Jeff
00:00:00.000
AI has a lot less layers of problem it needs to solve. At the same time, it doesn't need to do any of those, infrastructure techy things. It only needs to focus on, Business levels, business logic, if it's implementing a custom ERP system, it only needs to think about how inventory moves. It only needs to think about how the sales order should be structured. It only needs to think about how the cost should be calculated. So, in terms of no-code/low-code it's kind of a, guardrail for, the AI To only have to think about the business layer, only have to think about the business logic of it all.
Darin
00:01:42.512
Viktor. It seems like every week there is a brand new tool that promises you don't need to write code anymore. This keeps on happening over and over and over again. Today's version of that is vibe coating. Right?
Viktor
00:01:55.472
I mean, you always write code. It's just the question how when a white code I write code, if a drag and drops something in Dream Weaver, I wrote code. And when I use Vim, it's also code.
Darin
00:02:08.038
Some people might argue with you on that. On today's show, we might have a person that might argue with you a bit more. We have Jeff quo on from Raic, that is R-A-G-I-C. Jeff, how are you doing?
Darin
00:02:22.682
What did you think of, Victor's statement there? That no matter how he's doing it, he's still writing code, whether he is using them or emax, which is what you should be using, or drag and drop or today's world of vibe coding just through text.
Jeff
00:02:38.848
I think he definitely should be writing code. So at today's stage, more or less, you should be writing at least a little bit of code.
Darin
00:02:47.889
but he was arguing that no matter which way he's doing it, whether he's actually physically writing the code or letting a tool do it, he's always writing code. cause you're coming at it from a no code perspective, right? I've seen no code play out for years. I'm still not a big fan of it.
Viktor
00:03:05.373
Just a reminder, I, I remember conversations when people see people. Were claiming that writing Java in its first days is not writing code. That's too much of a obstruction.
Jeff
00:03:17.268
yeah, from a historic point of view, it's like from the first gl, second GL, all the way to four GL and like pretty much the fifth GL vibe coding and AI generated coding. conversational coding is like the next generation that the generation that was promised years ago. So it's kind of still like coding, although it's conversational.
Darin
00:03:40.807
It's amazing. you went back to two jail. I'd actually forgotten about. I remember three and four gl, but two, it's like, oh wait, yeah, there was a two before the three. Okay. So you've been working in no-code for a number of years? close to 20 now? Probably 18 to 20 is what it feels like.
Darin
00:04:00.023
oh dear. is it ever gonna become your doctoral, whatever the word is for that one? I don't.
Darin
00:04:06.418
Okay. So that's an interesting point. Why isn't it graduating? Why does it seem like no code, low code is just sort of in this, what I'm gonna use the word rut as an outsider looking in, I haven't seen a major shift in low code, no code in years.
Jeff
00:04:23.842
Yeah, that's kind of an interesting point because even back in 2000 there were no-code tools, like Microsoft Assets was kind of a no-code, low-code tool and FileMaker of course. So, was always, there it is, it's there to try to solve people's problem of. You writing code because writing code's not for everybody. it has been there for such a long time to solve this very problem, although it looks like people won't be needing to write code as much. professionally, I, I must emphasize professionally, for me, I think generating code is still. For professionals. I think, AI assisted coding is for people who understand the infrastructure understand the underlying what the code is about. it's for them to do, it's actually quite counterproductive for non-developers who has no idea what the, IT infrastructure is doing to. Generate a lot of code. They can, but they can't maintain.
Viktor
00:05:31.429
I have a lot of thoughts on vibe coding on for non-engineers, right? Software engineers. But before I go there, kind of why would vibe coding not be for everybody else? It seems like the most intuitive way for, ah, create the Tetris game for me. And, uh, you go there and you watch a few YouTube videos and you're back and you have a touches game. Right now quality is a different thing, but now we are not talking about professionals. Right.
Jeff
00:05:58.293
yeah, quality would be kind of important if it's not a one shot thing, if you want to run it for, uh, years. So. The problem would be maintenance and if the whole project ever kind of scales. I've been doing, these kind of AI assisted coding. Both professionally and a lot of side projects because I just like coding. I like creating things. I have quite a few experiences when I was doing these fun side projects and I, don't like to look at the, code that closely as I'm doing professionally when I'm vibe coding, like when I'm just generating code telling AI what to do. that usually. End up pretty bad if I don't look at the whole direction where AI is going. So the whole structure often gets pretty bad and it gets stuck in like, endless loop of fix this and AI say, oh, okay, I see the problem here, here, here's the problem. And I say, no, no, fix this. And oh, okay, okay. I, I found the root cause. I found the root cause. Here's what the fix is. until I go in and look under the hood, I see. Oh. The ai Got it. The, the direction is all wrong.
