Debug Information

Processing Details

• VTT File: 75b11be9.vtt
• Processing Time: September 11, 2025 at 01:32 PM
• Total Chunks: 1
• Transcript Length: 70,652 characters
• Caption Count: 596 captions

Prompts Used

Prompt 1: Context Setup

You are an expert data extractor tasked with analyzing a podcast transcript. 
I will provide you with part 1 of 1 from a podcast transcript. 
I will then ask you to extract different types of information from this content in subsequent messages. Please confirm you have received and understood the transcript content.

Transcript section:
[00:00:00.400 --> 00:00:04.560] Hey, it's Arvid, and this is the Bootstrap founder.
[00:00:09.040 --> 00:00:18.240] Today, I'm talking to Nick Groenevelt, a designer and user experience expert who has been a valued collaborator on PodScan, my own software product.
[00:00:18.240 --> 00:00:24.320] Nick works as a freelancer for a lot of companies, and all of them don't really have any in-house design knowledge.
[00:00:24.320 --> 00:00:32.320] So, he's been navigating the rapid changes that AI tools like Lovable and vZero are bringing to the design world for himself and his clients.
[00:00:32.320 --> 00:00:46.000] With 10 years of professional design experience and a deep understanding of design theory, and I can attest to that, he really knows his stuff, he brings a unique perspective on what happens when machines start creating interfaces that look surprisingly good.
[00:00:46.000 --> 00:01:00.160] We also tackle deeper questions about whether AI is making us dumber or why human judgment remains irreplaceable, and how you can use AI as a learning tool rather than just a productivity hack, both in design and beyond that.
[00:01:00.160 --> 00:01:06.160] A big shout out to the sponsor of today's episode, Paddle.com, my merchant of record, payment provider of choice.
[00:01:06.160 --> 00:01:13.840] They're taking care of all the things related to money in businesses so that founders like you and me can focus on building the things that only we can build.
[00:01:13.840 --> 00:01:17.920] Paddle handles all the rest, sales tags, credit cards failing, all of that.
[00:01:17.920 --> 00:01:19.360] They do it so you don't have to.
[00:01:19.360 --> 00:01:20.880] I highly recommend checking it out.
[00:01:20.880 --> 00:01:23.680] So, please go to paddle.com and look into it.
[00:01:23.680 --> 00:01:26.320] Let's dive right into our conversation with Nick.
[00:01:26.320 --> 00:01:27.200] Let's talk design.
[00:01:27.200 --> 00:01:28.320] Let's talk AI.
[00:01:28.320 --> 00:01:29.680] Here we go.
[00:01:31.280 --> 00:01:33.680] Nick, it's really nice to have you on the show again.
[00:01:33.680 --> 00:01:37.520] It's always nice to catch up with you, both on the record and off the record.
[00:01:37.520 --> 00:01:40.080] But today is one of those on-the-record days.
[00:01:40.080 --> 00:01:41.840] Thanks for coming on and thanks for chatting.
[00:01:41.840 --> 00:01:44.240] Oh, well, thanks for having me a second time.
[00:01:44.240 --> 00:01:57.360] Yeah, it's always nice to talk to you because you just give me perspective into the things that I'm working on and may not even be working on yet that I just could not get even from doing research or thinking about stuff.
[00:01:57.360 --> 00:02:01.000] I never get the same quality insight that I get from you.
[00:01:59.840 --> 00:02:01.880] So, thanks for that.
[00:02:02.120 --> 00:02:14.360] I'm really excited to talk to you about one particular thing that I've been running into the last couple of weeks because I've been playing around with stuff like Lovable and vZero, these extremely cool AI generation tools.
[00:02:14.360 --> 00:02:26.520] And one thing I noticed that I hadn't noticed about AI-assisted coding or building before is just how clean the interfaces look that these tools generate.
[00:02:26.520 --> 00:02:31.160] And as somebody who's struggling with white space, let's just call it that.
[00:02:31.160 --> 00:02:35.240] Like, I don't know what the right margin is, the right padding, and all of that.
[00:02:35.240 --> 00:02:44.040] Whenever I have an AI tool generate something for me, it looks really, really clean, stylish in some way, something that I could never accomplish.
