Can You Vibe Code a Secure App?

November 19, 2026 • 00:54:36
Can You Vibe Code a Secure App?
Rogue Startups
Can You Vibe Code a Secure App?

Nov 19 2026 | 00:54:36

/

Show Notes

Craig Hewitt and Colin catch up after a stressful week. Castos was taken offline on purpose for about half a day while Craig's team implemented security patches for two large PHP vulnerabilities, and the two talk about security in a world where attackers and defenders both use AI. They move on to getting back to work after a holiday, being reactionary versus intentional, and whether strategies and plans help. Craig also explains Ultraclip, a new AI video editing service, and what it means for SaaS and for pricing a service.

Topics discussed

Takeaways

Chapters

View Full Transcript

Episode Transcript

[00:00:00] Speaker A: Colin, hello. Welcome back to Rogue Startups. Colin Gray, how's it going? [00:00:04] Speaker B: Yeah, not too bad, Craig, not too bad. How about yourself? How are you getting on this week? [00:00:08] Speaker A: Yeah, yeah, I mean, of course there's always. In a podcast, there's always the like, oh, before we started recording, we talked about this and I stopped this. So 12 years in, I'm like, finally a responsible podcaster. But it's been a very rough week. So yesterday there was a release of a two large vulnerabilities to php. I don't know if you saw it on the news, like government websites down. Castos was intentionally taken down for about half the day as we were kind of waiting on and implementing the security patches and. Yeah, just massively stressful. Just, just, you know, at one point I was, I was like literally shaking. Like, I was just so, so stressed. And I don't know any way around it other than get used to it, man up. That doesn't seem right. That doesn't seem like the right answer. But yeah, I think the reality that I hear a lot of folks kind of pointing to is like, hey, this is just like what life is like now. I think this was a responsible disclosure, from what I understand. But then they patch it and when they patch it, obviously PHP is open source. You can see what they changed. You can kind of reverse engineer and like, you know, the bad guys can kind of like orchestrate an attack. And I'm sure it was leaked ahead of time and all this kind of stuff, but I'm very happy with how the team did. We blocked traffic to the app basically at like Cloudflare level. We were waiting on the issues. We had upgraded to the latest PHP version right before then, so we tested that kind of. As we were waiting. We upgraded once the actual Docker image we use anyways, we did all the stuff communicator with customers. I think customers were appreciative and understanding, but still just like. And then later on in the day, like Railway, like a pretty popular managed hosting provider was hacked and like API keys were, were revealed and stuff. And then they were blocking deployment. So even if you rotated your keys, you couldn't redeploy and it's like, Jesus Christ. Like, [00:02:28] Speaker B: yeah, [00:02:30] Speaker A: I think we always had a level of stress, like going back to when you started alitu of like, security. And we've certainly had our brushes with it, but it seems much more urgent now. [00:02:44] Speaker B: It seems so much more fast, doesn't it? [00:02:47] Speaker A: Yes, that's what it. Yeah, [00:02:48] Speaker B: yeah, just things are coming up much more Quickly and the severity. Maybe I'm no cybersecurity expert here, but having run a software app, my kind of outside view of it was you can do all. You can. You can do the penetration testing, you can do the checking, you can do all that kind of stuff, but there's only so much you can do because there's just so much buried in these heaps of tools that we use as part of our apps. Like you're saying it's php, there nothing to do with how you made your tool, is it? It. It's to do with the actual infrastructure on which it's built. So it's like it's almost a bit of helplessness as well, isn't it? [00:03:27] Speaker A: Yeah, yeah, yeah. I mean we literally were waiting on a guy who manages this particular. I don't even know what it is, you know, wrapper or instance or something like he wraps PHP in this way that we can use it very easily [00:03:45] Speaker B: as [00:03:45] Speaker A: a web app. Um, we're just waiting on this guy to deploy a thing to GitHub and Docker. [00:03:50] Speaker B: Yeah. God, can you. I don't know whether it's worth doing this or not. You tell me what you think, but maybe it's too technical. But is it worth explaining what it is? Because I saw a little bit in the news. So I was just saying to you before we got on the call, I'm just back from holiday today, so I missed a lot of the furor around this because I was [00:04:08] Speaker A: traveling. Lucky you. I [00:04:10] Speaker B: know. So. So [00:04:14] Speaker A: as far as like the vulnerabilities or like our technical [00:04:17] Speaker B: setup. Technical, yes. So. Well, no, actually, sorry, what I'm interested in more is like what would happen if you hadn't. So you took Kastos offline? [00:04:24] Speaker A: Yeah, yeah. What [00:04:26] Speaker B: would happen? Yeah. [00:04:27] Speaker A: I obviously am not technical for anyone who I'm like technical enough to understand, but I'm not a developer. The vulnerability was around large multi part uploads, which we definitely have, and an attacker could overload the memory, I don't know, of the instance or the server or whatever, and gain access to the server unauthenticated through that vulnerability. There's two? Yep, there's two. That, that, that was the one. Now, in terms of severity, that actually is not massively dangerous because we could just shut the instance off and spin up a new one because like we're on aws, we do blue green deployments, we just shut an instance off and it spawns a new one. If we didn't know this, they could install like crypto mining software or some crazy shit on there. They couldn't access the database, I don't think easily. So, like, in terms of really bad stuff they could do, I don't think. I don't. I don't think it's like that bad. But. But then, like, I don't, like. I don't know, man. Like Railway. Did. Did Railway get hacked by this? Like, they're pretty. Like they're a top 10, I would say, like managed infrastructure provider. Did they get hacked by this or something else? Like, there was a Node JS disclosure yesterday too, that needs to get patched. So, yeah, that's the very dumb version of what's going on. [00:06:12] Speaker B: God, it's hard, isn't it? I don't know. Do you have an equivalent of. In the UK if you set up a software company? In fact, if you have any kind of company that deals with customer data, which basically means any company ever, you have to register