Ditching Hourly

DH 395 David Guttman

===

Introduction to David Guttman

---

Jonathan Stark: Hello and welcome to Ditching Hourly. I'm Jonathan Stark. Today I am joined by guest David Guttman. David, welcome to the show.

David Guttman: Thanks for having me.

Jonathan Stark: Well, ladies and gentlemen, today we're going to talk about AI, your favorite topic, specifically OpenClaw, but I think some bigger picture things too. But first, for those who haven't met you before, could you tell folks a little bit about who you are and what you do?

David Guttman: Sure. So I'm David Guttman. I run Superstruct, a consultancy that right now is focused on setting up AI employees for solopreneurs and founders. But for a long time, Superstruct concentrated on setting up teams of human software engineers for startups and founders.

From Superstruct Teams to AI Employees

---

Jonathan Stark: Wild. That is a great jumping off point. You're replacing people with bots, right? Actually, let's start there. So I know you wrote a book called The Superstruct Manifesto a few years ago. I don't remember the exact. When would that have been?

David Guttman: Oh, I want to say probably 2022.

Jonathan Stark: Okay. And what was the general premise of that book?

David Guttman: So Superstruct Manifesto really came from the 10 principles of Superstruct. So when I work with a startup, what I'm doing is I'm providing a team of engineers and an engineering manager. But part of the difference is just how the teams of software engineers are run. So it tends to be very systematized. One of the things that is most striking, I think for a lot of, I don't know, people used to working with devs in that type of context, is that there's no voice calls, there's no video calls, there's no daily stand-ups. All communication is over text in Slack. Private conversations are actively discouraged unless the content of that conversation is actually private. So everything defaults public. So a lot of those different ways of doing things got codified into principles and each one of those winds up being a chapter in the book.

Jonathan Stark: Really, it's a wild contrarian way to manage engineering teams. Very interesting, very different. And it strikes me as something that would lend itself very well to a universe in which people are chatting with AI agents in Slack and public channels and getting work done, you know, coming out the other end and getting pushed into GitHub and pull requests and that sort of stuff. How similar, am I making that up, or how similar is that?

David Guttman: I mean, I saw something recently about how the biggest issue for new graduates right now and them finding difficulty getting work is not actually the rise of AI, but it's the rise of remote work. And what I find so interesting about that is that those two wind up being really connected. The types of ways that you interact with AI are so similar to the way that you interact with AI employees, right? The default way, when everybody first encountered ChatGPT, you didn't get on a video call with it. You know, that wasn't a thing, right? The chat interface is very similar to, you know, much the way that I was talking about how Superstruct runs things. It was all over Slack. And there's a, if people aren't familiar, one of the different agents or software development agents out there right now is called Devin by Cognition. And then that was one of the first ones that the agent was on Slack. So you could talk to it on Slack. It had its own computer, so it could use a browser, could clone repositories locally to its machine. It could run commands to get whatever your project is up and running. It could document things like that. And so Devin, you know, which is out for a number of years now, was like the first time where it was like, you kind of couldn't, like, if you squinted your eyes, you couldn't really tell that you were talking to an AI. I mean, you could by the way that it was talking and things like that, but the whole shape of it was not different than the way that you would work with a Superstruct engineer. Because again, Superstruct engineers, you're restricted to talking over Slack, you know, working with them over GitHub. All of these different interfaces were things that the AI was totally capable of working with. No, I think probably more to your point, which is like things that work well with working with AI, you know, do those work well with remote employees? And that's totally true, right? I think part of the success of Superstruct was how we had process for everything, how our onboarding was very systematized, so that when we brought a new engineer onto the team, they knew exactly what the coding style was. They knew exactly what the different steps to get a feature reviewed and then QA'd and then eventually put into production. And all of those like steps and systematized behaviors really work well for AI as well.

Testing Devin Against a Human Engineer

---

Jonathan Stark: Did you use Devin at all? Have you experimented with that?

David Guttman: Yeah, I mean, I think Devin, so I personally do not use Devin on my own, like side projects and things like that. It is very heavily used in some of my clients. The first turning point, I think, for me realizing that Devin was like massive was when I gave the same exact task to a human engineer as I did to Devin and didn't let them see each other's work or anything that the other one was doing. So they both created a real feature in isolation and then compared the, you know, the work that came back. And Devin was like six times faster or something like that. And in some ways it was better and that it was more complete and full featured and things like that. In some ways that it was worse in that it had a larger blast radius, it did things that weren't asked for. But on the whole, the fact that you could get something, you know, six times faster for about, you know, what wound up being, I forget, like the same hourly cost or something. I forget exactly how it worked out. So it's not like it was a screaming deal in terms of how much we were paying for the feature, but how fast it came back was incredible. And then I guess I should also say that the, it probably does come out to a screaming deal when you compare it to a US based software engineer versus, you know, distributed geo arbitrage engineer, which is what I used at the time.

Jonathan Stark: Okay. So when you say the cost, it was a pretty low cost for the human compared to a US based engineer.

David Guttman: Yeah, extremely. Yeah.

Jonathan Stark: And Devin, so Devin was kind of like comparable with that level of pricing.

David Guttman: Yeah.

Jonathan Stark: Okay. So that Devin, I remember hearing about Devin years ago, like you said, it's been a couple of years at least. Would that be considered, I guess these be considered an AI agent specifically for coding, right?

David Guttman: Yeah. You know, it's been a while since I was keeping up with exactly how they were sort of running it. At the time, you know, when I did this test, it wasn't like it was using Claude Codede or something like that. It was, it was using Claude. I think at the time it was probably, you know, like Claude Sonnet 3.5 or something like that. I have to imagine that they probably have their own harness that's comparable to Claude Codede and Codex now, but they are using the frontier models. And then they still have the VM where the agent has access to its computer. So I know we're going to get into OpenClaw, but I'd probably say that Devin was the first more like a fully featured agent that is similar to OpenClaw where you've got the models themselves. So that would be like, imagine you go into your browser, you go to Claude.ai and you start talking to it. That's kind of the closest thing to talking to the model itself. That would have been more true back when chat.cppd came out because all of these companies have added more and more to their interface and more and more in the way between you and the model. But that would be the most stripped down version. Then you can go to a harness like Codex or Claude Codede. And then those are typically, you know, terminal based things. And so that's a harness. So you're talking to the model and then wrapped around the model are all of these things. that the model can use to operate your computer, manipulate files, folders, things like that. And, you know, there's a, you know, other ones that are more user-friendly, like something like Claude Codede. But then you get to the next phase where you wrap all of that in something that has, you know, a communication layer, like hooking it up to Slack or Discord or Telegram or whatever it is. And then you get into the territory of something like Devin and, you know, what I use much more heavily these days, which would be OpenClaw. So it winds up being that harness plus a whole bunch of other things tacked onto it.