[00:02:44.040 --> 00:02:45.960] How do you feel about this kind of stuff?
[00:02:45.960 --> 00:02:53.960] Like, I always felt in our conversations over the last months and years, really, you have this innate intuitive understanding of design.
[00:02:53.960 --> 00:03:00.120] How do you feel when a machine designs and people like me tell you that it looks good?
[00:03:00.120 --> 00:03:02.520] What's your take on this whole development?
[00:03:02.520 --> 00:03:08.280] Well, I have mixed feelings, or maybe a better way to put it is the feelings I have change from day to day.
[00:03:08.280 --> 00:03:16.200] At some point, I feel like I think everyone will feel at some point is, oh, help, it will replace me.
[00:03:16.200 --> 00:03:18.200] You know, that's a feeling I have sometimes.
[00:03:18.200 --> 00:03:25.960] And other times, I feel like I have to up my game, which is a good thing because then I will learn and it will force me to be better.
[00:03:25.960 --> 00:03:31.800] And then the third category of feelings that I have is that it's not that good.
[00:03:31.800 --> 00:03:35.480] It's about to hit the ceiling and we're all safe.
[00:03:35.480 --> 00:03:37.400] So it's those three.
[00:03:37.400 --> 00:03:40.360] And on top of it, there's also some sort of excitement.
[00:03:40.360 --> 00:03:45.600] This morning, I had the first time where I'm like, huh, that upcoming update sounds interesting.
[00:03:45.600 --> 00:03:47.280] I can't wait for it to be here.
[00:03:47.280 --> 00:03:48.160] Okay, what is that?
[00:03:44.920 --> 00:03:48.960] It's about Lovable.
[00:03:49.200 --> 00:03:54.880] They have something coming up in terms of using a design system instead of it being random.
[00:03:54.880 --> 00:03:59.840] Ooh, that's one of the things that I always wondered about because consistency is still a problem, right?
[00:03:59.840 --> 00:04:07.440] Like for any AI tool to be generally consistent in its output over time is always an issue because of the context that it lacks.
[00:04:07.680 --> 00:04:10.560] A designer has this in their mind, but the AI might not.
[00:04:10.560 --> 00:04:13.600] But you're telling me that they're now implementing this into Lovable?
[00:04:13.600 --> 00:04:16.080] Yeah, there have been a few teasers coming up this week.
[00:04:16.080 --> 00:04:19.200] And I think the end of last week, I saw it on LinkedIn and Twitter.
[00:04:19.200 --> 00:04:28.160] They're teasing something coming up around design systems, which is quite interesting because, as you say, it's still a bit random at times.
[00:04:28.160 --> 00:04:34.960] So it's for me personally, it's not ready yet to be used like on a production place, like in a project.
[00:04:34.960 --> 00:04:37.760] But one of the scenarios is that it will be at some point.
[00:04:37.920 --> 00:04:40.720] It's quite likely that it will be good enough at some point.
[00:04:40.720 --> 00:04:47.920] And this feels like it's, if it works well, you know, as well as they tease, then it feels like it's a step into that direction.
[00:04:47.920 --> 00:04:49.120] Yeah, for sure.
[00:04:49.120 --> 00:05:00.480] I mean, from the perspective of somebody who does not feel like I have a design system at all after a year and a half of building this, I know that you have in the things that you have designed.
[00:05:00.480 --> 00:05:04.640] It's just on a technical level, it's never been clearly expressed anywhere, right?
[00:05:04.640 --> 00:05:13.680] It lives in the collective knowledge of the Figma file that exists out there and your brain, but it's never been codified into the system.
[00:05:13.680 --> 00:05:20.880] It would be interesting to think about this almost as a configuration step in AI-assisted coding, right?
[00:05:20.880 --> 00:05:30.760] Because all these tools, like the clawed codes and the cursors and stuff, they have these markdown-based configuration files that are almost like system prompts in a way.
[00:05:29.680 --> 00:05:41.800] Like every single prompt draws them in, so they know, okay, I'm actually building a Laravel project or a Next.js project, and this is how we deal with dependency injection or whatever, these specifics, right?