with a company called the. Or an organization called ico. What does that actually stand for? Ico. The Information Commissioner Office, I believe it might be. Is that something like that? Can't even see it on their website. It's not up top. It just says ico. But they're an organization that helps regulate data protection. Data management in the uk. [00:06:51] Speaker A: Yeah. [00:06:52] Speaker B: So you have to pay a fee. You pay like 40 quid a year and they register you as a company that deals with data and so they give you procedures. So if you ever discover any kind of security breach, the IC will have procedures that you have to go through and then report to them what happened. And then they basically analyze it and give you advice on what needs to be done to rectify it. And one part of that is that if you have been discovered to be, you know, if there's more you could have done, you get fined as well. So that's quite a common thing. Like, if there's a big data breach, emails escape. You might get some kind of fine as well. [00:07:28] Speaker A: Yeah. [00:07:29] Speaker B: Is there something like that in the US as well? [00:07:31] Speaker A: No, not that I know of. Which, like, wouldn't surprise me. I think the US is kind of generally less kind of regulated than most other countries. The only thing I could think of is like, we definitely have, like, cyber. Cyber insurance as a company. [00:07:47] Speaker B: Sure. Yeah. I suppose where my head was going with that is that you have to. So they have, again, they have procedures around reporting breaches. If you don't report within. I believe it was seven days. I could be wrong on that, but it was pretty quick. You had to report it Pretty rapidly. Then it gets worse, like fines and all that kind of stuff. I just wonder how many people report these things. And I guess over your way, even if you don't legally have to report it, there's probably some kind of regulations around how tell customers if your data's lost. I'm presuming there's some kind of laws around that. [00:08:24] Speaker A: I'm sure. [00:08:25] Speaker B: How many people discover these little intrusions? There's small things that can happen on hobby sites. Like you discover there's a little Trojan horse or something or other installed on your WordPress PHP server that's been there for a little while and then they could maybe get some commenters, email addresses out of it, something like that. There must be very few people who actually report those things compared to the ones that do. I wonder. [00:08:52] Speaker A: I mean, I think the scary thing about that, you know, is following Twitter last night just because I was maniacally plugged in yesterday, you saw people saying railway has been hacked, but they don't know it yet and haven't said anything. [00:09:11] Speaker B: Oh, right, okay. Yeah, [00:09:13] Speaker A: and I can see that. I can see because like, if I went to my team right now and I kind of did this morning, I was like, hey, I'm hearing that like, you know, people were exploiting this PHP vulnerability up to the 17th, so, you know, a week ago, is there any way for us to know that? And they're kind of like, what do you want? Like what do you. And not, you know, our team's amazing, but they're kind of saying like, there's not just like a button to push to say, like, is anything weird? Like how do you even look for the. Obviously like logs and data loss and weird stuff on the servers. But yeah, I don't like, I think that's part of the challenge is like, these are smart folks that are doing this stuff and it could be very kind of elegant and say like discreet, like what they do. [00:10:03] Speaker B: Yeah, yeah. The smarter they are, the more discreet it's likely to be. Because if it can hang around for years, then the more they're going to get out of it, most likely. Okay, [00:10:12] Speaker A: yeah, [00:10:14] Speaker B: maybe there's smash and grab type ones where they just get in and extract a million email addresses or credit cards if it's purely directly money based as well. [00:10:23] Speaker A: Yeah, [00:10:25] Speaker B: I know, it's terrifying. Do you think the optimistic view is that, yes, hacking is going to be quicker, more urgent, faster, all that stuff that we mentioned, because the nefarious folks have the tools at their disposal to perpetrate it, but the optimistic view is that we have the tools to defend as well. [00:10:47] Speaker A: Yeah. And that's where the positive and smart people from our world that I follow are like, yeah, they have the bots, we have the bots. So we're doing bot war kind of now. And I think that's true. I mean, we have had for a while, very extensive scans of our code base and it brings up stuff that we didn't know about. Now I have a Grokbot that scans four times a day, all of our packages and dependencies and platforms for any kind of security issues. We didn't have that before yesterday, so we're better today than we were yesterday. And that's what brought up like, oh, there's a node thing that we need to fix and there's all this kind of. There's all sorts of stuff that we need to fix now that we weren't really, really aware of. Because, you know, thinking back to like, a couple years, one of our developers would spend 30 minutes every morning just bouncing around to a bunch of websites to see if there's anything they needed to know about. [00:11:47] Speaker B: Yeah. [00:11:49] Speaker A: Like now you can imagine. Oh, Grokbot sees the thing, it opens an issue, it spawns a Codex instance, it fixes the thing, it bumps a version, it puts it on staging and we deploy it. Like that can all just happen. Of course, that's an oversimplification, but not that much. [00:12:05] Speaker B: Yeah, but the more there's the bot wars you're talking about there. I don't know. Again, it can happen so fast that, as in discover a vulnerability getting within the next five minutes. It seems like there will be much more of those types of breaches in Future. Just because [00:12:24] Speaker A: 100%. [00:12:26] Speaker B: Because it can be scaled so quickly, as soon as something like that comes out. And there must be so many. I was hesitating whether to say or not, but I know people that work in the NHS in the uk, so. Nhs, The National Health Service here in the uk, and see if you sign up to go to your GP or you go to the hospital or something, the care here is amazing. But as soon as you sign up for something like that, you realize how old the technology is that they use in these hospitals. Not sure if that translates to where you are as well, Craig. [00:13:03] Speaker A: 100% it [00:13:03] Speaker B: does. They are literally Windows 95 and stuff like that. You see stuff sometimes, which is just ludicrous. And some of the folk that I know in