Viktor
00:07:14.922
isn't that the argument that holds, give or take equally for low code? No code as for vibe coding or what's the difference?
Jeff
00:07:24.073
I think the biggest difference is that ai, right now, the, at this stage, it's not really good at looking at things from every single level at the same time. When it's generating it usually does a lot better if it looks at things only one level at a time. If it's looking at like security and it can think very well about security. If it looks like, uh, look at performance, it can look at performance very well. If it look at the overall scalability of the, code of the logic, it can do that pretty well, but it's, it doesn't seem to be doing. Everything at the same time. Very well,
Viktor
00:08:05.118
that, I would tend to agree. But is that something we can solve with low-code, no-code or it's, still the same problem
Jeff
00:08:12.379
that's kind of what I'm thinking, that what local can, solve is that AI has a lot less layers of problem it needs to solve. At the same time, it doesn't need to do any of those, infrastructure techy things. It only needs to focus on, Business levels, business logic, if it's implementing a custom ERP system, it only needs to think about how inventory moves. It only needs to think about how the sales order should be structured. It only needs to think about how the cost should be calculated. So, in terms of no-code/low-code it's kind of a, guardrail for, the AI To only have to think about the business layer, only have to think about the business logic of it all.
Viktor
00:08:59.595
Uh, so it would be something similar to AI or lms, but with very strong skills. In a way, right, that guided, so if a company would create skills, let's say, or CPS or commands or whatever kids are using today, uh, that kind of guide it and maybe some kind of, you know, semantic search from company references, documents then we would scope the problem to a similar size, like, uh, low-code, no-code is trying to solve, am I in the right direction, give or take,
Jeff
00:09:32.910
I think, what you're saying is in the right direction, but it's not gonna be solved by skills because, with skills, you're still doing it in one prompt. So doing multiple layers in one prompt. It's just not good for today's LLM.
Jeff
00:09:54.061
yeah, but they're still processed in, prompt, in NOCO local. the output of the LLM is not actual code.
Viktor
00:10:07.558
Ah, okay. Okay. Did not get that part. Okay. So basically, let me rephrase it to see whether I understood. Right, right. It's kind of basically that we are talking about. In both cases, AI or l LMS doing the work, but interface being different, right? Uh, whether it's a chat based agent versus, some kind of, I'm, I'm imagining graphical user interface where, where you are guiding it through selecting specific features or components or something like that, right?
Jeff
00:10:41.788
Yeah, sort of. for no-code, low-code, I think especially no-code, AI won't be generating code in that sense. AI will be generating whatever the no-code platform like magic is. Having to define all those, applications is gonna be able to generate all those definitions
Viktor
00:11:02.746
then the low-code, no-code tool behind the scenes would actually generate code, so it would be. Like a layering between code and LLMs and Logic or some similar solution would be equivalent. I'm just trying to picture in my head the architecture, uh, would be equivalent to LMS using tools like any other, right? Instead of, uh, I dunno, uh, talk to this database or instead of talk to A-W-S-A-P-I, it is essentially talking to your API.
Jeff
00:11:31.148
yeah, that's kind of what NOCO Tools like US is, doing. We're wrapping all those, moving parts of our, application into tools that it can just use and not worrying about doing any of those work underneath.
Jeff
00:11:51.854
well, when using those tools, it doesn't really need to test the tools themselves. it needs to then do the end-to-end test, but not the tools themselves. Like when clock co calls the. Jre or call calls or any of those tools. it just assumes that the tool is sound but it needs to do end-to-end test and like unit test for things that it generates, but it doesn't need to test tools underneath it.
Viktor
00:12:18.455
That's not necessarily true. I mean, what you're said saying is true, but it does constantly in a chat like experience, it does constantly test the outcomes. of the outputs of the tools, right? So I just call this tool, the tool, writes the code. I just call some either tool, maybe write population to create a Python script, and that Python script will test the code that I just wrote. Right? And I guessing something similar is happening there, right?