[00:05:41.800 --> 00:05:51.720] These technical specifics, they're codified in there, but it usually doesn't extend to anything further than, oh, we kind of use Tailwind most of the time for design, right?
[00:05:51.720 --> 00:05:56.040] That's kind of the core maximum of what design is laid out as.
[00:05:56.040 --> 00:05:58.920] So that could change into something much more elaborate?
[00:05:58.920 --> 00:06:01.960] It will be better, I think, is the safe way to put it.
[00:06:01.960 --> 00:06:17.080] Because at the same time, I know from my experience as a designer that let's say you have a super clean design system, and there are always edge cases where a designer will say like detach instance and then they will make something custom and then that's not consistent anymore.
[00:06:17.080 --> 00:06:25.800] Just this morning, I worked on a bit of tailwind code and it has like the H2 and the H3 like for you know to structure your content.
[00:06:25.800 --> 00:06:36.440] And I used my default design system for it, but then I ran into a situation where the H2 followed directly after a styled list where each item was a card.
[00:06:36.440 --> 00:06:43.240] But then I felt with my design eye that the H2 following that final card was way too close to the card above.
[00:06:43.240 --> 00:06:46.440] So I had to add some custom margin there.
[00:06:46.440 --> 00:06:48.840] And sure, that's something that could happen on any page.
[00:06:49.000 --> 00:06:55.640] You can define it as like whenever you follow a card, this should be your margin instead of the default one.
[00:06:55.640 --> 00:07:02.760] But that's just one of hundreds of examples where things might be slightly different compared to the default design system.
[00:07:02.760 --> 00:07:05.560] So that's, I think, a challenge that will remain.
[00:07:05.560 --> 00:07:09.960] And then you also have existing code bases for other projects.
[00:07:09.960 --> 00:07:15.000] I work on a code base plus the design for something that's been around for five plus years.
[00:07:15.520 --> 00:07:25.680] And there are all sorts of quick fixes and inline code and stuff people have done before me that I have no clue about, like why they did something, like why they did A instead of B.
[00:07:26.000 --> 00:07:30.320] It's hard to be consistent there because there are so many little details.
[00:07:30.320 --> 00:07:39.520] And I'm not sure how to make that super simple or how to have an AI that's smart enough to be like a super designer in each of those situations.
[00:07:39.520 --> 00:07:45.520] But I'd love to see them try and get as close as they can because I'm just curious about how it would work.
[00:07:45.520 --> 00:07:50.880] I would like to see that too from the other side of things, from the technical implementation side of things.
[00:07:50.880 --> 00:08:09.520] Because if the people at Lovable or vZero and whatever other tools might be out there that I'm not even aware of, if they come up with a way to codify design as something that is as specific as you were just saying, like it has all these rules, and then there's also the exceptions that are also kind of rules, right?
[00:08:09.520 --> 00:08:15.360] There are meta rules, rules of composition, not necessarily of individual dimension, right?
[00:08:15.360 --> 00:08:18.480] Like if this and this exists, it needs to have white space.
[00:08:18.480 --> 00:08:20.240] That's not about the A or the B.
[00:08:20.240 --> 00:08:22.160] It's about A and B in combination.
[00:08:22.160 --> 00:08:24.320] So it's a rule of rules.
[00:08:24.320 --> 00:08:44.560] And if we can codify this in some way, or even if we can just express it in some way, and then the machine internally codifies it into whatever tailwind code or HTML that comes out of it, then if we can have this, we can also give an agent the opportunity and the task to just look into a whole code base.
[00:08:44.560 --> 00:08:45.360] That's my dream.
[00:08:45.360 --> 00:09:00.440] Like, go through the whole code base, find the inconsistencies, and with a system that we have well-defined that you can actually implement, re-implement this stuff and re-implement it in a way that still passes all the tests and that still is visually pleasing and so on.
[00:09:00.760 --> 00:09:12.520] I think the first agent that does this like 95% well, that's going to have a massive impact on existing code bases and the tools out there in a very visually present way.
[00:09:12.520 --> 00:09:16.200] Like you can clearly observe this because it's going to change a lot of tools.
[00:09:16.200 --> 00:09:17.080] Yeah, I think so too.
[00:09:17.080 --> 00:09:20.680] And you see today that they're all trying the same thing.