the nhs, they either work in it or close to it. They're absolutely terrified right now because the technology is literally so outdated that they just can't apply AI to it. They can't apply, get Fable 5.1 on or whatever and say how do we fix this? And it's like, well, you need to rip out the entire thing [00:13:35] Speaker A: from scratch. Write it in Rust. Yeah. [00:13:41] Speaker B: And that is a job for something as big as that. NHS is obviously one of the biggest flipping organizations in the world. That was a big job. But even for just corporations, 1,000 plus employees, something like that, ripping out the infrastructure structure of a place like that, any kind of larger company on the planet is going to be such a big hard job. So [00:14:03] Speaker A: yeah, I mean even us. I think Castos is a reasonably complex application. Nothing like the NHS or the IRS here or something. Even running payroll for a thousand person company that's like managing. Anyways, it would take us months to like, we're a pretty modern platform, you know, it's Laravel and runs on AWS and all this. But like say we wanted to rewrite the application in Rust or Node or something like that. That would be crazy. And I don't think it's any better. Like [00:14:40] Speaker B: yeah, [00:14:41] Speaker A: yesterday was PHP's day, but it's been JavaScript's day and Ruby's day and all these languages have their time, so I don't think anybody is impervious to it [00:14:53] Speaker B: totally. I think this is one of the defenses against the death of sass, to be fair. Kind of perversely, ironically, because obviously it's a way you can kill a SaaS product, but the security side of things is the part that I think so you can vibe code your own app, you can decide to release it to the public, you can potentially ask AI to do some security audits to make sure it's secure. But I have been developing tools for myself along the last three, four, five months. You have as well, Craig and I, quite often because of the experience I had running a software company similar to you. I'm sure you ask it to check on certain things, you know, some of the technical terms for it. You ask it to do certain like things that you know have been done by specialists and your own team and then it discovers that you say something, oh yeah, that's been there for the last two months, even though I asked you to do a security test like last week. But it's because you didn't ask for this very specific thing and because, you know, there's so much speciality in those kinds of activities that I don't think you can just ask an AI model to Say, make this secure. Is this secure? It needs much more specific guidance than that. [00:16:11] Speaker A: Too good this time. Yeah, [00:16:12] Speaker B: exactly. I think there's just. It's too general. Yeah. [00:16:15] Speaker A: Yeah. [00:16:15] Speaker B: I don't know. Do you think that's nonsense? Can you make a secure app vibe coding it these days without that knowledge, without that experience? [00:16:24] Speaker A: Yeah, I think you can. I think, like, the latest models are so good. One of our team members said something yesterday that I've thought about for a long time, and they said, we were on a call, just kind of like working through everything. They were like, you know, what it seems like is just these guys just throw so many tokens at a problem that something comes up. And that's kind of how I think about building product is like, I'm building a. I'm building a mobile app to track, like, my running and my working, my workouts. [00:16:56] Speaker B: Oh, yeah, Cool. [00:16:57] Speaker A: And it's. It's super dumb. Like, I'm super dumb with it. I'm just like, I want to make this app. And then it makes the app. I was like, oh, but that's not right. Do this instead of. And over three weeks now, it's pretty good. It's just because I've spent a bunch of tokens on it. It's not because I'm good. And I think you could make it secure that way. I think hackers can find vulnerabilities that way. They're just expanding the amount of effort on both sides. We can be smarter, they can be smarter. AI in the right hands is more effective than in my hands a lot. But even I could do a good job. Yeah, [00:17:39] Speaker B: I suppose if you go about it in a mindful, deliberate way and actually say, right, my activity this week is to figure out what security is required for this app. And you actually have that as a whole project. Like, talk to Opus, talk to Fable, talk to Codex, whatever it might be, and say, I'm developing an app. This is going to be open to the public. I want it to be fully secure, compliant with all regulations in my country, blah, blah, blah. Let's create a full plan, a strategy for how to do this. Tell me all the different activities a cybersecurity firm might go through, might recommend. What would a specialist do? Tell me all the different kinds of specialists. I think that a lot of that is about. That's a lot of what it takes to uncover these types of things. You need to get down to the utter detail of every single role and figure out what that person would do and then apply that. That's the Plan. That's what you need to get AI to then do. [00:18:33] Speaker A: Yeah, yeah, build a plan, Create a document so it can refer to the plan so it doesn't lose it. Right. Anyways, yeah, so stressful day yesterday. Today's better so far at least. But yeah, it's just the reality all of us are going to have to deal with. Even just like vibe coding. Our little side projects like that may be even worse because it's just you like. I have dozens of little websites and NEXT JS apps that I've deployed over the last year. Most of them I haven't touched in six months. [00:19:05] Speaker B: Some of them will be connected to your email, some will be connected to your calendar. Yeah. It'll have vectors into your personal data. [00:19:12] Speaker A: Yeah, not good. Not good. So, yeah, anyways, you mentioned you have been on vacation in Italy for a week and a half, which is amazing. How was it? [00:19:25] Speaker B: Yeah, it was great. It was lovely. It was, yeah. Eight days of just kind of wandering around, eating a whole lot of pasta, at least two or three pizzas a day, enjoying the warm weather as Scotland descends into the frozen north that we are every. Every single winter. So, yeah, it was lovely. But yeah, back today and back on it and yeah, yeah, the question I had around, it was really the question I have every time I get back from a holiday, which is really, what in the hell do I do day to day? What's my job? What was I doing before I left? What was my plan? What was my strategy? And figuring that out. Do you have a routine that you go through when you come back from time away around reorganizing, figuring out priorities, figuring