Viktor
00:12:51.712
I have so many questions. I'm guessing right now that you are instructing models, kind of. Okay. Those are the components you can work with. What's not right, but eventually, probably the goal is to be trained on it. Right. so it, it doesn't really need, you don't need to teach it how to use those low-code, no-code tools, but it is part of its training. Right. how would you tackle that?
Jeff
00:13:16.200
well, luckily when we're doing these kind of, integrations, most of these models are already kind of trained on IC documentation. It already Reddit because, you know, when we call it, it already knows what magic does and the. Basic training part is pretty much done and we just need to do a little bit of, uh, rag for it to find more obscure information on, on raic, and by a lot of custom prompting it, is able to do pretty sophisticated no-code application generation.
Viktor
00:13:55.380
reason why I'm asking that is I saw a tremendous difference. So for example, I'm, uh, I'm probably one of the very few people who actually like, really new shell, nobody even knows about it. It doesn't matter replacement for Shell, right? and the outcome that I get from s when working with no Shell is. Inferior on so many levels to bch, Even though I don't like BCH code. Right. and I, I attribute that to the fact that simply the source of information and knowledge that it has about no is just very, very low compared to, you know, infinite amount of examples and conversations and what not, it found on, on b. and I feel that this is not necessarily related to low-code, no-code, but in general, right? How do we, doesn't matter whether it's your tool, my tool, this system, that system, how do they get on the same level as a common system otherwise? Because otherwise LMS will always stick to, today's equivalent of mainframe kind of like, oh, this what I know the best, right? This is what I do.
Jeff
00:15:03.684
we faced the e exact same problem because of course, even though they know magic, is not like their favorite thing, honestly. So that's kind of a technical challenge that we need to tackle. So it's. Still solvable. So we need to break the problem down for the LOM by multiple prompts and have more clear description of the tools and breaking the tools down into simpler parts. So. It's mostly a technical challenge for platforms like us or any tools like us to actually break the problem down for LLM to solve one at a time if it's apparent that is not really good at lists. After some testing, we have to try to break the problem down and make it into simpler and more generic parts that doesn't require as much, specific tool knowledge.
Viktor
00:15:57.515
Could it be content problem instead of being technical? Maybe, if you would bomb Reddit with, uh, with the, or Stack Overflow with the information about re maybe that would be the solution. Kind of. Maybe it's just a question of how can we propagate information about this everywhere.
Jeff
00:16:18.054
I think that kind of will be the future because like when we are re rewriting our documentation, we are literally telling AI to help us generate the documentations for first, for humans to read. Secondly, for crawlers to read, and thirdly for. Ai, LM story. We want them to be able to learn rad really well. So because we, we think they're, they know rad, they know Rad, API, they're just not that good at it. So I think how to propagate those. material for AI to learn how to propagate those, examples and documentations would really be important for, any software so that, a LM learns their product well.
Darin
00:17:03.476
I'm gonna go the other direction. I, I think you're going in, in the wrong direction. I think it should be AI first, web crawler second, and humans third.
Darin
00:17:16.496
No. In the order of importance, that's what it is. Um, you gotta make sure when you write that. Th this is a Viktor, and I have been talking about this problem. It's not a problem, but it's a problem. It's like, I wanna write something that's specific for AI to consume, and that's going to sound very different than how I would want a human to read it.
Viktor
00:17:45.821
When was the last time? Okay. Be be honest, Darin, I'm going to attack you here. Right? When was the last time you actually spent an hour reading something yourself? Yeah,
Darin
00:18:04.153
I, will say as of today when we're recording this, there was a blog post that came out of one of our favorite tools that they did, a version two. So I read, I found it through Reddit, and then I went to the blog post to read that. Because it's too early for it to show up in my every 30 minute chat with Claude. or at least it probably was at that point in time. 'cause it had just been published. You know it. I'm looking at all this and I'm listening to you, Jeff, and what I'm hearing is you are not anti AI at all. my question to you because you and Viktor are probably close in age, ballpark, much younger than me. Uh. Has this been the most fun that you've had coding in a few years? A decade, probably, if not longer, or is it because you've been working in low code? No code. Everything's been fun.
Jeff
00:18:58.866
well, first of all, the, the, the NOCO Pool magic is mostly for business. So, that's true. I have, I haven't really think about it that way because like for the past. Few months or even year. I've been doing more side projects than ever because yeah, I, I had a lot of fun, right? And vibe coding or, you know, just AI assisted coding, creating a lot of even stupid things I've been even be able to do it with, with my kids who hasn't learned any coding. So it's kind of fun to be able to do that with them and trying to create stupid stuff.