[00:09:20.680 --> 00:09:25.880] And then in the end, I think one or two like main players will survive.
[00:09:25.880 --> 00:09:32.520] Yeah, that's kind of already happening in the platforms for AI inference world, like with OpenAI and Anthropic, right?
[00:09:32.520 --> 00:09:36.040] Maybe some DeepSeek and maybe some Gemini, and that's kind of it, right?
[00:09:36.040 --> 00:09:37.880] Mistrol is struggling and so on.
[00:09:37.880 --> 00:09:39.320] But yeah, I think so too.
[00:09:39.320 --> 00:09:41.160] And honestly, Lovable has been really cool.
[00:09:41.160 --> 00:09:47.560] Like I've been building two lovable projects for PodScan over the last couple of weeks, one of which is on the homepage.
[00:09:47.560 --> 00:09:52.600] It's the rotating world that shows wherever new podcast episodes are released.
[00:09:52.680 --> 00:09:54.360] Took me a couple of days to figure this out.
[00:09:54.360 --> 00:10:07.560] And I think I had to upgrade my lovable plan from $20 to $50 a month to be able to do all these little things, these little extra prompts with like, yeah, okay, this is nice, but it should kind of fade in maybe, you know, those things.
[00:10:07.640 --> 00:10:09.720] Took me a while to get that right.
[00:10:09.720 --> 00:10:24.120] But that was really fun to build, like integrating my own API, the PodScan API, the Firehose push API that throws 50,000 data packets right onto it every single day, and seeing these things come to life in the right locations.
[00:10:24.120 --> 00:10:33.880] Like the prompt that I gave it, it was so bizarre how simple and enjoyable this was, was just, I want a world map, and there will be country codes, two leather country codes.
[00:10:33.880 --> 00:10:38.040] Put the image on the right spot on the map.
[00:10:38.040 --> 00:10:39.480] And then it did.
[00:10:39.480 --> 00:10:40.760] How wild is that?
[00:10:40.760 --> 00:10:41.000] Right?
[00:10:41.000 --> 00:10:43.080] Like, that was one of those things where I was like, what?
[00:10:43.080 --> 00:10:44.960] I don't need to keep that list.
[00:10:44.440 --> 00:10:51.520] It's just going to generate a list over 190 countries with each having a central coordinate and it's going to put the image there.
[00:10:44.680 --> 00:10:51.760] Okay.
[00:10:52.400 --> 00:10:54.240] So that was one thing that I built.
[00:10:54.240 --> 00:11:01.440] And that is now running as a kind of landing page visualization of the speed and the intensity of the data that we have.
[00:11:01.440 --> 00:11:07.680] And the other thing that I built was for a prospective customer that I had a conversation with, a demo call.
[00:11:07.680 --> 00:11:09.520] They are in the entertainment industry.
[00:11:09.520 --> 00:11:11.040] They have an analytics company.
[00:11:11.040 --> 00:11:16.400] They just track listens and views and social media presence of artists.
[00:11:16.400 --> 00:11:19.280] And they were wondering, well, podcasts look interesting.
[00:11:19.280 --> 00:11:21.440] Lots of people are talking about artists on podcasts.
[00:11:21.440 --> 00:11:22.960] Might we be interested in that?
[00:11:22.960 --> 00:11:26.560] So we had a whole call, an hour-long demo call.
[00:11:26.560 --> 00:11:46.800] And what I did at the end of it, and this might just be really interesting as a concept for building prototypes for customers, I took the transcript of the call, I threw that into Claude, and I told Claude, write me a prompt for Lovable so that Lovable can build a prototype of the application that this customer already has with our data included.
[00:11:46.800 --> 00:11:49.600] So it went through this hour-long conversation.
[00:11:49.600 --> 00:11:56.480] It turned that into a prompt, which I copied, pasted into Lovable, and had Lovable work on it for like half an hour or so.
[00:11:56.480 --> 00:12:11.760] And what came out of this was kind of a click dummy prototype version of a part of their website with their own branding of the company that I was talking to already baked in that was showing what our API could facilitate on their website.
[00:12:11.760 --> 00:12:19.840] So I just took that link, sent it over the next morning to those people, and have been in contact with them ever since about building something together.