What OpenClaw Adds to AI Agents

---

Jonathan Stark: Cool. So yeah, let's fast forward to OpenClaw. When did that, was that the beginning of 2026, this year that that was released?

David Guttman: So I kind of forget exactly the whole history of it. It grew out of a project by a different name by Peter Steinberger that I think was much more WhatsApp focused. Changed names a bunch of times. And I think by January or February of 2026, it became OpenClaw and, you know, its adoption has just been, you know, straight hockey stick vertical.

Jonathan Stark: Yeah. Yeah. And I'm a big fan. I'm set up on OpenClaw and the, for people who haven't messed around with it, it brings that sort of chat based interface to something that has dramatically more power. So it has this like this crazy high ease of use. And then it has all this power in the backend to actually do things for you, you know, depending on what you granted it. But in my case, like I granted it access to a bunch of things, my podcast and my Google drive and a bunch of different things. And it can move files around and it can put chapters in an MP3 and it can upload stuff and publish things, you know, to an RSS feed and just really, but it's the same easy use that anybody who's tried ChatGPT is familiar with, but then it can go and actually do work for you, especially the tedious sort of death by a thousand paper cuts kind of work that knowledge workers deal with on a day to day basis. So how important to you was that particular piece of the puzzle when OpenClaw came out and it was a, basically a telegram or WhatsApp style interface or Slack style interface or discord. I mean, it's probably got dozens of them now, but how much of a big deal was that to you versus, Oh, well I could just do all of this in Claude Codede if I didn't mind getting up and sitting

Vibe Coding from Anywhere

---

David Guttman: at my computer. Yeah. So I think for me, you know, I definitely had this white whale that I was hunting, which was to be able to vibe code from anywhere from, you know, my phone, you know, while walking the dog or out with the family or any number of examples of how I have zero work life balance. But since about 2023, that was just something that I was constantly trying to do. And the reason was because in 2023, I used cursor, which was one of the first popular software engineering agents that was built into a code editor. That was the first time where I pretended I didn't know how to code and pretended like I was just a product manager and told it what I wanted. And it actually made a usable app. So I've actually barely made any edits to it. And that is the mobile app I use today for archery scoring. I did it in a couple of hours just by telling cursor what I wanted, pretending like I didn't know how it should be done or anything like that. It was successful and I still use it today. But what that showed me was, well, damn, if I don't have to look at the code, I don't have to review it. I don't have to see what it's doing or telling it how to do what it's doing. Why do I need to be at a computer? I could just be talking to this thing. Now, this brought up a whole host of problems because cursor was built into a code editor. It had to run on your laptop or whatever it is. And so then that started this huge, long, painful journey of trying to use remote desktop on my phone and zoom and pan to get to the chat box and tell it what to do and whatever. And that worked if you put heavy quotes around the word worked. And it got a little bit better with Claude Codede, different ways of remote accessing it, but still you had to have a laptop or a PC or server or whatever running all the time. Um, but the big, big change came with OpenClaw because OpenClaw had all the power of Claude Codede or, you know, cursor or whatever it is. Um, but it was made to be hooked up to, in my case, Discord is what I use, but, you know, and also it was made to be hooked up to WhatsApp and Telegram and all of these other things. And that was a huge, huge upgrade because unlike something like Claude Codede that really wants a directory or is really set up in a way to be a, in many ways, a tool, like you are working in this directory or this repo or this code base, uh, and here is this objective that you want. So you start up Claude Codede for that objective, you tell it what you want and then it does the thing that you needed and then you shut it down. Um, OpenClaw is very different. It doesn't live in a folder or like a project. It behaves much more like, you know, dare I say it, an employee that is always on, always ready, waiting for you to ask it, you know, for something or to do something. And then it has like, if you set it up like I do, full run of your computer to accomplish whatever it is. It can use your browser. It can, you know, look at all your files. It can access whatever it needs to do. And, uh, the way that it's also set up is it has much more of, um, you could call it a personality or, you know, if you just want to be technical, it just has a very long system prompt that guides it to be way more proactive and way less willing to just stop, uh, before, you know, finishing what you want. And all of those things together made it a complete game changer for, uh, you know, eliminating work-life balance.

Pretending You Do Not Know How to Code

---

Jonathan Stark: Achievement unlocked. Yeah. Okay. I want to loop back to something you said earlier, and I think it's going to come up a few times, but you said you pretended like you didn't know how to code. Like, what do you mean by that? Why would you do that? Um, let's see. Okay. So

David Guttman: what do I mean by that? So I wanted, I wanted this mobile app, um, for, um, you know, archery scoring. Um, and so, you know, I had some ideas of, of what was important to me. Like it was really, so, you know, when, when shooting at the target and need to collect your arrows, other people shoot at the same time and they're collecting their arrows and you don't want to be like out at the bale, um, messing around with your phone for a long time where they're kind of like waiting for you to get off the range so they can shoot again. So it was really important that I could enter in the, you know, the scores very quickly, get my arrows and get back to the, get back to the line. So those are the things that I communicated to it, but I wasn't telling it, Hey, I want you to use react native and I want you to use this, you know, particular, you know, UI framework or whatever it is. I wasn't, I wasn't looking at its code and telling it, you don't have enough tests or, you know, this is sloppy. Your functions are too long or whatever it is. Um, I was just, you know, using it the way that I wanted to use it and making sure that it fit my, um, you know, my requirements. Um, and I would iterate that way and it works really well now. Like, why would I do that? I mean, in some ways it's like, why wouldn't you like, you know, for context, so not to interrupt, but for context, the dear

Jonathan Stark: listener, um, David has written books about, so, you know, and in my experience, it takes them, it takes a hot minute to break the habit of micromanaging it when you do know how to write code. So what, and like, what caused that light bulb to turn on for you? where you basically went, you almost went past product manager mode and straight to user and like, I just want an app that does this for me.

David Guttman: Yeah.

Jonathan Stark: Like when you could have easily said, use this React thing or this routing framework, blah, blah, blah, blah, blah.

David Guttman: Right, right, right. So, I mean, I think it was kind of natural for me, you know, Superstruct is a dev shop and I think it's pretty natural for me to delegate to engineers and an engineering manager. And so I think in some sense, it was just natural to treat it like that situation in many ways, like can this, can this like write apps and build software in a way that's analogous to how I run Superstruct or how a client would interact with Superstruct. So, in that sense, I don't think there was necessarily a lightbulb, it was just, is AI up to the task of having this much ownership and I guess running with it, you know, to that level without being micromanaged.

