks we'd always break it down into two week sprints anyway. Probably even less to be honest. What can you do in a week? I remember seeing a, a quote from the founder of Sierra. Do you remember Sierra Games? From our youth. Did you ever play any? It was a really successful games company. Well, not that successful I suppose. I don't know if it's still around or not, but it was at the time certainly. And the founder wrote quite a good book of which one of the big quotes that I remember was something around the lines of a one hour task takes two hours, a half day task takes about two days, a one day task takes about a week and a one week task takes forever. So it's like an exponential increase. The longer a task takes, it doesn't just scale linearly, it scales exponentially. So the longer it is, it just means there's way more unknowns in it. There's way more things that can go wrong. Probably you haven't defined things very well because you think that's going to take a week. So there's probably a bunch of stuff and to make you estimate a week or longer means there's probably a bunch of stuff in there that you haven't actually figured out properly. So that was why we always tried to define stuff down to even just day tasks in the end, like to try and make it the most predictable thing possible. And yet still things always went wrong. And a day
task ended up taking four days sometimes. But I found it was a lot less common when we started doing it that way.
61
00:34:27.000 --> 00:34:35.000
Interesting. I was going to ask how important is the velocity of shipping product, but I don't want to get into that.
62
00:34:35.000 --> 00:34:51.000
Yeah, we could come back to that. That is an actually really interesting thing to come back to. I was talking to another founder earlier on today and he was talking about some product keeps shipping features. They just keep putting things out there. And I'm like, you've already got a really good product here. You need to just market this thing.
63
00:34:51.000 --> 00:35:54.000
Yeah, yeah. I mean, I think that's the challenge of, like, a lot of companies. And you see this, like, you know, there's been the bundling and unbundling of SaaS a few times. It's this, like, sine wave. And I think we're in the bundling phase now because it's so easy to build a product. We all think, and we're no better or different. Like, you think, oh, well, we need recording and editing if we're going to be a proper podcast tool. And like you, you guys did the same. You did recording and editing. You add hosting. Riverside has done the same thing because it's easier than marketing is the. Is the real reason. And like, I don't think it's wrong entirely. Like, we look at it and we like, hey, this is valuable. Would people pay $10 a month for this? If so, that's easier. We have thousands of customers. That's easier than getting a thousand more customers. So I think that's just the math a lot of people have done is like, hey, it's easier to build more product and get our existing customer, deliver more value to our existing customers, and they pay us more than to go get more customers.
64
00:35:54.000 --> 00:36:03.000
And there's a perception that building more product or improving the product will get more customers anyway. But actually that's not the case because you then need to market the improvements.
65
00:36:03.000 --> 00:36:38.000
It's both. I mean, it's not guaranteed, but it is both. Like, for us, legitimately, this will unlock potentially like a big group of people, like a big group of Libsyn customers or Buzzsprout customers or whoever. Customers are like, gosh, I don't want to use two tools. I don't want to record in freaking Zoom and then put it into whatever and then upload it to Libsyn. Like, can I just do this all in one place? Some people will be like, that's dumb. Your editor is going to suck. And like, yeah, like, relative to Adobe Audition or Final Cut Pro or whatever, it will. But, like, that's not the goal. The goal is not to compete with them.
66
00:36:38.000 --> 00:36:57.000
No, it's a savvy point. More marketing probably beats more features than usually presuming you've got a really solid product. In fact, in some cases, adding more features just muddies the waters. It actually can kind of ruin your existing marketing, can't it? Because it can change positioning, it can confuse things around who you're actually aiming for.
67
00:36:57.000 --> 00:36:58.000
Yep, Yep.
68
00:36:58.000 --> 00:37:55.000
I wanted to ask you a couple of more things about the Grokbot stuff. Like, one of the things I've been dabbling with. I heard somebody use this phrase the other day, actually, the knowledge librarian for your company. I'm thinking about, like, the knowledge librarian for me. So it's like, one thing I've always struggled with is ideas. As in, like, when you have an idea, how do you log it and how do you make it come back up again in future? At the right time, you can have the kind of standard product approach where you've got a backlog. So any idea that comes up just goes in the backlog. But then you have to set up a schedule to review the backlog. You have to have criteria to be able to upgrade someone from the backlog to active development, all that kind of stuff. And it just isn't really that it's hard. The backlog always just is a graveyard of nonsense that people just never look at again, how are you, like, when you have an idea just now, like something you're not wanting to do right away, what do you do with it? Is that something that goes into your Grokbot kind of assistant now? How. How do you resurface it beyond that?
69
00:37:55.000 --> 00:38:29.000
If so, like, if it's YouTube video, I have a notion table and I put it in there. If it's a product idea or a marketing approach or something, I'll stick it in Slack to our team. And if it's like a tweet or a piece of content or something, I'll just do it right then. So I don't have what you're talking about, like an idea library. I think that's probably really hard because, like, you attach so much intuitive context to that little nugget of an idea that would be hard for an LLM to understand and pull out at the right time.