[00:12:19.840 --> 00:12:34.760] And for both of these purposes, I didn't spend any coding effort at all on this, but out came two really, really strong visual things, well designed from my perspective, that communicate very clearly what the business can offer.
[00:12:29.680 --> 00:12:36.280] It's quite magical.
[00:12:36.600 --> 00:12:37.480] It is magical.
[00:12:37.480 --> 00:12:45.800] That's exactly what I wanted to say: like magical, because I feel like old school coding that's really like ones and zeros, and it works or it doesn't.
[00:12:45.800 --> 00:12:47.720] And then if it works, you can optimize.
[00:12:47.720 --> 00:12:50.440] Like it feels much more scientific.
[00:12:50.440 --> 00:12:52.360] What you're describing is magical.
[00:12:52.360 --> 00:12:55.960] It's also what they call a 10-star experience.
[00:12:55.960 --> 00:12:59.080] They think of things before you think of it.
[00:12:59.080 --> 00:13:02.680] And then when they do it well, you're like, huh, that's nice.
[00:13:02.680 --> 00:13:03.960] I didn't even think of it yet.
[00:13:03.960 --> 00:13:05.160] Like, how cool is that they think?
[00:13:05.160 --> 00:13:12.200] Like, you go to a super expensive restaurant and then they remembered that it's your birthday and then they bring you a cake without you asking for it.
[00:13:12.200 --> 00:13:13.080] Like, oh, that's nice.
[00:13:13.080 --> 00:13:15.480] I will give them a big tip at the end.
[00:13:15.480 --> 00:13:18.520] So that's the UX of this whole thing.
[00:13:18.520 --> 00:13:20.600] Like, you feel like, wow, that's impressive.
[00:13:20.600 --> 00:13:22.040] And it's useful too.
[00:13:22.040 --> 00:13:25.880] At the same time, you know, you mentioned like this is design that's good enough for me.
[00:13:25.880 --> 00:13:30.200] And I noticed within myself when I saw the globe spinning, my head was spinning too.
[00:13:30.200 --> 00:13:31.720] First of all, because of how cool it was.
[00:13:31.720 --> 00:13:43.640] And then, secondly, like, oh, we can do so many things here to make it really flow within the whole landing page that I've had in mind for months for PodScan, but haven't been able to work on yet.
[00:13:43.640 --> 00:13:46.280] Like, I really want to take it to the next level.
[00:13:46.280 --> 00:13:54.200] And that's why I think that my job is safe, to give a bit of peace of mind to the people who are afraid that their job will go away.
[00:13:54.200 --> 00:13:58.840] Like the layer on top, that's what I'm eager to build and design now.
[00:13:58.840 --> 00:14:00.600] Yeah, it really is a shift.
[00:14:00.600 --> 00:14:07.640] And if you're noticing that already in what design is, because it's more like at this point to me, it's almost managerial.
[00:14:07.640 --> 00:14:11.160] Like you're managing the systems, you're managing processes.
[00:14:11.160 --> 00:14:20.160] You're not necessarily shifting pixels around anymore, but you're allowing others, humans or machines, to implement things better and more consistently.
[00:14:20.160 --> 00:14:21.840] Same for the tech side of things, right?
[00:14:21.840 --> 00:14:30.480] When I code, I don't know when I've written my last actual line of code, probably two or three days ago, but I have shipped four features since, right?
[00:14:30.640 --> 00:14:31.440] That's the difference.
[00:14:31.440 --> 00:14:42.080] And maybe I wrote a line of code yesterday, but it probably was just a log somewhere in a file for me to understand if the data at that point was right before I yelled at the machine to fix it in a different way, right?
[00:14:42.080 --> 00:14:47.280] Sometimes I just tell it, hey, write more logs and then read the logs and fix it yourself and then remove the logs.
[00:14:47.280 --> 00:14:58.400] I don't need to do it myself anymore at all, but it becomes almost like you're a shepherd of sometimes smart and sometimes very dumb tools, depending on what you tell them to do, really.
[00:14:58.720 --> 00:15:04.480] And I wonder if designing will also change in that way as a whole industry.