Jonathan Stark: Right. Yeah, it's really, I personally, and I don't write that much code even with my OpenClaw or anything, but it was a huge shift to stop thinking at that sort of tactical level of like, I'm going to use whatever Laravel and this is going to be my coding, like even caring.

David Guttman: Right.

Jonathan Stark: And finally, the first time I went to like really use Claude Codede and having kind of come to the same conclusion that you had, like I found that, in fact, it was one session where I was micromanaging it essentially and I was telling it like, here's what I want you to do and then it would screw it up and I'd get angry and I'd tell it to do it again. And then finally, I just sort of threw my hands up and I said, I just went way up like two levels of abstraction and said something like, I wanted to do this, you figure it out and the results were way better and less frustrating and almost instantaneous to the point where after having that experience, when I do have it code something, I tell it to pick the language because it knows what it knows best in terms of language. I'm not going to look at the code anyway and, you know, JavaScript and PHP are the two languages I know, so I'm not going to have to do it in PHP and, you know, so it's like, you know, and it wants to go do TypeScript or whatever. So that's close enough for me and it's like Peter Steinberger, is that how you say it, Steinberger, I saw him say on an interview that he ships code that he hasn't read and I'm like, yeah, well, so do product managers, so does everyone. So anyway, so that to me, that's a huge unlock, so I wanted to kind of linger on that point for a second because I know a lot of software developers are listening and they might have had the experience that, you know, like I did at first where I was telling it what to do and micromanaging it and it was just making me mad and then when I kind of like let it just do its thing, it worked way better.

Software Quality, Risk, and Real Requirements

---

David Guttman: Yeah, I mean, I think it touches on identity and role and if you are thinking and that's probably why it was easier for me to like think of myself as a non-technical product manager, which meant that it's not my job to look at the code. It is my job to evaluate whether or not it is meeting the requirements or the criteria and it's useful, but it's not my job to, you know, scrutinize the code. Now, you know, I think that that does change a bit when the app changes, right? Like if this was an app that had a database, it was taking people's social security numbers and there was ways that it could kill people and all kinds of stuff like that, I think I would have been not so eager to be, you know, that cavalier, but also not something I would have tested. I think, and I think this again, you know, to go back to the how does this tie in with, you know, human engineers or human teams or remote teams or anything like that, I mean, it, you know, it's what's the, like what's the worst that could happen? Like how good does this need to be? How solid does it need to be? Like if this isn't the most pristine maintainable code in the world, like what is the real world impact of that? You know, sometimes, sometimes you do need to have a bridge that is never going to fail under load because that would be terrible if it did, but other times, you know, you're building furniture that doesn't really need to hold up that much weight because it's more of a set piece. Like it just has to look nice in a room. And I think understanding what you're building, why you're building it, and what the real requirements are, uh, is important and not just, oh, this is software. Therefore, it has to be the best software ever written because that's how you write software.

Jonathan Stark: Yeah, it reminds me of, of, geez, I don't know, it was probably like 10 years ago now, but when it seemed like junior developers would, you know, read some article about sharding or something, and now everything had to be sharded or like everything had to handle.

David Guttman: Web scale.

Jonathan Stark: Oh my God, that video.

David Guttman: Yeah, everything had to be web scale all of a sudden.

Jonathan Stark: And it's like, you only have 10 internal users of this workflow software. Like it doesn't need to be web scale. So, you know, back away from the Kubernetes cluster.

David Guttman: So it's, it's just another, a newer, more modern example of like build it in the appropriate way. And like, it's really easy to overbuild stuff, but other times, you know, in certain high risk situations, you don't want to under build it or under engineer it.

How OpenClaw Has Changed

---

Jonathan Stark: Right. Okay. So, okay. So we started talking about OpenClaw, uh, sort of officially named earlier this year and, you know, it got your attention. I know you, uh, I saw in your LinkedIn that you're like contributing to the source from like the very first week. How has it changed in the six or so months that it's been? How have your attitudes changed around it? As you know, is it, is it getting better? Are you like less excited by it for some reason? Are there other competitors that have cropped up? So like what's kind of happened in the last six months around OpenClaw? Okay.

David Guttman: So how did I first come to hear about it? I mean, there was so much buzz about it so quickly. Um, I mean, I just don't even really know how that happened. But, um, yeah, I mean, I remember seeing kind of like a tweet or something that was like somebody talking about how their biggest challenge with OpenClaw was like explaining to somebody else why it was so amazing, but everybody who trusted them and did it agreed that it was, you know, life-changing. Yeah. And, you know, something like that will get you curious. And I honestly cannot remember what got me to pull the trigger to test it out. But, um, you know, again, once I had it, um, or I guess let me put it a different way. You know, something that, that obviously made it attractive to me was that I had been hunting for a while, you know, since 2023. Um, so, you know, two and a half years at that point, I've been hunting for the best way to be vibe coding or whatever it is from my phone. So something that was a, uh, an agent that could do whatever Claude Code did theoretically, um, but was like had a natural connection to a chat app, like discord was very attractive to me. Um, I guess I really should drop a footnote because Devin has been here the entire time and Devin probably could do, you know, most things that, um, something like, uh, you know, OpenClaw could do, you know, especially early on, um, that was just like really expensive. And so that is a, that's a very heavy tax for using something personally. Um, I think as a business, you always, you know, you always have an ROI, uh, for, for using it, you know, especially if you already have revenue coming in and there's any idea of like, all right, this is going to make us more efficient or faster or more productive or whatever. You already have that revenue that you can add a multiplier to, and then, you know, put the cost against it. When you're doing things in your personal life and you're really just messing around, um, I think it's much harder to justify this, like, always on agent that you, like, as an afterthought, have it do, uh, do a particular thing for you. So, I did just want to talk about Devin really quick there where Devin could have done some of these things that I was hunting for, but for me was very cost prohibitive or way too cost annoying for me to rely heavily on it.

Jonathan Stark: Yeah, hard to justify.

David Guttman: Yeah. So, OpenClaw, one of the very first things that it did is that it really solved this piece that I was looking for, which was how to, you know, just have this always on agent where I could just give it features that had access to all my projects and I could just move, you know, move the ball forward on any number of projects at any time, you know, like before I go to bed, I could just write off a message or whatever and it could start working on something. Now, you know, I think that's something that showed its value very quickly. Now, in terms of how the project has changed over time, so I think, you know, I came in somewhere in January, I came to it before it had threads and Discord, that was actually my first contribution.

Jonathan Stark: Cool.