out next steps? What. What's your kind of way of dropping back into it? [00:20:20] Speaker A: Yeah, I mean, I guess especially for you kind of coming off this, the exit and not having as much. I know you have some kind of responsibility, but like less than you, than you used to. [00:20:31] Speaker B: Yeah, [00:20:31] Speaker A: I can imagine. This is. This is pretty. Flexible, right. Can. Can be pretty loose, like your. Your work life. [00:20:45] Speaker B: But I'm trying to make it not so loose now. Actually. [00:20:47] Speaker A: I've [00:20:48] Speaker B: struggled with that the last couple of months because I want to start doing some work again, but if it's too loose, it just winds me up and stresses me out. [00:20:55] Speaker A: Yeah, I mean, what I'll say is I have a lot of demands on my time and attention. Customers need stuff, team needs stuff, partners need stuff to where it's pretty easy to get drawn back into stuff. Whether that's the right stuff or not. Like that, that's a different question. I also am pretty bad about taking real time off. We were in Europe for a week and a half this summer. I was totally unplugged for three or four days of that and then a couple of days of just my kind of check in once a day and keep the ball rolling on a couple things. Um, yeah, I think, I think like in the reactionary sense, you know, things like email and slack are kind of how I work and that's, that's super unhealthy, I'm sure. But like things that are important kind of live there or kind of are spurred by that when it. I think the bigger question maybe where like you, you are is like you are starting something new, you know, like you, you or expanding on something that you already have. So that's like proactive work. I don't have a great answer there other than like, I try to think back from like, where do I want this to go? And then like, what are the steps I need to take? And then like, what's the next step I need to take? So like, if it's, you know, you want to build out your coaching. Okay, I want to build on my coaching. I want to have 20 clients. [00:22:30] Speaker B: Yeah, [00:22:30] Speaker A: well, I have three. And whatever you have, you know, like, how do you get more clients? How do you have a system around this? How do you deliver better coaching? That's kind of how I think about it and that's pretty loose, admittedly. [00:22:44] Speaker B: I get this, I used to get this so much, particularly when I was working. Exactly the same situation as you, working with the team, customers, all that kind of stuff. When you come back from a week away, two weeks away, and you kind of want to. I always look forward to come back to work, honestly. I really enjoy holidays, but I'm always quite ready to come back into work and looking forward to routine again. And I sit down on the first day back and think, oh yes, fresh new page. Let's start some, let's figure out where we are. And I open everything up and it just smashes you in the face with notifications and you just start answering them and then five days later you're like, wait a minute, what happened? Don't even know what I did this week. And that's really hard. It's hard to avoid that. I think the. So about two weeks ago, it might have been, actually since we last spoke because we've had to miss a couple of weeks, I've gone back to month based planning. So when I used to work for myself 15 years ago, up to about 10 years ago, I think I used to always have a board pack for myself every month. And it was four weeks worth of activities, one month worth of activity, and I did it every month. Have I told you about this? Am I going to bore [00:23:59] Speaker A: you? Yeah, we talked about it in recent episode. Yeah, [00:24:01] Speaker B: so I did that very recently for October. And so my activity today was actually calls. This is another thing that you get smashed in the face with when you get back, isn't it? Calls all queue up because you have to postpone everything until after. So today's just been loads of calls actually. But one of my activities between them has been to just get back into that, look at that plan, look at that October plan, look, do the kind of reorganization, like what's been done, what's not been done, what's off track, what's on track, what opportunities have come up, what threats have come up since then and really just do that review. So a couple of hours on a Monday morning, reviewing that plan and seeing exactly where I am and then essentially trashing my entire to do list and todoist because I've got like about 15 things in there, trashing them all back to the inbox so that I then have to reprocess them all. Because that's, you know, that's one of the parts of the process in the last couple of months that I've got much better at and much more appreciate is having the traditional gtd, you know, the get things done approach of like you've got your inbox, which you religiously studiously deal with the same time every single week. And therefore it doesn't matter what ideas come up, what, you know, any random thing you can get straight out of your head and into that inbox because you know for a fact you're going to go through that inbox next Monday, next, whenever it is, you do it and you will handle it with a peaceful mind at that point. [00:25:36] Speaker A: Yeah, [00:25:36] Speaker B: yeah, [00:25:37] Speaker A: yeah. [00:25:38] Speaker B: And so I just kind of trashed my entire to do list back to inbox, knowing that I'm going to process it all again. I'm going to take a couple hours and do that next week and that's fine. So that's kind of been what I've done today. [00:25:48] Speaker A: Well, that's good. So those items that were in the to do list don't just go away, they just get kind of relegated back to the pool and they have to earn their spot back on this week list or whatever. Yeah, [00:26:00] Speaker B: exactly. Yeah. Because my alternative is always the standard. You just push it to next Day and next day, next day. And it's just this ongoing pile of things. [00:26:10] Speaker A: Yeah, [00:26:12] Speaker B: I was curious. Yeah, I just wondered if there was anything you did in particular for a longer time off. But, yeah. [00:26:19] Speaker A: Yeah, I am just pretty reactionary, which I think is not good, obviously, because one thing that I think happens is work fills all available time. And you can spend all day on calls and doing email and all this kind of stuff, or if you're disciplined, you can say, I'm going to do these three things which are the most important things for my business this week, and you do those, then you have a better business in a few weeks or a few months. And I think that if I'm honest, that's part of maybe not making as much progress in my career as I'd want, is I'm too