70
00:38:29.000 --> 00:39:03.000
I wonder if that's more possible. I think your three examples there were really good, actually. That might make that more possible. Now, in that I want to have, like we said earlier, I want to have this one place where I can just voice note into. And that's it. That's the idea. Logged and it's filed. That's one of the big things, I think, making it filed in the right place. Often one of the biggest hassles is me thinking, right, this is an idea for. You just had a YouTube video there. So this is an idea for a YouTube video. Right. I've got to remember, where did I log that stuff again? Was that in Slack or was it Evernote or was it Notion? Oh, it's Notion. Where in Notion was it again? What's it called?
71
00:39:03.000 --> 00:39:08.000
I got five tabs of Notions and three different accounts and all this kind of bullshit.
72
00:39:08.000 --> 00:40:00.000
Yeah, exactly, exactly. So you want to be able to just pop it into Telegram or Slack, however you communicate with your, your AIPA and for it to say, oh, YouTube video. Right, that goes there. There's a rule for that. I always know where that
goes. Similar for your ideas for product that's always posted to this Slack channel. So I think that's one part of it actually that makes life a lot easier is just having a thing that knows where to store your 20 different varieties of information. Probably 100 different varieties. Eh? But the resurfacing part, yeah, I haven't. Haven't quite figured that out yet. I wonder if that is something. Just about as long as you then still use that assistant to assess ideas. So you say, oh, I've had an idea to do this, or I want to try and grow the app here. Then it starts to resurface those things automatically, as long as it's the same place again.
73
00:40:00.000 --> 00:41:12.000
So this is one thing with Grokbot that's really different than Hermes or openclaw is. Like, there was a lot written for both openclaw and Hermes about their memory system. Like, it ingests so much information. How does it kind of prune the crap it doesn't need and keep? And I don't say tag, but contextualizes the bits that it does need so it can retrieve it later. That's really opaque in Grok because you don't know, like, you can't see what it's done. You know, it's memories and stuff like that. Like, I just yesterday had it start, like posting a markdown file to a GitHub repo every day for all the Stuff that it did. And it's like five lines, you know, for each of my kind of. I have a. I have an agent for myself and I have an agent for Castos. And below them they have like sub agents. But it's like five lines. I'm like, that is not all that we talked about yesterday, you know, like. But then how much, like, does it just export the whole contents of, like, our discussions and all of it model calls and all that? Like, that's crazy because then, like, how's it going to read through all that after two years? I don't know the answer, but that's the balance. It's like, how can it only pull the right stuff that it needs to at the right time? Yeah, I don't have an answer for that.
74
00:41:12.000 --> 00:42:13.000
One of the things I've built so far is the first scheduled part is now on a Monday, my assistant pops a post in Slack. I've started using Slack rather than Telegram for communicating, so that I can set up different threads and different channels for. It just makes. Just organize that a bit more easily. And now in the kind of the organizational channel, it'll pop in a. Here's the stuff you got done last week. And I'm using todoist. Actually, I've always used Todoist as my task manager. So I've got it linked into the API of todoist. So it populates my todoist, it manages my todoist, it watches what I've done in todoist, like what I've ticked off. So it can tell me what I did to help have a bit of context around progress, all that kind of stuff, which I do like. But it can also then tell me, like, what projects are making progress and what are not, which is quite good. And then it'll ask me a few questions about prioritizing. This week. It's like, all right, you did a bit on video assist last week, you posted the golf newsletter, but you didn't post the promo for rogue startups. So do you want to try and mix that around this week or you
75
00:42:13.000 --> 00:42:14.000
want to get off your ass and
76
00:42:14.000 --> 00:42:47.000
actually do exactly get. Get this done? So I get. I think my hope would be that if I continue to do that, it gets better. I can refine how it does the prioritization, how it helps me with that. And then that means at the start of the week, I always find on a Monday, I'm like, all right, well, I'm thinking about this and I'm thinking about that. And then it'll bring up Any previous ideas that relate to that? I'm hoping that would be the concept, but that does depend on it knowing. Yeah, exactly. Being able to delve into the memory, being able to manage the context, all of that kind of stuff.
77
00:42:47.000 --> 00:43:27.000
Yeah, just from a high level, like what I would say is like I would treat this like I would our team where like I don't want this to be my job to do your job. You know, I say that, I say that to our team in a very nice way. But like this is your job. Like my job is to help you do your job, but this is your job. And so in this case the job of your bot is or agent is to do this stuff. And so like I would kind of turn it back on it and be like, this is the goal. This is kind of where I feel like we are today. I need you to like figure out like how to do this and I'm here to help you and provide context and stuff like that. But like I don't want to put the pressure on myself to figure all this stuff out. I guess is the short version.
78
00:43:27.000 --> 00:43:56.000
Yeah, it's a good point. Yeah, I can just ask it. Any ideas? I've actually I've been asking, I've been talking to it a lot about different organizational methods. You know, there's like, you know, there's the classics like get things done by. You've got the para method, you've got like a bunch of these different ways to organize information and retrieve information and then you've got these really complicated tools that do it as well. Like is it Obsidian? Is it the really famous note taking app that is organized in a really strange way. You've used it?