David Guttman: So, certain like things like that, you know, I think would be pretty noticeable to me is now, you know, threads and sessions. I know a lot of voice has gotten built in there. So, both text to speech and speech to text. Yeah, there's a theme. Um, so, so, but like, I'm really trying to think about how it's changed. I think a lot of it has to do with, you know, the theme of the project itself and how it's changed over time. And so, there's a lot of things that I couldn't do. So, I think there's a lot of stuff that I could have done differently. Um, so, but like, I'm really trying to think about how it's changed. I think a lot of it has to do with stability and, um, OpenClaw depending on the model that you were using. I think in the beginning, a lot of people really liked using Claude and Claude Opus for it. Anthropic was always from the beginning pretty against you using OpenClaw with it. Uh, to begin with, they threatened that it was against terms of service and reminded people that they could get banned for using it. That was not enough of a deterrent, but eventually they, they figured out how to identify when the model was being used by OpenClaw and then they, uh, figured out how to like charge extra for it. And that, um, I think that killed off, finally killed off a bunch of Claude usage. So, when everybody switched to Codex and GPT, there were a lot of issues where it would say that it was going to do something and then it, it wouldn't do it.

Jonathan Stark: When you say switched, you mean they're still using OpenClaw, but they switched the model that was powering it?

David Guttman: Exactly. Yeah. Yeah. So, OpenClaw from the very beginning really made a huge effort to one, support a bajillion connector chat apps and a bajillion providers and models. So, it's one of those things where you, you can both set up a primary model, like a default model, but then have a bunch of fallbacks or switch mid conversation to a different model, or, you know, you can talk to it both on WhatsApp and Discord. So, it's really set up to be this very flexible way with how you talk to it and what's powering it. So, a lot of people, once Claude became, you know, too expensive by not, you know, just being on the all you can eat subscription plan, uh, people naturally switched to OpenAI and, and GPT, uh, because that was also about the time that OpenAI hired, um, Peter.

Jonathan Stark: Right.

David Guttman: And so, they had a really good relationship and were very public about the fact that you could continue to use OpenClaw on their subscription. And that was totally fine. But one of the things that happened about that time, I think it was GPT 5.4, um, possibly 5.3, but, uh, it had a, it had a nasty habit with OpenClaw in particular of saying like, you'd ask it to do something, right? Like update this file or whatever and be like, okay, I'll do that. And he'd be like, Hey, did you do it? And be like, Oh no, you got me, but I'll do it now.

Jonathan Stark: Did you do it?

David Guttman: No.

Jonathan Stark: You got me again. Got me again. Yeah, that was frustrating.

David Guttman: It was really, really frustrating. Um, and that just had to do with how OpenClaw and the model sort of dealt with each other. Um, I could definitely get into the technical bits of it, but that's something that OpenClaw has spent a lot of time and attention on, uh, addressing. And so, it's been a long time since I've had that particular issue with it.

Qualifying Tasks for an AI Employee

---

Jonathan Stark: Yeah, me too. Yeah. So let's, let's talk about your secure OpenClaw setup service. And there's a couple of interesting things about this for the listener. One is, is I always like to bring examples of product type services, but the other is that you've had experience with, I don't know, a half dozen or a dozen non-technical people setting up this, you know, fairly advanced thing for them or this fairly powerful thing or power tool for them. And I want to get into some of the, the, your approach to doing that. When someone comes to you, how do you decide whether or not you think it would be a good fit for them to actually hire you to do this?

David Guttman: Right. So the, the first stop is the discovery call. That's totally free. That's a, you know, a half an hour call where what I'm doing is trying to suss out, is there a job that OpenClaw can take over for you? And, um, the, the best model for things that OpenClaw can do is a virtual assistant. So if there's anything that you're doing or someone on your team is doing or whatever, that could be handled by a actual human in the Philippines with a browser and good English and good internet access and everything like that. If there's something that that theoretical human could be doing for you, OpenClaw almost definitely can do it as well. Um, but, you know, sometimes when you're pushing somebody on it where it's like, okay, so what's, tell me something that you do and be like, okay, well, every week I send out this update to all of our clients about what's new with our product and, and whatever it is. And, um, you know, that, that's, that's, that's important and it's time consuming and, and whatever it is. And that on, you know, face value might sound like something like, oh, well, the agent, all they need to do is go see what's changed in the product, get this list of emails and write up an email with those things in it and send it out. Like, cool. So, you know, maybe, maybe this would work. But, you know, that's the type of thing that you wind up pushing somebody on and be like, all right, well, like, let's just pretend that, you know, you got a, a, you know, someone in the Philippines or whatever. Uh, why can't, you know, they do it. So like, why, why not hire a virtual assistant? And it always winds up being something related to taste and judgment in that, like, well, you know, like we have a certain way that we talk about the product and it's really important that like my personality comes through because like it's a, you know, like founder led brand. And if I gave it to somebody else, then it would just come off differently and it would just, you know, fall apart or whatever. So it's really important to, to find those, those things. And so sometimes it has to do with like a touch point between, you know, clients or, you know, uh, you know, the high value connections or something like that. Um, other times it has to do with just kind of like inside knowledge, like, you know, like, oh, well, I happen to know this about this particular company and the way that we deal with them. And, you know, there's all these like little edge cases. You know, the way that we enter the stuff in for each individual, you know, partner is different or, you know, all those little things. And those, those can come out in the conversation. But, you know, other times, uh, there's lots of things that have something that even if it's not already written down, you can tell that it is wrote, it is systematized, someone could follow a checklist. Um, you know, if you had a virtual assistant come in, you could imagine running through it with them. And as you're running it through, you just like write down every little step. And if they're following that checklist, they'd, they'd be able to do it. And those, you know, if I can find those, then that's usually what I want to set up as a target. So something that's time consuming, done repeatedly, and does not really involve a lot of your judgment, that's something that I'm gonna be like, okay, this is a good fit. Uh, this is, this is something that can be taken off your plate and start saving time.

Jonathan Stark: Okay, perfect. So that's, that's the sort of qualifying round when you first meet with somebody, it's like, can we, do you have a checklist for something that's annoying, recurring and delegatable, I guess I'd say.

David Guttman: Yeah. Yeah. Delegatable is probably the big one.

Jonathan Stark: Okay. So once, once that happens, what do they, like, what does the process look like? So if you were going to do this for, I don't know, it's like you're trying to think of someone who, you know, Elton John, the first famous person that came to mind, I'm too old. Um, and he's like, oh yeah, uh, I need to have all of my media appearances, you know, pulled together into a report to send to, um, my agent or something, you know, I don't know. And it's like, okay, great. Like, what would Elton John have to do to prepare for, well, actually this is a better question. Like, what ha how do you, how do you do this? Like, what is the, what is the mechanism for setting this up for them?