Darin
00:19:33.327
Let's assume for a moment your kids are gonna go into computer science or some variant of that. Are you concerned about what they're actually need to learn nowadays? I mean, do you, do you really need to go to college? I mean, we can argue about college all day long, but I, I don't. We used to say, okay, get involved in an open source project, you know, build up your chops there while you're in high school, college, and then you're probably set for an okay career. We can't say that anymore. I definitely can't say that anymore. What are you thinking? Because again, I'm in the states, Victor's in Spain and Jeff is in Taiwan. Right. what does it look like over on that side of the planet? Because that's just seems I, I don't know what we're gonna do. I mean, we know we need electricians, we know we need plumbers, 'cause somebody has to build the data centers and we need masons. Do we need software developers anymore?
Jeff
00:20:28.621
now, I'm still gonna guess yes, but less, but I'm gonna wonder how they're gonna be trained because, um, to write good software, you still need to read a lot of code generated by ai. I. You, you have the reading code, it will be a, a more important skill than writing them, of course. But you have to look at a lot of code, like three times as many as before in our age when we are writing code, you have to look at so much more code yeah, you're, you're doing a lot of reading of those AI generated code,
Viktor
00:21:10.895
I, I think you're right for today, right? I can easily imagine how we will be reading less code, in the future. That being said, I think that the need for software engineers will not drop. It'll. If anything increase. Because the critical skill, and this is my perspective, right in the future even now, is taste, and taste is based on experience and skill and, and knowledge, right? Yeah. Oh, we all have a taste. What is a good painting, right? But somebody who is really experienced with paintings can tell you easily kind of, this is worth a hundred K and this is fresh, right? I do feel that we will be reading less code and writing less code, but understanding code is critical to distinguish somebody, you know, like a design kind. Can everybody, this is long before ai. Can everybody with some tools like Photoshop or Carava create a design? A hundred percent, yes. That still does not mean that when I create it, it doesn't suck big time.
Viktor
00:22:23.060
Right that that's simply because I'm not trained on it. I don't understand design. What I'm asking, kva, and this is before, nothing to do with ai. I'm not asking the right questions. I'm not being critical as much as somebody who is in design, right.
Jeff
00:22:41.232
that's kind of the, question I'm trying to answer when I'm, and doing these projects with my kids. because I, I was thinking, oh, so my, my years of experience doesn't matter anymore. So I, I wanna find out like if, if, you know, my kids can do their Teenage kids. So if they can do pretty much the same thing, but in reality the questions they ask, the way they talk to AI is just still fundamentally different from how somebody who knows what coding is, is still fundamentally different because what they're describing is always on the behavior side, it's never about how the code is. How the overall structure of the, code is. And I think for now I still have to kind of peek under the hood to understand what's actually happening, to make better prompts, to kind of steer it in the right direction. So what I'm saying is for today's models, but if you're able to trust the model completely, when they summarize. The code. When they tell you what their code is doing, then probably. But right now, right now, no. The, the, the summaries a lot of times, and they're kind of misleading. They, it doesn't give you the whole story how the, whole application is working. I, tend to believe at the time when I generated, I always do, but I, but when days goes on, I say, wait a minute, are you sure that's what you're doing? Then I find out, yeah, actually no.
Viktor
00:24:17.143
I feel that the problem is different. That we will eventually get to the point where I can trust AI with code. What? I cannot come to the point, probably let, let's say, or I dunno, 90% or whatever, high trust, I mean, I don't trust people with code just to be a hundred percent transparent here. So I, I dunno what we're asking here for, but the real problem is different. The real problem is that, can you. Two things actually. Can you read my mind? Because I'm definitely not capable of expressing everything I need, in one go. And I don't think that anybody's working on mind reading, type of LLM And second issue is you have all the public knowledge in the world. You have no idea what we're doing here. Right. You have no idea what, exists in this company. You know, you have no idea how it's structured. You have no idea what our business is and so on and so forth, right? And I feel that that's the biggest missing piece, How do we extend public knowledge? With internal knowledge. ' imagine that you just hire a genius. You, you hire the best coder in the world and you tell that person, okay, cool. Do stuff first day. And that person will do something and it'll end up in AWS Lambda. and you happen to use only Azure and you happen to use only Kubernetes and you happen to use only the rust for whatever reason. Right. is it functionally correct what that genius made? Yes. Has it, does it have anything to do with your system? No.