79
00:43:56.000 --> 00:44:03.000
I've used it a little bit, but yeah, we have. One of the guys who used to be on the team was huge into it. Like every talk about your brain, like
80
00:44:03.000 --> 00:44:45.000
everything was there, which seems really cool but potentially overcome. But no, I'm already finding benefit in having built the Slack integration so that now I've got that really easily on my phone and on my computer. So wherever I am I can really easily jot in a note. I can choose the channel really quickly in Slack as well to give it a wee bit more context that way. And then having this weekly catch up to like forcing me to think about it at the start of the week every single week, which is what I always tried to do, but I never really had a thing that knew
my priorities that would then kind of question it, you know, I think that's the main thing, isn't It. It's like questioning, right? What are you going to do this week? All right, why that?
81
00:44:45.000 --> 00:45:18.000
Interestingly, like, I find this is funny. I find Claude to be kind of a jerk like that. Like, I. I've asked it, like, hey, keep me accountable. Don't let me chase a shiny object and stuff. And now basically, anytime I bring something up, it's like, why are you doing this? That's dumb. Like, stay focused on this other thing. And I'm like, look like shit changes, man. Like, I should be able to. To like, evolve too, I guess. Like, one question on your open claw, like, what model are you running or using it on? Like, on Codex or what?
82
00:45:18.000 --> 00:45:28.000
Yeah, on Codex. Yeah. I had it on Claude before. They kind of changed the terms of service and everything. And I thought I'm just going to change it over to Codex because they've said you can use it freely, so.
83
00:45:28.000 --> 00:45:41.000
Yeah, yeah, it's. I mean, the Codex, like, yeah, Codex is really good. Like as of today, like 5.6 sol is, you know, I think comparable to Fable. It's really, really good. Like, it does everything that, like, you need it to do.
84
00:45:41.000 --> 00:46:00.000
Yeah, yeah. I found no difference, really, to be honest. And I'm interested to go and try the Grokbot approach, actually see if that's. See how it is. I installed Hermes, but I haven't linked it up with everything yet. So the plan was to migrate my current openclaw over to Hermes just because reportedly, I don't know what. Will you tell me the memory's a little bit better? Just a bit more aligned?
85
00:46:00.000 --> 00:46:24.000
Yeah, that's about the only difference to me is, like, it crashes a little less and the memory is better. I think if you want to open source, you know, version that you run and manage, I think Hermes is probably the way to go. I know a lot of people that got started on openclaw still run it and still love it though. So I don't think it's not, like, wrong. It's probably. It's probably just like the Claude versus Codex versus, you know, cursor Grok discussion.
86
00:46:24.000 --> 00:46:26.000
They're all small differences.
87
00:46:26.000 --> 00:46:29.000
Yeah, pretty. You can get the job done with all of them, maybe, is how I'll say it.
88
00:46:29.000 --> 00:46:38.000
Yeah, I kind of just keep dragging my feet because Claude is changing so fast and actually just building kind of all of it anyway. And I'd rather just stay in there because that's where all my stuff is.
89
00:46:38.000 --> 00:47:01.000
You gotta believe that, like, they're just gonna build this, right? Like, they kind of. They kind of did with Like Claude cowork. But like it had the big. To me, the big thing is like all this has to live on their servers for it to be really, really good. Because then that's like when, when OpenCloud came out, I was like, yeah, like the grown up version of this is going to be, is going to be anthropic. Building this in house and it's going to be awesome.
90
00:47:01.000 --> 00:47:29.000
Yeah. And it needs to, you're right, it needs to run in the cloud. I mean like I run Claude code on my main computer at home and I use it most of the time, but then I was away the last week and I wanted to still have the capability to run the projects, like to progress things a little bit, just to play around and it meant I just had to leave my computer on. I set up like tailscale so I can like link in via SSH really easily. I've got like a bunch of different tools set up so that I can screen share and run the computer at home. But it just.
91
00:47:29.000 --> 00:47:55.000
That's so silly, isn't it? I mean like, I know. Yeah. Yeah, that's silly. Yeah. Well, fun to catch up on this stuff, folks. Would love to hear your kind of thoughts and experiences with agents, how you're remembering stuff, how you're servicing it, how you're like keeping context long term. Drop a comment down below if you're watching on YouTube. If you're listening, hit us up on socials. I'm TheCray Hewitt on Twitter. Colin, where's the best place for folks to like reach out to you? LinkedIn.
92
00:47:55.000 --> 00:48:08.000
Yeah, LinkedIn's my most kind of common social media, so just find me on there. If you search Colin Gray podcast, you'll find me, no worries. Or if you just go to my name Colengray link, that's where you find just about anything. So all my social stuff and channels are on there. Colingray link.
93
00:48:08.000 --> 00:48:10.000
Awesome. Thanks, Colin.