The Secure OpenClaw Setup Process

---

David Guttman: Okay. So let's pretend, you know, we got this, this agreed on job that that's the goal. That's the target. This is what we're going to try to get your, your OpenClaw set up with a good foundation to be able to do. Uh, the next stop. Yeah. So the next stop is like, okay, we're going to schedule some time together. It, you know, it's typically like a three hour zoom screen share because it's pretty time consuming to go through all these things. But before that is the, the pre setup checklist that I do need the, uh, you know, my client to do. And so that's a lot of boring stuff. Like making sure that there is some computer somewhere that can run, uh, the OpenClaw. Like the OpenClaw does need a machine. It needs internet access. It needs somewhere to live. So being able to get that it's going to need a model. You know, we mentioned that very popular for OpenClaw to run with OpenAI and GPT. So you need an OpenAI account with access to GPT. Uh, you had mentioned, you know, the secure aspect of the server, uh, the service, um, that involves like whatever server it's running on being able to be secured. Uh, and you know, Tailscale is, is really useful for that. So having a Tailscale account, you're going to want to talk to your OpenClaw. So that's setting up something like Slack, uh, and having a workspace that your OpenClaw can be in that you can talk to. And so it's a lot of those things. And that, that winds up being a bunch of things that isn't, doesn't really need there for, but are things that the OpenClaw, like the AI employee will need. And so that's, that's all good stuff to get done before the, um, before the setup call. Then once we're on the call, um, that is where that server is actually, uh, set up in a way, uh, where it has OpenClaw running the Slack app has the, the bot integration and OpenClaw has access to, you know, run the bot and speak as the bot and take messages as the Slack bot. Um, Tailscale is set up. So whatever that server is, it's completely isolated from the public internet and only you can access to it, uh, have access to it, which, um, really improves security. And so all of those things happen on that, uh, screen share setup. And then once all of that's done and OpenClaw online and responding, you can talk to it and you can, you know, tell it to do things and, and it's, it's ready to go. That's then when we would step through all the items on the checklist for whatever that target job was. So if it was the media appearances, it would be like, okay, so if you were to do this or your assistant was to do this, what do they do first? They open up their browser and then they go where, or is this like in a Google drive or is this on a folder somewhere? Like what, what is that step, you know? And then, and then what's the next step and what's the next step all the way on to, okay. And this is delivered via email or it makes a password protected webpage or whatever it is. And then, you know, the, you know, you get the agent to do that. And then, um, you know, once that's, once that's all done, then there's a couple of steps to, to make it more reliable and then, uh, give it, you know, what's needed to, to make, make it do on its own initiative or, or on a, on a schedule.

Local Machines vs Cloud Servers

---

Jonathan Stark: Okay. So tons of stuff in there for, for people who haven't heard of it before, what's Tailscale and how does that relate to security on a, well, wait, actually before that, even before that, I'll come back to Tailscale. What's your, where do you fall on the, uh, spectrum of local machine versus remote machine, like in the cloud? Yeah. Okay.

David Guttman: Yeah. So Mac minis super popular option to the point that, you know, they like sell out. Uh, nobody can get one. Um, you know, I've done, I've done a setup for, for a number of people, uh, where they either have a Mac mini or, um, you know, pro tip, you don't need a Mac. There are these very Mac mini like, uh, you know, you know, computers that you can get on Amazon for four or five, you know, a hundred dollars or whatever it is that can also run, um, an OpenClaw. So the big advantage to having a computer that is running OpenClaw in your office in your office or your home is one, uh, you know, security.

Jonathan Stark: Um, okay.

David Guttman: So one, there's a whole bunch of different parts to security. So I'm just going to talk right now about the security meaning. Is that computer itself ignoring OpenClaw? Is that computer itself accessible? Can an attacker get to it and turn it into as part of a botnet or some spam farm or whatever it is, or yeah, regular network security. So by default, if you have a Mac mini in your office or at home, generally, that is not going to be, uh, a easy target for an attacker to get to compared to, as you mentioned, like the cloud, like if you have a server that you're renting as a, as a VPS virtual private server, server on digital ocean or something like that, those are usually, uh, usually, uh, rented to host a public service that like a web web app or something like that, that you do want the public to be able to access. Well, that's a double edged sword because that also means that attackers can get to it in a way that they cannot get to a machine in your home or office. So that's the first advantage that, um, you know, having OpenClaw run on a, on a Mac mini or something like that. Uh, the second is that if you run it, um, at home when OpenClaw uses its browser and accesses services, like let's say, um, LinkedIn or something like that, um, LinkedIn, uh, and pretty much any of these, you know, consumer, um, you know, apps and services, uh, they over time have become very hostile towards bot traffic. Um, you know, usually because you got scrapers and scammers and all fraud and all kinds of nonsense that, that, that they've had to deal with over time. And one of the heuristics that they use is, is this traffic coming from a residential source, residential IP, or is this coming from a data center? So if the traffic from your OpenClaw is coming from a data center, like if you get, if you are hosting it in the cloud, like on digital ocean, well, it is now going to show up to LinkedIn as coming from a data center. And LinkedIn is going to be pretty predisposed to not want OpenClaw to be able to do anything because it doesn't trust it. So that's another big advantage of hosting it at home. Uh, on the flip side, uh, if you are hosting your machine at home, you are responsible for everything that goes on in that machine. Like all of the hardware, if it fails or anything like that, that's on you. Uh, you have to make sure that it always has power. It always has internet. Um, you know, I think we're all in the first world. We tend to those things, think of those things as always on and very stable. Uh, the truth is they tend not to be, uh, especially when you sort of rely on the most, like there will just be some little blip that, you know, your power goes in and out. You mostly don't notice, but it took your machine down and now OpenClaw isn't back online, or maybe there actually is some issue with spectrum and, you know, it's. down for the day or something like that. And you are really relying on it. Like you are out, you know, at a meeting or something like that. And you are really relying on the fact that you had access to your OpenClaw. So those are not issues that you typically have with a data center because their entire job is to make sure that all the hardware is functioning, replaced when it gets old or it breaks. The power is always on. There's always backups. It does not go down. The internet, you know, backbone, you know, connection it has is always stable, always fast, does not have issues, has backups and so on. So in terms of reliability and always on 24 seven, you know, like it were a real service, I recommend the cloud, but there certainly are advantages to having it in your office.

Tailscale and Remote Access Security

---

Jonathan Stark: Now you did mention Tailscale, you asked about that.

David Guttman: Tailscale is one of the pieces of the puzzle for having it in the cloud, in one of those data centers, because you can turn off its access to the public internet while still maintaining access to yourself. Tailscale creates a private network that only you can access. So that's the value there. So it has the sort of security equivalent to having it in your house, but with all the benefits of the always on power and fast internet and hardware backups and so forth.