Jeff
00:25:55.122
Yeah, I think one of the great thing about today's, um, AI assisted coding tool is the plan mode, and I think the plan mode is. Kind of the solution to this kind of problem. But it's a good start, but it's definitely not asking enough questions in the plan mode. Like asking every, question about, you know, your current infrastructure, whatever's on your mind, because of course, AI cannot read your mind. But, uh, with. sophisticated plan mode, it should be able to, extract enough information for it to, you know, make good decisions about the design.
Darin
00:26:33.789
Let's assume you're a CEO. You are a CEO. Let's assume you're a CEO or a CEO of a 50 person company. No dedicated it, that's always been outsourced or as, as needed. This company needs a CRM, an inventory tracker and a customer portal. What gets no coded? What gets vibe coded? What in general gets built?
Darin
00:26:57.024
And get, let's get bought. Yeah. Finally, what gets bought, what do we do in 2026
Jeff
00:27:11.544
at least, me at today's stage in 2026, would the smart way would be to leverage no-code to do that Because with, no code, you, you will be able to generate most of the, the scaffolds and the, the applications you need with AI to assist you with all these applications. And you, you will still be able to go in and tinker with all the setups and all the forms and reports that AI generated for you. So at today's. Yeah, 2026, that would be a pretty good combination so that, you know, you could AI generate the, the hard work for you. And you can go in there and look at it from a no-code tools, perspective. So there's not a lot of technical expertise needed when you're talking about like creating a CRM application or a business application like that with the assistance of tools like rad. But if you're building something, not, business applications or like in other realms, I'm not sure, but in your the, the case that you mentioned, I think that would be a pretty good combination. in five years, I think. my guess is that, I will still trust that AI do most of the work, but I will certainly want to be able to have fine grain control so that I can go in there to the application and knows exactly what they built. I don't need to look at the code. I can just look at the, the no-code UI to understand every single rule and every single. configuration that has made for the application. Yeah. But I don't need to look at the code because there's no code. I would still need a comprehensive view of what is being built and the logic that is being run
Viktor
00:29:13.039
What is the. Background code, kind of like what is being generated that I don't see as a user, which language is behind the, tool.
Jeff
00:29:22.129
there's, no, no, there's no code. It, it generates, um, forms and report definitions.
Viktor
00:29:27.363
Okay. Okay. It's not, it's, it's not attacking no code, below code, whatever the term is. Anything I want is for specific types of applications. Right.
Jeff
00:29:35.953
Yeah. Yeah. It generate different types of definitions, like J configuration files, XML configuration files, even CSV files. So all sorts of, configuration files.
Darin
00:30:04.433
I wanna go back to what you were talking about. You were, you talking about with ai, right? We were going through that scenario of building something in five years ahead. Let's go backwards five years. Let's go back to 2021. We didn't have AI then. How much of a, I mean, what, what was it like then? Because we, we know what we can do now, right? We can sit there, we can have a chat with it. Something will get generated for us in, let's call it minutes to a half hour, we can see something, we have something clickable, but I would imagine five years ago it wasn't minutes to and a half hour to what we could have today. Is that correct?
Jeff
00:30:41.208
Yeah, we were doing tons of copy pasting, asking GPT GE generate code snippets and trying to copy paste to our code base. We were doing a lot of that, but 2021 we're, Yeah, that.
Darin
00:31:03.092
yeah, like it, to me, I'm, I'm thinking from his perspective today, onboarding a, a prospect or a client has to be a heck of a lot faster than what it was. Before you've introduced AI features into magic, am I wrong in that
Darin
00:31:22.533
meaning like, like the time to value has to be greatly reduced compared to five years ago, three years ago? Am I wrong? I
Jeff
00:31:31.868
It's reduced, but I'm not sure about the greatly reduced part. the problem will be with the inertia of, of users when they're beginning to learn about a new application. AI has certainly reduced the amount of things that people need to learn in order to make the first things work in a product. But. still, people are surprisingly stubborn that they, a lot of times they wouldn't even say the first thing to AI to tell them what they want to build 10, 15 years ago when we, when we introduced rad, that was the problem because we thought like creating a form on RAD was simple enough. Just type in the, field names that you want you just type in the field names that you want and it becomes a sheet that you can use. We thought that was simple enough and that was very little work, but Some people get it. some people do it, some people just, eh, I'll do it another day. With today's ai, it's still kind of the same thing. People tell AI what they need to do, but a lot of people just look at it and just leave. Leave it there. people are surprisingly stubborn in the. If they're going to do any of those typing.