reactionary and not intentional enough about, like, how I work. [00:27:13] Speaker B: I think it's really rare that people are very good at that, though. [00:27:17] Speaker A: I [00:27:17] Speaker B: think the vast majority of people, including people that do really well in life, you know, like, I think we've both done okay considering. Sorry, not considering, like, relatively to the world, and yet we still suffer from that. Like, I'm exactly the same. Like, it's. It's really hard not to be. And I really have to fight every single week to maintain those processes. And I don't do it every week as well. Like, I fail in that I don't. You know, sometimes I miss the inbox triage, sometimes I miss the monthly planning, those kind of things. [00:27:49] Speaker A: So, you know, I think. I think one thing that's interesting just kind of, like, reflecting on this. I kind of have, like, two work lives right now. One is Kastos, where I'm almost entirely reactionary because we have a team. And I feel pretty good about that because I can offload some of my proactive initiative onto the team. Hey, Jimmy, we need to do this. Can you kind of head this up and then let me know when you need my help? Or, hey, we need to develop this feature, you know, like, that. That's one side. The other side is, like, my kind of coaching and YouTube and all that kind of stuff. That's all me. So I have to be more proactive and, like, independent there. I think, like, think about, like, how you hear about Elon, like, leading his organizations. He's entirely reactionary. His whole thing is, like, what's the constraint right now? And he hyper focuses on that one constraint. [00:28:39] Speaker B: Yeah. [00:28:39] Speaker A: And then when that's resolved, he moves on to the next constraint. And, like, he sets the vision and has, you know, has mechanisms in place like get the ball rolling. But, but I think he's just a firefighter at his core and I don't think that's wrong. If you have capable people in a good kind of sense of what the vision is, your job is to just enable and unblock them. I don't know, that doesn't seem super wrong to me. That's a pretty good use of your time. [00:29:08] Speaker B: But then a lot of the supposed purpose of a CEO is to set that long term vision, isn't it? To set that strategy. And I certainly had this experience and other people have told me the same as teams get bigger and bigger as well, that when you have 20 plus people, 50 plus, especially when you're into triple figures, quadruple figures, nearly all of your job is just reminding people what that strategy is, keeping them on track. I wonder who does that with his team in that case, who's the person that just keeps them all on focus, on track? [00:29:47] Speaker A: Yeah, Well, I know SpaceX, there's a CEO and she's amazing. He's chairman of the board or whatever, he's her boss. But yeah, [00:29:56] Speaker B: she's okay. Yep. [00:29:58] Speaker A: Yeah, yeah. So, I mean, yeah, you think about all the stuff he does. He, he is like the ultimate visionary, I think, and is able to go super deep on rocket technology and stuff like that. Right. Because he's an engineer. But, but I don't think that he's running the day to day of anything. [00:30:17] Speaker B: Yeah, fair enough. Yeah. I found that a struggle when I was running Ality in the podcast host. We would do our strategy meetup once a year because that's the cadence you're meant to do strategy meetups, isn't it? You set your strategy for 2025, you set your strategy for 2026. You lay it out and then you throw it in a drawer and you just get on with the work. I found a struggle to kind of keep bringing that back. And it was one of the things that my leadership team brought up a few times in the last couple of years before we exited it. The fact that I think that is a place where I was weak and I got a [00:31:01] Speaker A: lot [00:31:02] Speaker B: better at in the last year because of that. There was something around. I did monthly reports. Every month I would put something out. I did it for investors actually. Do you do that just now, Craig? Do you have to do monthly investment reports? [00:31:15] Speaker A: Every month [00:31:16] Speaker B: I would put that together, I would send that off to our investors, but I would repurpose it. I wouldn't change it much. At all, actually. It would never need much changing to go out to the team and say, here's the activities we did in the last month, here's what a few your colleagues have been doing, here's what everyone's been doing. But in the last six to 12 months and through basically 2025, I have so much more started to write in how that tied to the strategy, to the overall goal. [00:31:42] Speaker A: So [00:31:43] Speaker B: it felt cheesy and kind of boring to me because I kept repeating the same things, but in one paragraph I would say, so Allegra from Customer service did this great thing the other day where she developed a little bit of the onboarding which helped people to access this feature more readily and it helped people stick around. And you remember that that was a big part of our strategy for this year is retention. Our strategy this year is to really increase retention so that we drop churn to blah, blah, blah. And so it was even a bit more high level than that generally. But it was always tying it back to what the goal was for the year. And it was something I was rubbish at for quite a few years. I think I would just get in there, I would just do the work. I would probably act a bit like you're talking about Elon there, which is just reactionary, fixing problems, because I thought that was what your job was as the leader of the company. But I found the feedback from the team was actually really positive. Even though I thought it was cheesy and boring writing these little comments in. They responded really well to it. [00:32:44] Speaker A: Yeah, it's definitely both. Right. Set the vision and then unblock the team to achieve it. Those are two really different things in your mind. Set the vision really esoteric and ephemeral. Right. And then like unblock the team is like practical, like, oh, we need to adopt this technology or can I give this customer a refund or whatever. [00:33:10] Speaker B: Yeah. Do you have a strategy? Well, I've got a couple of things there. Do you have a strategy written down? Do you do that every year? [00:33:18] Speaker A: No. [00:33:19] Speaker B: No. Do you feel like you should? Is that a deliberate choice not to do it? [00:33:32] Speaker A: Don't. I don't feel like it would do much good if we had a strategy. I don't know that. It's not my style to be like, guys, remember the strategy is xyz and this thing we're doing relates to that. The closest thing we've had to that is we have what we call seasons. You