Jonathan Stark: Exactly. Okay. And what, it just occurred to me actually, how did, well, that's a technical question for later. It's maybe at some point you can explain to me how Slack can communicate with a local machine on your computer. I guess it's the bot. Okay, that makes sense. So once your client, Elton John now has his VPS set up, it's got Tailscale configured,

David Guttman: it's got OpenClaw installed, it's connected to GPT and all of that, all that stuff that he never had to have before. That, and like you said, it's time consuming to a certain extent, but probably faster than onboarding a remote employee.

Jonathan Stark: Yeah, I would say so. Yeah, there's no healthcare, no health insurance. Yeah, no health insurance and they're around the clock and they have no ego whatsoever. Right. Okay, so once they get set, so the first task that they sort of qualified themselves by proving to you that they had a delegatable task, once that's done, I mean, surely they're not, surely that's not the last thing they're gonna use it for. That would be silly. So how, like, what is the unlock for them or how do you try and get them to, this is a non-technical person who's, you know, maybe not even used to making, they're just doing stuff by the seat of their pants and they just know how to do it or whatever. How would they set up the next one without you there where maybe they'd need to log into TMZ or some other, you know, what was it, SoundScan or something that Elton John cares about? How do you, how do they not get stuck immediately and what do you teach them to kind of eat fish instead of giving them fish? Yeah, exactly.

Teaching Clients to Train Their Agent

---

David Guttman: So, I mean, so luckily they don't have to set up all of the, you know, the server and OpenClaw install and everything like that. That's, once it's done, it's the equivalent that they have a computer, like they have this thing that they can use. And, you know, the goal is not to have a computer that only runs one app, you know, much in the same way that we're talking about. The goal is not to have just an OpenClaw that runs one, you know, recurring task or only does one thing. But in showing them how to get to whatever that task or that goal, that first one, that pattern is exactly how you could set up as many as you want. I talked a little bit about, you know, the steps, you know, that you go through, which would be what you're really trying to do is you're thinking of all the different steps on that checklist, you know, whatever it is, you're having it do each individual step, you know, you are asking it, okay, go log into the site. Do you see where it says, you know, the username? Oh, you do. Okay, tell me the username. Like, okay, do you see where the messages are? Okay, click on that. Tell me like the first message from, you know, who you see. And so you're just like all these little steps, having it prove that it's able to take the next step and then it's seeing something and doing what it needs to do. And you take all of those steps all the way to the whatever that deliverable is, like a report emailed or, you know, sent to you or a webpage or whatever it is. And then once it's proved that it can do all of those individual steps with you handholding it, then you ask it, okay, create a skill, which is, you know, shorthand for it writing up its own checklist, such that, you know, I can ask you just like do the thing, not all the individual steps. I can just tell you like, go do the thing, you know, that gives me the outcome I want and you'll be able to do it quickly and reliably. You know, without the handholding. And it'll write up its own documents for itself so it can do the whole thing. And then you can do another test with it. And instead of doing all the individual steps, you just tell it, you know, all right, do the thing. And then if it's successful, great. You now have something that you can tell it to do on a schedule or some repeatable, you know, cadence. Most of the time it can do it when you're handholding it, but when you try and do it all in one shot from its documents, there's something that it'll get hung up on. And so that's just, you know, normal where something that it was able to gain from the context from you handholding it did not transfer to its notes. And that's fine. You just note whatever it missed. You tell it to fix its notes, test it again, and then it probably will work. And then you can put it on a recurring schedule or cadence like that. So having, it's like once you go through that entire process with a client, they will understand that that's the same process, so they want to do TMZ or SoundScan or whatever it is, they will start doing the same process. Like, okay, go to TMZ, login or whatever it is. Do you see the, you know, something that you know is after the login screen? What does it say? And then it tells you, okay, you're past the login screen. All right, do you see the, this page? What is it? You know, whatever it is. And then you go through all the little handhold steps and then again, the skill and then the recurring.

Jonathan Stark: Okay, so a couple of questions there. So one, if this were a local machine, is it spinning up a virtual browser or a headless browser to do this or would I actually see the browser?

David Guttman: I can do either. It can do either. Yeah, I mean, it's totally a config sort of deal. I have some OpenClaw where I default to headless. I have other ones where, you know, I just, I can see the browser pop up when it is doing whatever it's doing. Having the browser open is useful for debugging. You know, you might get some clues if there's, if it's not able to do what it's doing, but you can see the browser, you can sort of more quickly recognize what's going, what's going wrong. But once, I don't know, once it knows what it's doing, you know, headless without it popping up can be, you know, more efficient.

Jonathan Stark: Yeah. Okay, and I mean, as you're describing that, it reminded me of days doing like sort of customer support with non-technical clients, employees, or non-technical employees at like a client where I'm, you know, I've got like a, some new software deployed into their workspace and I'd be saying, okay, you know, look over, see the, on the top left of the screen, there's a red button and it says, it sounded exactly like that.

David Guttman: Yeah, 100%, yeah. So really the message to your, to Elton John, is you just have to train the thing a couple of times.

Ask the Agent Before Asking the Expert

---

Jonathan Stark: And so what happens if, what happens if it just like, I know there are scenarios, at least with me, where it's hitting some bug and I can't figure out what it is. And then I switch back into that mode of like, same thing, like pretend I don't know how to code, pretend I don't know anything about LinkedIn or anything and just zoom back up to the level of kind of like, here's what I want, here's the outcome I want, you figure it out.

David Guttman: Yeah.

Jonathan Stark: So like, what's that, what's a scenario, if you can think of a scenario from like an actual client engagement where someone was maybe micromanaging it or asking you questions instead of asking their claw?

David Guttman: Yeah, I mean, that's, okay. So that last bit touches on something that's very common. That's a little, that's, it's very much some muscle memory that I'm trying to build on the call. It's very common, where someone's on the, on the, on the call with me, I'm the expert. And so they want to know like, well, how, you know, how would I do X, Y, and Z? Um, I'm always, you know, I'm not always going to be there with them. And so what I want them to internalize is that they can ask the, the agent. And I think that's, that's a little bit of, you know, what you're, what you're getting at. Um, so agent, you know, that, that whole, um, sort of pattern that I mentioned of like stepping, stepping it through each individual thing and then writing the, you know, the skill and then putting it on, you know, a schedule, you know, if you were to ask an agent, like, Hey, what's the best way for me to like teach you how to, you know, do this task for me, the agent would probably tell you the same thing, you know, it's, it's, it doesn't really require me to say that. Um, and so the meta skill, the important thing for a client to have is that whatever the situation is, as long as they know what the goal is, and they can articulate to the, the, the agent, they can generally trust the agent to be able to, um, give them good advice or instructions for how to work with the agent to, to achieve it. Um, you know, in terms of like how much to micromanage when getting the agent to do something versus letting the agent figure it out, um, you know, I think if you have, if you have a process that you know works, you know, meaning somebody is already doing whatever this is, and they have a process, I think it is generally valuable to just have the agent do that process, because you cut out all the uncertainty of whether or not the process will work, you add the uncertainty of whether or not the agent can follow the process. But generally, you know, if you're limited to the browser, or whatever it is, like, decent, decently high confidence on that. Okay, that said, if you don't have an existing process for whatever it is, then yeah, I think the agent is in a good position to know what its capabilities are, it can often find avenues of attack that are not going to be apparent to you. So in a browser, there's all the interface that you have where you're clicking buttons and navigating and entering things into boxes and hitting enter. But your agent has access to the internals of the browser. And so it can actually see all the real network traffic that's going back and forth. And it in many cases can completely circumvent the UI and not even need to click buttons or or use the keyboard to type into boxes, it can just directly trigger what the button click is triggering or directly trigger what entering into it, you know, a text field and hitting enter is doing. And sometimes that can be a lot more reliable and faster and efficient. Other times they can get you into trouble. But, you know, talking about it with the agent, and even if even if it starts to get technical, you can always, you know, you're in control, you can always say to the agent, like, Look, I don't understand all that techno, you know, babble, explain to me, like, I'm five, and it'll do it, you know, it is there. It is there to help you. And it will do it. And in any way that is most helpful to you.