Darin
00:32:50.702
That seems strange to me. I mean, it, it shouldn't surprise me after 40 years of doing this, but
Jeff
00:32:56.567
Yeah. people don't like using AI like dialogue features when they're encountering a new product. That's kind of our finding, so we're trying to work around that. But most of the time people don't like to. Engage in dialogue with AI in a new product, unless it's in the familiar setting like they're already using. If they're already using GPT there, the behavior, the, the habit that they're already in.
Viktor
00:33:22.897
are you trying to say that the, one of the problems is that people are not necessarily inclined to use AI in products they're already using? But they're very happy trying it with products that they never used before. Because obviously there is a user base of ai. Right. I'm just wondering how much, you know, kind of, oh, I'm a long time Photoshop user. I don't want ai, uh, whether it is this new tool, I dunno, nano banana and then kind all of a sudden, I'm very much into it. Right. Because I never experienced that. I have no muscle memory, I have no experience with it. It's okay. Is, is that the direction where you're going?
Jeff
00:34:04.942
yeah, I think it's kind of like that, because that's both the numbers that are seeing and what I'm actually doing, because especially. With any tools that I use, like any source applications, they all have AI assistance now. They all have AI things, AI dialogue there. My first intuition, you know, to ask myself is just to try to close them, try to get away from, get them away from me. I feel that, oh, I'm, I'm doing that too. So that kind of matches the number and the, the patterns that we're seeing from. The people on our tool, their initial thought is to just close that dialogue. They don't want to talk to ai.
Viktor
00:34:47.361
that means at the same time, and I, I'm in the same boat, just to be clear, I'm now criticizing myself, that if usage of AI continues increasing, which it is kind of like we never saw the numbers, like those in the history before, that means that the future is actually not in existing players, right? Because if you say the numbers are going up like crazy. And people are rejecting AI in existing application systems, services, whatever. That leads me to a conclusion that the future is in nonexisting application services that are being built now.
Jeff
00:35:29.083
I think it's, I. it's non-trivial to build a habit of talking to certain ai. for example, I think clock code has crossed that threshold that people are comfortable with talking to it. As an AI dialogue tool and like in years ago, GPT that has crossed the threshold that enough people are comfortable enough talking to its AI as the main tool. It's difficult for a person to trust that AI tool. But if you want people to engage with the AI dialogue box in there, it needs to cross a certain mental threshold.
Darin
00:36:08.212
Is that threshold? The only thing you get to see is the text box. You don't get to see anything else. if you want to see all the knobs and switches, you have to go into a completely different mode of the app of the front
Jeff
00:36:20.407
yeah, if, if it's only a chat box, the session will be high. So we find that we have to kind of, not. Only be a chat box. We have to blend a lot of other things into with ai. Like if an AI feature is like kind of hidden, like there's AI inside, but it's just like a normal knob that you use, then the adoption will be much better. But if you ask the person to like really write a full sentence or a paragraph, the adoption will be lower.
Darin
00:36:54.489
It just seems silly too, right? Because is it the blank page syndrome? I mean, all they see is a box, so they don't know what. To do versus I haven't gone into the back end of ra, but I'm assuming there's, okay, click this block, drag it here, grab these lines, drag them here, and that's a more familiar scenario versus, Hey, I wanna form so I can capture customer information. That's a, a solved problem for decades.
Viktor
00:37:22.014
I will push on that with two examples, and you figure it out that in Yahoo, Google. Yahoo gave us the full list of stuff that then categories and what's or not, which we could navigate. And Google came up with a blank page And a box. Do you still have stocks in Yahoo? Darin?