know, some seasons are about security and performance, some are about marketing, some are about product development. And in that time. And it's not a year, it's a season and it lasts however long the season needs to be. I think that's really appropriate. That is when I really get behind. Like guys, remember this is what we're going for and this is kind of how we want to orient most of our work. [00:34:19] Speaker B: I think that's good. I think that's great. I think that's exactly how I started to think about it. Like our last couple of years, 24, 25 actually we had six month strategy seasons as you call it. Actually. I like that because there was a couple of things we decided in our sort of big strategy meetup that we wanted to get done and work towards. And it wasn't a year long project, it simply wasn't that. So we said, right, we planned it in January because that's the arbitrary time you plan strategy again. But we said, I think we need to be, we need to be done with this by June. So we need to, we'll give ourselves six months for this and that is written down and we'll re meet in June to see what's next after that. So yeah, I think you're spot on there. I think that's like I was kind of sarcastically saying the arbitrary time there because I think it is nonsense. Like things don't always last a year and actually it's way too long to actually take to revisit what you've looked at like such a long period of time. Things all have changed so much. [00:35:25] Speaker A: Yeah. And I think that I've seen from my business and from others, folks who try to box things into confined time frames create more friction than benefit. This is a two week sprint or this is a six month plan or whatever shit takes, however long it takes. And you're creating stress on your team that may or may not be appropriate or necessary. And just all these anyways. Yeah. But also it's just my style. Like I'm not, I'm an intense person, I like to win, but I don't work with deadlines well at all. So that's kind of how I want to run the business, I guess. [00:36:05] Speaker B: Yeah. I put this out on social just recently I remembered it and it always gets a nice reaction. One of the best things I ever read in a business book was in a book by the founder of Sierra. Remember Sierra Games? [00:36:24] Speaker A: No. [00:36:24] Speaker B: Did you ever play a Sierra game? There were some great ones. The one that's always remembered was a really dodgy game called Leisure Suit Larry which was a sleazy ass guy in a Weird suit. So don't play that. It was rubbish. Really, really dodgy. But they made some really good games as well. And the thing he mentioned in it was it was around setting deadlines, around setting timelines. And his rule of thumb was that a one hour task will probably take two hours. Half day task will almost certainly take a day. This is obviously you estimating this is going to take me half a day. It will really take a day. But a one day task will generally take about a week and a one week task will generally take forever. So it was just that rule of thumb that it completely exponentially scales. The longer you go, the worse the estimate is by a lot. Like by a huge factor. And what I took from that was just anytime you set projects, you just always have to break it down to the smallest possible increments. Which is kind of pretty obvious project management approach to be honest. But it's so common though that we'd still get to planning something out over a couple of months and think, right, we're going to give this one two weeks and then suddenly realize, no, that's just not going to work. [00:37:50] Speaker A: Especially in software. Especially in software. And then it's so hard to measure those things anyways. Yeah, we are very bad at all of that. [00:37:59] Speaker B: Estimating is hard. [00:38:00] Speaker A: Yeah, estimating KPIs around team performance and stuff, like just so hard. I wanted to talk about a new product we launched this week. I wanted to get your take on it because I'm a bit at kind of a crossroads. So we launched a AI video editing service. [00:38:20] Speaker B: Okay. [00:38:21] Speaker A: I think one of the, one of the themes behind this that like everyone can take away is like most SaaS should transform into a service. So we were to flip where the S's are. So service disguised as software is. Is. Is I is going. And so like whereas Kastos is a software platform, it hosts and distribute podcasts. This is and we call it UltraClip. UltraClip AI is a video editor. It does everything that a video and audio editor does automatically. And it's really good. It's edited our podcast for the last month about and so if anyone has listened or watched like it's. It's really good. We added motion graphics yesterday. So like this episode we'll get motion graphics. So using Hyper Frames for motion graphics, it does active speaker and side by side and switching and you get audio, you get video, you get clips, you get show notes, you get tweets, you get all the stuff right. All automatically in about like an hour. The question I have is a technical one actually. And I know you've done quite a bit of work on this. I kind of stumbled on this because like DaVinci Resolve and other editors have MCPS and that's how we're driving this. So we have kind of the layers are like there's a website where you submit your stuff. It runs the service itself runs on my Mac Mini with some kind of watchers and Claude and DaVinci resolve. And Claude kind of drives DaVinci resolve via MCP. And then Grokbot is like the customer service agent. [00:40:12] Speaker B: Yeah, [00:40:13] Speaker A: it works really well. I have a whole separate project that's all cloud based and it's not very good. I don't know if you've run into this at all, but that's obviously a big constraint because I can only edit one episode at a time unless I go buy a bunch of thousand dollar machines. [00:40:33] Speaker B: So what are the technical differences there? Are you running the same system on both bits of infrastructure? Because there must be some. [00:40:47] Speaker A: No, I'm sure I'm not. I'm sure like that the, the two, the two kind of started at the same place and have like diverged a little bit. And I guess this is like a meta learning with AI is like without AI you're more disciplined around like, okay, we're going to do this and then we're going to do this, we're going to do this and whereas with AI it's just like, hey, I want this to edit my video. And it goes and makes a bunch of decisions down this path. And it goes and makes a bunch of other decisions down this other path. Maybe, maybe that's it is we need to take what is in the kind of active ultra clip process and say like, hey, how can we make this run in the cloud? [00:41:25] Speaker