Jonathan Stark: Yeah, I want to emphasize that I don't want to zoom past that explain it to me. Like, I'm five. That's not that you look mean that literally.

David Guttman: Yeah, I mean, so on Reddit, that's a that's a very popular. I don't what initialism Eli five used all over the place and Reddit, Reddit, Hugh and other places on the internet, a huge part of these LLMs of their training. So it's, it is something that it is very, very comfortable and good at doing has no problem has no problem with using with you telling it to explain it to you, like you're a moron, it is happy to do so.

Jonathan Stark: Mm hmm. Yes. And to kind of expand on the, the kind of like, you figure it out, and it maybe has an even better way than, than your checklist, or maybe you don't have a checklist yet. Like, there's been a couple of cases, you know, probably a half a dozen times where I was kind of in over my head with like, I, you know, I was a web developer for a long time, but like the, the DevOps and network security and all of that stuff. Was not my forte. Yeah. And it's great. Yeah. Yeah. Yeah. So it just be like, I don't know, just, just, I want to, I want to be able to send a URL to someone else so they can see this, figure it out. And it might ask a couple of questions or it might just do it. And, and it's wild when you cross one for me, when I cross over that boundary line from like, when maybe I could micromanager, maybe I could have done it myself if I felt like it into territory when I really couldn't have, like, it would take way too, I wouldn't, I would just stop doing the thing. Right. You know, it's like, I'm not gonna figure out how to exchange SSS keys over a socket, you know, I'm just making stuff up, but like, I'm not getting into the socket layer. Right. So, you know, and it just, it's normal for that. It's just like, it can all of a sudden at the edge of my expertise, all of a sudden it's, it's expertise goes way outside of mine and it might not be a global expert or have like the right judgment or taste or understand the whole context, but it's way better than me. And so it's good training for me when I get to one of those places and I'm just like, I don't even know what to tell you to do. So I'm just going to tell you what outcome I want and you figure it out. Sometimes I'll say, give me three different ways that you could do this. Cause I want to kind of, there might be something security-ish that I'm nervous about or something, but yeah. Okay. So, so you've done a bunch of these setups for, for people who really, really run with it, like clients who are really like, Oh, this is, this is perfect. You know, like it turned out that they get it, they understand they got the muscle memory. What do normies think once they get it? Cause it was kind of like you said earlier, people are like, why should I care about this? And they kind of have a sense that it's a big deal, but they really, I find it hard to explain to people. So it's no wonder. And then once they see it, they see the light, like what, how do they react to that? Do they, you know, do they add new automations and new tasks for it to do?

Life-Changing Client Use Cases

---

David Guttman: No, like I have one, one client who literally said that it changed his life. Um, for the better, I hope. What's that? Yeah. Yeah. Um, yeah. Uh, and, and he's, you know, he just goes, goes nuts with it, you know, because there's all of these things that in the past he would have wanted to do, but it was too expensive, too time consuming, would have involved somebody else. It was almost like all of these things that he wanted to do required permission from somebody else. Like he had to get somebody else on board and this, you know, this is a founder CEO, but, you know, even in larger organizations, you know, or maybe especially in larger organizations, like even if, uh, you're at the top, it's very difficult to just change priorities, pull somebody off of like an important project just because you, you know, have a whim, you know, that you want explored. And so it just gave him license to sort of exercise that, that creativity that would have been harder in the past. And so, you know, went from, you know, initial ideas where he'd be out at a conference and he wanted to just send it, uh, photos of, of people he met, their business cards. And then his open club would take the photo, see who it was, search around for them on the internet and then fill out the CRM in a pretty detailed way, like more detailed than he would have. Uh, and then set up a reminder for him to contact them like a week after the, um, the conference, you know, so that was kind of where it started, but then eventually he just like pulled it into like every aspect of his life. Like he had a, he has like a, a web app, which is like a OS for his life and his family and set up WhatsApp that his, his wife can talk to. And he's just like building, you know, just all these apps like that are, are, and, and sometimes I think they're like single, single, single use, like disposable things are just like a particular thing that he's like, oh, this would be fun for this one particular, you know, pitch or demo that I'm doing. And it'd be really cool to, to show it off in this particular way, but he's never going to use that again because that was totally, you know, set up for this one pitch and this one client that he was, that he was doing. Um.

Jonathan Stark: So, I mean, it's kind of interesting that when the, when sort of like the cost of, uh, of, of the software, you know, what it costs him or costs you to do if you have, if you have OpenClaw, it winds up being so low that you can treat, you know, software that is really well built, um, and, you know, really functional app, you just treat it as disposable or in the past, that would have been unthinkable because it would have cost thousands of dollars. Like you wouldn't have just built something like that just for use for an hour or two and forget about.

Disposable Software and the Future of Developers

---

David Guttman: Yeah.

Jonathan Stark: So that's maybe a good place to, to start wrapping up. Let's zoom way out, uh, outside of OpenClaw, just more, maybe even as agents, it's pretty important.

David Guttman: I like the agent's piece is really key because cutting and pasting code from ChatGPT is not the same thing. Um, doing like having these inside of a computer that can do things on your behalf, manipulate files and send things around and all of that and network requests.