Darin
00:37:46.113
Unfortunately, no. Okay, so I'm wrong or am I right? I'm not sure which way I was going with that. think Victor's thing is correct in a very twisted way of, if we can make it simpler, people should use it more. But the problem is, is that they become accustomed to, alright, so this, I think this is the, going back to what Viktor was saying, it's gonna be the new companies that come up with this. I'm going to take it one, one more step. It's gonna be the new employees in old companies needing to use tools. They're not gonna wanna do the block and drag the things they want to just type into a box. So I'm wondering if that that annoying corner popup AI thing, or one of the apps that I use, it's now started. Anytime I go into it, the whole sidebar opens up to do ai. I don't want that go away. Whereas if it just gave me that and gave me no other choice, it'd be like, oh, okay. I'll do this now. But I've been using the app for two years versus now if I've got a new guy coming in, they just graduated college. What IDs do you think they were using in college? Uh, at best, a terminal maybe. Right. I'm just thinking ahead a couple of years. It's like, I think that's where tools like Raic and others, I'm not gonna throw Raic under the bus. I think we're gonna have to evolve with the times to be thinking about, okay, people aren't thinking this way anymore. Now they're thinking just like they're playing me showing old olden games, they were playing Zork and uh, Hitchhiker's got the galaxy. Now we're back to text-based all over again.
Jeff
00:39:22.536
I think for me it's sort of from. Some of the bad experiences that we have with these AI chat boxes from the early days of AI that you know, most of them kind of sucked. People try to rush and have AI features in their product, so almost all of them sucked in the beginning. those are kind of bad memories of these AI chat boxes. We have. So maybe in the future, if they're generally more sophisticated and more convenient than turning the knob yourself, then perhaps the overall habit, that that behavior might change. But I think for me, like a lot of it is like, I just kind of know that most of those chat boxes don't really work well.
Darin
00:40:12.277
But again, my argument is now you're gonna have brand new people coming into companies that it's their first job and this is how they know to interact with computers. It's going almost going back to green screens. that's what it feels like all over again. it's all brand new. Maybe we can get rid of all the computers and go back to green screens. That's what we should do. you know, I'm, I'm thinking about all of this again, no code, low code's been around for a number of years. I'm thinking at some point there's ceilings, There's a vibe code ceiling. There's a, a low code, no code ceiling. Where do you think, are those in parallel? Are they close? Does one outrun the other? Where do you think they land today? It depends, right? What kind of app is being built?
Jeff
00:40:59.817
I, definitely feels like I could be wrong, but I have always felt that we're already at the ceiling but people just keep coming up with more clever ways to, put together. Because I, I don't feel that AI has been substantially so much smarter, but people have been able to, find new ways to Stack different AI calls together so that they become, oh, now they're thinking, now they're, they're able to do all those more complex things. They always surprised me on how they can make AI smarter with. A combination of a lot of AI calls, like for AI thinking chain of thoughts. they're basically putting together a lot of stupid AI responses, but they refine themselves and and make them smarter. I don't think AI is. Fundamentally getting a lot smarter. They are kind of incrementally getting smarter, but the way that we're using AI is a lot smarter than like two, three years ago. So they're generating drastically different results. But what changed is. The way the whole AI infrastructures, they're chaining all those thoughts together, how they're chaining all the, these things together. I think that's what's making AI smarter for models, they're only incrementally smarter. I, I feel, I would just feel that, oh, this is probably, mm. pretty close to the ceiling. I would never say the ceiling, but it's pretty close. but, you know, three years from now, I would, the, the, the paradoxical things that I, I would predict three years from now, I'm wrong about this, that's just my honest opinion.
Jeff
00:42:57.807
it's still kind of frustrating working. It, it's still magical, but it's still frustrating working with today's AI to code.
Darin
00:43:06.417
Well, I think that's something you just used there. We could have applied to no code 10 years ago.
Darin
00:43:12.807
It's magical. We don't know what's happening, but it's wonderful. I'm not saying AI as an evolution of no code, but to me putting those two things together. 'cause yes, I can sit down and I can vibe code A CLI or a desktop app, but there's a lot of problems with all those things versus if I'm vibe coding something into something that's got guardrails built to the nth degree, I can see good value for that, especially for internal applications.
Viktor
00:43:43.327
here's a challenge for both of you, and this is my reply to this. It's challenging to work with it, which technology? Isn.