[00:15:04.480 --> 00:15:12.320] If it will always have an almost an automated component or if there will still be the human touch the same way that it was, I don't know, 10 years ago.
[00:15:12.320 --> 00:15:18.880] Because here's the thing: I remember wireframing and then implementing prototypes and maybe just drawing stuff on a piece of paper.
[00:15:18.880 --> 00:15:32.720] But if you have AI tools that can do that for you and that can create this 10-star experience, I wanted to say something about this too, because it reminded me that when, for my prospective customer, this app was built, it had three bars on the top.
[00:15:32.720 --> 00:15:37.040] And these were kind of these kind of standout things for you have to act on this now.
[00:15:37.040 --> 00:15:40.880] And in there were things that I personally would have never thought about.
[00:15:40.880 --> 00:15:45.280] It presented me like, here is a spike for this artist that happened recently.
[00:15:45.280 --> 00:15:50.080] Or here is a scatter plot of which artist is currently the most talked about.
[00:15:50.080 --> 00:15:56.080] And if they're going up or down, things that I personally did not think about could be done with the data.
[00:15:56.080 --> 00:16:06.200] But just from having this full conversation and the prompt in there, the lovable tool figured out that this might be useful and actionable data for my customer.
[00:16:06.200 --> 00:16:17.560] And that was like, okay, not only is this tool generating this for me within five minutes, which would have taken me two weeks to build, it also is giving me business insights that I would never have had.
[00:16:17.560 --> 00:16:20.920] Like, again, magical, but in a totally different, unexpected layer.
[00:16:20.920 --> 00:16:23.400] Well, you know, I want to be very realistic.
[00:16:23.400 --> 00:16:26.680] Sometimes people are very protective of their craft and their field.
[00:16:26.680 --> 00:16:32.280] But just my observation is that I do very little UI work for PodScan.
[00:16:32.280 --> 00:16:34.840] At the same time, PodScan is growing.
[00:16:34.840 --> 00:16:36.520] I get your notifications.
[00:16:36.520 --> 00:16:41.720] Like, I'm part of the team, so I can see the insights and how things go up, which is great.
[00:16:41.720 --> 00:16:47.080] Question is, would the lines go up quicker if I did more UI work?
[00:16:47.080 --> 00:16:52.360] Does the UI really matter to that potential client you talk about?
[00:16:52.680 --> 00:16:59.800] If you have like a zero to 100 level, like if you play a game, like you're a level one design, like it's not that good.
[00:16:59.800 --> 00:17:09.080] But if it's between like 40 and 70 out of 100, like that's good enough that most people will be like, yeah, I will sign up for this PodScan thing.
[00:17:09.080 --> 00:17:12.200] Like, does it need to be 95, which I can do?
[00:17:12.200 --> 00:17:13.720] Well, that's an honest question to you.
[00:17:13.720 --> 00:17:14.680] Like, what do you think?
[00:17:14.680 --> 00:17:19.000] Would a nicer looking UI have benefits for PodScan?
[00:17:19.000 --> 00:17:19.560] Yeah.
[00:17:19.560 --> 00:17:24.200] Well, obviously, yes, but is that worth the effort and focus?
[00:17:24.200 --> 00:17:28.760] Particularly with you being a fractional designer that also has other projects.
[00:17:28.760 --> 00:17:32.680] To me, it's always like, is it worth also your time to spend on this?
[00:17:32.680 --> 00:17:37.960] And does that have as much impact for PodScan as your work would have on somebody else at the same time?
[00:17:37.960 --> 00:17:39.000] That's a whole other level.
[00:17:39.000 --> 00:17:42.200] But let me answer the actual question that you asked me here.
[00:17:42.200 --> 00:17:46.640] I think our customers, and currently we're looking at like two main customer groups, right?
[00:17:46.640 --> 00:17:58.720] We're looking at the PR departments, the PR agencies, the marketing departments, marketing agencies, everybody who has a kind of a FOMO when it comes to people and brands and products being mentioned on podcasts somewhere.
[00:17:58.720 --> 00:18:02.400] Those are one user group and they use the UI quite heavily.