B: Yeah. Who's Ultra Clip aimed at? Is it YouTubers? [00:41:31] Speaker A: Good question. It does its best at podcast episodes and solo talking ahead. Stuff that don't need as much as many segments and B roll and custom animations and all that kind of stuff. Our episodes are you talk or I talk or we both are talking and it has a little bit of now motion graphics and that's it. And that's all that a podcast episode needs. [00:42:00] Speaker B: Yeah, yeah, Totally interesting. Yeah. I don't know, I'm not sure. I've not come across that myself. No. The kind of difference. The thing I was thinking about most recently was going your way though, which is I've got a Mac Mini set downstairs and I was looking at setting it up for a few different Activities. One of the things I'm playing with just now is how to run a newsletter, like a daily newsletter, using some more automated processes, but with my kind of personal input into it. I [00:42:33] Speaker A: have a client that does this. It's very successful. [00:42:35] Speaker B: Oh, cool. Nice. I feel, though, that I would like it to be running on a constantly running system, to be able to do the research to. Yeah. So I'm going to set up a Mac Mini, basically, to have that process running on it the whole time. I've tried it with a couple of different cloud services, but it just doesn't seem to work as well as it does on my desktop Claude code. So I think I need to do it that way. [00:43:01] Speaker A: Isn't that a trip? And I even find, and I'd love to hear people's comments to this, I find a pretty significant difference between Claude code in the Claude app versus in the terminal. [00:43:15] Speaker B: Okay, right. How so? [00:43:17] Speaker A: I know you use the Claude app and I kind of use both for different. Just how I use [00:43:22] Speaker B: both a fair bit. Yeah, yeah, yeah. I've not noticed. [00:43:24] Speaker A: Do you notice a difference? [00:43:25] Speaker B: I don't think so, no. What kind of difference do you notice? [00:43:29] Speaker A: I find that Claude in the terminal just goes and just does it's. And it's all Opus 5.5.5, medium effort, auto everything. And Claude in the terminal kind of just goes and does its thing. And like, I get a result. Claude in the app is. It takes a lot longer to do anything and often comes back to me like, hey, what do you think about this? Or I was going to do that, but then I decided to do this other thing, which doesn't make sense. I mean, it's got to be the same. It's the same harness. [00:44:04] Speaker B: Yeah, yeah. It's the same language. Sorry, the same model. Yeah. I haven't used a lot of the terminal lately, so maybe I'll give that a go, actually. Because I have to admit, I have found I've been looking at changing over to codecs a bit recently, actually, just because I've found Opus and Fable just to be quite. I don't know, it's actually very similar to what you're saying there. Kind of indecisive, not much autonomy, not much agency. An agent without agency, because it just actually comes and asks your opinions on just about everything. And I keep telling it, look, I'm not very technical. Don't ask me what Fleming. Do [00:44:49] Speaker A: whatever you think is best. That's what I say. [00:44:52] Speaker B: What's your opinion? Oh, yeah, okay, go and do that. But if the terminal is better at that. Then I'll have to give that a go. I was actually looking at antigravity the other day. Was it you that mentioned to me that you were using that? [00:45:09] Speaker A: No. [00:45:11] Speaker B: Somebody else then who does a lot of AI development and runs companies. Who was it? I'll try and remember who it was and pass it on to you after this. But they mentioned to me that they're using Anti Gravity and it surprised me because it feels like, I don't know, looking at the sort of chat around it, it felt like it was a bit of a kind of beginner's tool almost for AI. [00:45:35] Speaker A: But [00:45:36] Speaker B: I don't think that's actually true anymore. And they seem to be bringing in the top models too. I just saw something earlier today that the latest codecs, the latest opus, all that stuff is coming to Anti Gravity soon as well. So I don't know if that's a new harness that might make something like that simpler too, but. [00:45:54] Speaker A: Yeah, [00:45:55] Speaker B: but on the other hand, I'm always very reluctant to change these things because it just jumps around so much. [00:46:01] Speaker A: It does. This is the best one [00:46:03] Speaker B: this week and this one's the best one next week. [00:46:05] Speaker A: Yeah. My current setup, and it does change by the day, is Claude and Opus for most everything. If I want to plan like a really big thing, I'll use Fable and then it will spawn like sub agents in OPUS to do the work. That's pretty effective. And then for like code, Codex checks all of my code and it's a very. I find it a very good checker. And so you can do that locally or you can do that through like GitHub Actions. So every [00:46:42] Speaker B: write set, Opus writes it and then you have codecs review it, essentially. [00:46:45] Speaker A: Yeah. And the kind of like there are specific review tools, but just like any business, they're incentivized to use the cheapest model they can to get a good enough result. Whereas I have plenty of usage on my hundred dollar a month Codex plan. And I could never use all of it. I don't think so. Yeah. So I think this is a less and maybe an opportunity for a lot of founders with thinking about, is SaaS dead? No, SaaS is not dead, but maybe the bigger. This is just back to our video editing service. Maybe the bigger opportunity is instead of creating the tool for your customer, can you just do the whole thing that they want done? And that's kind of the premise with our Ultra Clip is can we just, you know, what's the TAM of video editing? It's way bigger than Podcast hosting and can we just take 1% of the video editing market and provide that as a service? [00:47:56] Speaker B: Yeah, [00:47:57] Speaker A: that looks like software and there's zero humans involved. [00:48:01] Speaker B: I heard that term came up for the first time for me maybe last week. I think it was service as a software. Yeah, like you said, switching the essays. I think it was an SEO service that I heard about. You [00:48:15] Speaker A: might [00:48:16] Speaker B: know the guy, what's his name again? He launched it and it was kind of similar to a friend of both of ours, actually, Brian Castle's old service. He used to run audience ops to basically help people do their content marketing. So he'd write articles for it. And it was an actual service, it was human written. But he's setting up this new service, is setting up something similar. AI written but service kind