Viktor
00:43:53.084
Apart from Coval, okay, in Pearl, apart from No Pearl or no coval, what? What is it? Did we ever have something that is not challenging? Because the, the moment we get something, our criteria changes. if I say, Hey, if I would apply my criteria from 10 years ago, then actually infrastructure is not challenging at all. Today. It's, it seems, it's so simple managing infra today, if I would have the same criteria as 10 years ago. It is challenging because my criteria, my requirements change drastically. And now, uh, the tools that frustrate me frustrate me for very different reasons. Because if I would like to use AWS in the same way as I used my own infrastructure, I dunno, like 10, 20 years ago when I had two servers and, uh, cable between two of them. I mean, there is nothing challenging about AWS yet. I suffer with it every day. Because what I want is very different. And the same thing like goes with business apps. I remember a long time ago when I was a kid, very young, kind of like teenager, I was building apps for businesses. And, uh, all they needed, uh, from today's perspective is an is is, uh, Microsoft Access and 17 fields over there and the button to save stuff. it was very challenging back in the day. It's not challenging today. If I would do the same,
Jeff
00:45:23.314
Yeah, the, the expectations get changed so fast because I remember talking to my family, like it feels like, and when the moment I started doing all those AI assisted coding, every day, my, workday just got longer and a lot more tiring. wasn't AI supposed to help you, like do most of your work so you could, your work is easy, but No, it just got a lot more tiring and my workday got longer and longer.
Viktor
00:45:54.227
imagine that you are doing, now, you mentioned early in this recording that you're now working on using ai, whatever you're using, on some pet projects right now. If you would have the same expectations of those pet projects as when you thought of them first, time, probably kind of like, oh, I, I can do this in two weeks. You probably would find it delightful today.
Viktor
00:46:20.395
expect it to be done in a day, and then you get frustrated and then you work overnight, but because the,
Jeff
00:46:29.515
Exactly. Yeah. Because we start like, I would never start like multiple side projects at the same time. if I were, you know, really coding because yeah, it's just different expectations. And, 10 years ago when you were coding, we all know that software developers tend to underestimate how long the project would take. with today's ai. I think there's still the same symptom that We will totally underestimate how long it will take in the long run. In the beginning, it's, it's so magical. It's faster than you expect. But if you, make the project a little bit bigger than, it's gonna take longer than you think. If you, if you're just making something small and stick with it, then it's super fast. Faster than we think, than you usually, usually think. But kind of estimating how long this project will take is just like 10 years ago. You're gonna underestimate by quite a bit and that's kind of what's making us. work so hard and feel so tired.
Darin
00:47:40.362
It feels like, 'cause Viktor and I have talked about that. It's like I have never been more excited to work on different projects. And I have also, n never, rarely have I ever been so mentally drained at the end of each day.
Jeff
00:47:54.297
I think another thing is that currently AI runs for quite a while, so it's kind of natural for us to do multiple things at the same time. But doing all those context switching for the human mind is taxing most of the time that's what makes me so tired and I try to do less of that. But the, the wait time for AI is just. Naturally begs for doing another task at the same time.
Darin
00:48:21.228
even though you know, you know, you know that it's not gonna be enough time to get anything done. So you're just getting the next thing spinning until it eventually falls off.
Darin
00:48:33.003
if you're talking to somebody that's listening today and it's like, you know, what's the one thing they should think about before evaluating? No code, vibe, coding, or just hand coding? where do you think they should think today? Is it anything different than we've done in the past 50 years?
Jeff
00:48:50.329
at the high level, it's almost exactly the same. Like I've always advised, customers to think about maintenance when their input. Maintain any projects like before, they, always have to think about the maintenance. The maintenance part will be the most expensive. The maintenance will be the one that keeps you awake at night. whether you're doing a project by buying a, a package software, or if you're doing. An AI coding project by creating new applications, the maintenance is the thing that you should, you should be thinking about who will be doing the maintenance. How will you, maintenance this product? Because you know, with ai, the whole code base gets very large, very quickly, As of 2026, it's gonna be a pain to, to maintain when the project gets to a certain size. I think we have to think about how will you maintain this in the beginning when you plan the whole project, and to make sure that, this whole thing is. Scalable and maintainable. I think that will be the the number one concern because making things is no longer an issue. You can make things restart from scratch every time. You can just make things from scratch all the time. But if you wanted to grow into a certain size, is you have to solve the maintenance issue. You have to solve the architectural problem.
Darin
00:50:22.676
Of course, none of us will still be at the company by the time it actually gets to the end, so we don't care.
Darin
00:50:40.676
Oh, that was my. Gonna be my statement there. That's, but I was trying to be nice. So all of Jeff's contact information is gonna be down in the episode description and Raic can be found@raic.com. Again, that is R-A-G-I-C, thinking of it as rapid magic 'cause Jeff told me that raic.com Jeff Thankss, for being with us today.