Jonathan Stark: And it, I agree with you. I think that, I think that there's a lot of software, maybe the majority of software that gets written in the next couple of years, two to three years, let's say, would never have been written before. Uh, because the, the, the cost to do so has completely collapsed and you can make, it makes sense to have a one-off app or something that, that is only available on your local network or on your Tailscale that is just single user just for you, you know, or just the people in the household, whoever's on the, on the, on the wifi. And so like, what, if we're going to speculate for a second here, what does this do for the soft software as a business? I mean, you know, superstruct makes a lot of money renting out devs. So does this become a hybrid thing? Are they centaurs or like, or do you deploy for some clients a swarm of bots and no humans? Or like, what, what does a software developer in 2026, how should they be thinking about this? Do you think in your opinion?

David Guttman: I mean, like the, the visual image that I have is like, you know, Paul Bunyan or something like that. Right. Like the, like, you know, the American myth of, of capital and, and industry. And, you know, if you're an amazing, uh, software engineer and you were, you know, Paul Bunyan and all these trees, you could chop down with your axe and everything like that. And then, you know, I don't, I don't even remember if I'm getting the myth right or something like that. From what I remember, it's anyways, basically it's like the idea that you got this lumberjack who's like huge and strong and, you know, can do everything, you know, with his axe, like by hand. And then pit it against, you know, these machines that, you know, do it all, you know, more, you know, in a, in a industrial, like automated fashion. So, you know, axes, you know, axes versus, uh.

Jonathan Stark: Steam shovels or something.

David Guttman: Yeah. Or, or, you know, chainsaws or whatever it is. So if you really pride yourself on your craft and don't think that any automated way of production is going to be as good as your artisanally handcrafted, whatever it is, um, you know, I think it's probably just going in the same direction. Now, that doesn't mean that as an industry, you know, production is going to halt or that humans aren't going to be involved, but it just winds up being higher leverage, right? You can, you can wind up with somebody who has whatever expertise it is, chopping down trees or something like that. And now they are managing just a lot of machines and capital that is, you know, in charge of, you know, lumber. Um, and I see no reason why that wouldn't be the case with software that you can have any particular software engineer who has an interest in, in operating this way. Um, you know, just have a lot more leverage. Like they don't have to be the one putting their hands on it. They can be amplifying their output through the agents. Now, in some cases that may not really look like software engineering anymore. That may look more like product management. Um, but it kind of depends on what that output is. I mean, I said this before, you know, if your software, uh, can kill people, if it's not working correctly, I don't think it makes sense to be as hands-off and, and, you know, just trust that it, it looks good from the outside. Um, I think there's a lot of, you know, more, um, attention and care that would, would need to go to it. And I think that probably increases the amount of like in-depth software engineering knowledge that, that has to be involved. But I think there's a lot of products out there that probably can get away with fewer, um, software engineers in the mix. Like I think a lot of solopreneurs and founders can get, get, uh, very far, um, just on their own with the agents that they have. And so I do think that demand for engineers is probably going to decrease in a lot of areas where they are no longer needed for things that are less serious and have less, I guess, risk of failure or less risk, uh.

Jonathan Stark: Yeah, the impact is low.

David Guttman: Yeah, the impact. But, uh, I see, you know, I don't really see the demand for software as a whole decreasing. And I think there's probably lots of areas where that expertise is very valuable. And because every, every software engineer can have their value amplified with these agents, they do become more valuable because, you know, an hour of their time, uh, can produce more. Um, so, you know, they wind up being something along the lines of like an engineering manager whose value, uh, is, is, I guess, is more like their, their percentage, uh, efficiency or output increase to a team of engineers, uh, rather than just, you know, an engineer whose value is just how much they can produce, uh, on their own.

Productivity, Pricing, and Value

---

Jonathan Stark: Yeah. And I mean, I agree with all that. And while you were saying that, all of a sudden I was just like, I remember when it made me think of when Rails came out and, and the entire login module, which used to be a huge fricking thing that you had to write by hand was just like a gem. And I was like, wow, this is my reaction was, this is great. Cause I didn't build by the hour. So business model comes into it a little bit, but so I didn't build by the hour. So I was like, oh, so like this is in WordPress. Like I used to make websites by hand and then WordPress comes out and it was pretty great for people who didn't have enough money for a webmaster as they were called back in the day. Yeah. So it's not like these sort of, in one sense, it's basically just a new productivity enhancement. Really? I don't see any significant, I don't see anything that differentiates it from any other kind of productivity enhancement. It's like it increases your productivity. If you trade time for money, that is very bad for you. You do not want to increase your productivity, but your clients do want your productivity increased. They want their productivity increased. And you know, this is just one data point, but I, I'm just looking at my coaching Slack, 100% of the non-developers who I'm coaching right now are using AI in this fashion. Like I have one that just built apps on lovable, like without even knowing what it was, just somehow made an app on lovable, completely non-technical, a non-software person. Another one built a prototype of an app with a Postgres backend. That's just, I'm like, guy's a doctor. They're both doctors actually. So, and the list goes on. Like it's a bunch of software that never would have gotten built. It's like soft, it's, I feel like it's like the developer, no developer lost money because of that. Because they wouldn't have paid for, like it wouldn't have ever just existed.

David Guttman: Right, right. Yeah. I do think just to pull it back to the premise of the show, I do think that it might. just be a perception in the public, but there's enough headline front page news about AI on an almost daily basis that I do think that, and there'll be enough people who make some kind of cool app unlovable by themselves, even though they're a doctor and not a software person, that the perception of value of software will go down because they'll be like, oh, well, I can make this myself if I really tried. I just don't wanna really try. It's just like the FileMaker days when all of my clients had kind of made a FileMaker database for themselves, and then they painted themselves into a corner and had to call an expert in to untangle it.

Jonathan Stark: Sure.

David Guttman: So the value was sort of like, well, I could do this if I felt like putting the time in. So you kind of always had these people that were new enough to be dangerous, and that brings the perception down, the perception of the value down, but if it comes down by half, but your costs go down by 10x, then there's still plenty of room in there for you to be super fast and productive and make a really good margin if you're giving fixed prices based on value.

Jonathan Stark: So with that said, I just wanna thank you so much for sharing all that expertise. It's just incredible how fast things are changing, and I love that you're so sort of up to date and on the cutting edge of this to share with the audience.

David Guttman: Yeah, happy to be here. It is a wild, wild time. Yeah, it's software's fun again.

Jonathan Stark: Where should people go to find out more about what you're doing?

Where to Find David

---

David Guttman: Sure, so Superstruct, my consultancy, that is superstruct.tech, probably in the show notes, and then you can find me on LinkedIn, David Guttman, and my fun projects, david.app.

Jonathan Stark: Yes, one of the coolest websites I've seen in a long time.

David Guttman: Yeah, check out david.app first, and then you'll be like, oh man, this is crazy. Thanks.

Jonathan Stark: All right, folks, well, that's it for this week. I'm Jonathan Stark, and I hope you can join me again next time on Ditching Hourly, bye.