of provided in a way. SEO as well though reporting all integrated. And it sounds like it's just quite a roughly put together automation, basically written with AI, it sounds quite simple to put together, to be honest. Presuming you've got a human to kind of help manage it along the way. But you can then offer the service that maybe cost five or ten grand a month, three years ago, five years ago, and now you can offer it for 1,000 because it might only take you a couple of hours a week just to kind of keep on top of it and keep feeding in the right direction. [00:49:22] Speaker A: I [00:49:22] Speaker B: don't know though. I feel like even at the question I had about it was Even at say 1,000amonth, I think it's still likely you need a human to do the onboarding, to do the kind of trust building. Are you going to buy a thousand a month service without properly speaking to somebody and having a bit of a trust building activity of some sort? [00:49:46] Speaker A: No, I mean, the way I look at it, and I'm sure there's a lot of ways to slice this one is just keep the price the same and just have AI deliver the backend. I think that's reasonable. Like, hey, we're an agency. We're going to do a much better job than we could do previously. And we're going to use some AI to help along the way. Just shift where people are involved from the boring dumb stuff to the super interesting stuff. I think that's like our cast as production service is probably going to persist and just in that way we're going to do really great trailers and sizzle reels and stuff like that. Things that by themselves would cost $1,000. And we'll just include it because we have the bandwidth now, because editing is 80% automated. [00:50:35] Speaker B: Yeah. [00:50:36] Speaker A: We also like Ultra clip right now. Is 50 bucks an episode [00:50:41] Speaker B: okay? Yeah. Cheap. [00:50:46] Speaker A: Cheap, Yeah. I mean, cheap. And then people on Twitter are like, I can't believe you're charging that much for this. All you're doing is like, inference. I'm like, yeah, for you. Like, yeah, you could go set this up. If that's what you believe. You could go set this up, but that you're not my customer. I think we could drop the price and it would still be profitable. It costs US about between $5 and $10 in tokens to do an episode. [00:51:08] Speaker B: Yeah, that's pretty good. Add on. [00:51:12] Speaker A: Yeah. So I don't know how you think about pricing a service if AI is driving most of it. I don't know the answer. Keep the price the same is probably reasonable. Drop it by 90% is maybe also reasonable if volume is the play. [00:51:32] Speaker B: Yeah. I'm not sure what I think on that because I think I would love to say, yes, keep the price the same because you're still offering the same results and it should really be. Pricing often works best if it's results based. It's value based. What value is this person getting from it? And they're [00:51:48] Speaker A: still [00:51:48] Speaker B: getting the same value. But on the other hand, I think the uncomfortable truth is that there's just going to be more competition because more people can now do that. More people now have the time to do that. More people have the capability to scale a service doing that, and therefore, to be able to maintain it and make money yourself, you're probably going to have to compete a bit more on price. And I hated competing on price. I never competed on price with Alitude. Like, we were always more expensive than everything else in the market. And I liked that because it meant I could say we could. We can build in more stuff, we can give away more free things, we can add in more free features, not worry about tiers, all of that stuff. I hate competing on price, but I wonder if a service like that, if you wanted to make it kind of automated in that way, would have to. [00:52:37] Speaker A: Yeah, I think there's something to being the winner of the race to the bottom. Like, there's some arbitrage there. If you can undercut the other guy and get some customers of theirs and charge less and still have healthy margin. It's not a forever plan, but. But I think it's a plan and it's going that way anyhow. So, like, you can fight against it and you'll have like you did and, like, we have with our cast disruptions. We have customers that we charge thousands of dollars a month. [00:53:06] Speaker B: Yeah. [00:53:07] Speaker A: Where the end result is actually similar to $50 an episode. But they want to buy the thing that's thousands of dollars a month. [00:53:13] Speaker B: Yeah. Yeah. [00:53:14] Speaker A: Not that we're hoodwinking them or tricking or anything, but that's. That's how they want to work. [00:53:19] Speaker B: No. And yeah. And people will maybe even more so in future, pay more and more for the real human touch, for knowledge that this has all been doing done by humans. [00:53:30] Speaker A: Yeah. [00:53:31] Speaker B: Yeah. And they're happy to pay more for the craftsmanship of it and the knowledge that is has been. Every word of this article has been looked over by an actual human and is definitely ok. [00:53:42] Speaker A: Yeah, yeah, totally. [00:53:44] Speaker B: But that's probably going to shrink, frankly. [00:53:48] Speaker A: Yeah. [00:53:49] Speaker B: But there's a space for it still. Yeah. [00:53:51] Speaker A: Yeah. It's interesting. It's interesting. Cool, buddy. Well, welcome back. I'm glad vacation went well. And yeah, folks would love to hear your thoughts on kind of what we've talked about with pricing and value and how you organize your kind of day and month and get stuff done that actually needs done and not be reactionary. We'd love to hear how you all manage this. So if you're on YouTube, leave a comment below. If you're on Twitter, hit Colin and I up on socials. Our links will be in the show notes [00:54:20] Speaker B: indeed. Yeah. Thanks, Craig. Always good. Fun to catch up. And yeah, back in the groove. Let's know they're until Christmas. We've got two months until the next holiday. [00:54:28] Speaker A: Yeah. Time to get serious. Yep. Awesome. Thanks, Colin. [00:54:34] Speaker B: Cheers. [00:54:35] Speaker A: Yep, Cheers.

Other Episodes

Episode 0

October 29, 2018 • 00:50:05
Episode Cover

RS148: 6 Questions That Will Drive Your Business Forward (Part 1)

This week Dave and Craig catch up after a brief break for traveling and conferences.  In this episode, they catch up about a few...

Listen

Episode

April 24, 2019 • 00:40:31
Episode Cover

RS170: Doubling Down on What’s Working

This week Dave and Craig catch up on their businesses after a few weeks without updates. Craig is looking forward to the end of...

Listen

Episode 0

April 13, 2017 • NaN
Episode Cover

RC16

This is the RC16 version upload new

Listen