[00:18:02.400 --> 00:18:18.720] And then there's the other user group that's founders, entrepreneurial SaaS founders who need that data to pull into their own system because they're building a note-taking app, a podcast listening app, a podcast discovery app, or have part like the customer that I was talking about, have the need for the data inside their own existing infrastructure.
[00:18:18.720 --> 00:18:20.240] Like they don't care about UI at all.
[00:18:20.240 --> 00:18:21.920] Their UI is the API.
[00:18:21.920 --> 00:18:24.240] And that one is mostly well designed, right?
[00:18:24.240 --> 00:18:31.040] It's definitely documented well enough that you can throw it into an AI coding tool and it will implement it well because that's what Lovable did.
[00:18:31.040 --> 00:18:35.920] I literally took the docs for the fire hose, threw it in there, and it implemented it right away.
[00:18:35.920 --> 00:18:47.680] But the PR departments and the agencies that have to work with the interface every day, I get very little feedback on UI stuff for them where they express frustration at all.
[00:18:47.680 --> 00:18:52.720] If they express anything, it's we need this feature because we need this data.
[00:18:52.720 --> 00:19:05.840] So even those companies, I think they're just so used to working with clunky tools that a tool that is not necessarily clunky, but also not like the most beautiful it could be somewhere in the middle is just acceptable to them.
[00:19:05.840 --> 00:19:10.000] They have a horizon of too bad somewhere and good enough.
[00:19:10.000 --> 00:19:12.080] And we're somewhere there in the good enough stage.
[00:19:12.080 --> 00:19:13.040] That's what I'm saying.
[00:19:13.040 --> 00:19:15.840] Obviously, it's not the most consistent tool.
[00:19:15.840 --> 00:19:26.320] Consistency is a problem that I see because now that I have been building things with AI, occasionally things look very different from part of the system to another part of the system.
[00:19:26.320 --> 00:19:35.240] But the complaints that I get have caused me to work on it and actually implement stuff that makes people's usage experience easier.
[00:19:29.680 --> 00:19:36.760] That was mostly in the search interface.
[00:19:37.320 --> 00:19:40.760] It was mostly in search is weird, type ahead search is weird.
[00:19:40.760 --> 00:19:46.360] We want to be able to type a phrase, use and and or and that kind of stuff, then hit enter and get results.
[00:19:46.360 --> 00:19:49.800] And then we want to have one button where we can add all of the results to a list.
[00:19:49.800 --> 00:19:52.600] That was something that I didn't even expect anybody to do.
[00:19:52.600 --> 00:19:55.640] Like lists to me were an afterthought for the whole tool.
[00:19:55.640 --> 00:20:10.600] But it turned out that for a PR agency that has a customer that needs a list of podcasts, well, yeah, they need a list to put the podcasts in and then export that list into their own whatever system so they can present it to their customers or their clients.
[00:20:10.600 --> 00:20:19.800] So it was from these conversations with customers that I was like, okay, they need this and it needs to look good enough to know where to click to get it so they can get the CSV export.
[00:20:19.800 --> 00:20:24.200] So it's still a very functional, almost a very procedural approach.
[00:20:24.200 --> 00:20:26.600] Like they have a procedure that they need to follow.
[00:20:26.600 --> 00:20:29.960] Find thing, add to list, export list, present, get money.
[00:20:29.960 --> 00:20:31.880] That's their job to be done, right?
[00:20:31.880 --> 00:20:37.160] So if I can facilitate this, then what we currently have seems to be good enough.
[00:20:37.160 --> 00:20:41.720] Although, gotta say, like, I switched out the whole search systems a month ago.
[00:20:41.720 --> 00:20:50.120] Like, the search that is now currently running is way better and technically completely differently implemented than the one we had before.
[00:20:50.120 --> 00:20:53.480] And it shows because it just gives you better results.
[00:20:53.480 --> 00:20:55.320] So that's the UX stuff.
[00:20:55.320 --> 00:20:58.040] The data is better, not necessarily the interface.
[00:20:58.360 --> 00:20:59.480] Yeah, well, that's the thing.
[00:20:59.480 --> 00:21:04.200] Like, in some situations, like the UI doesn't matter as much.
[00:21:04.520 --> 00:21:15.280] And because of it, the impact of a certain tool, an AI-powered tool, you know, Lovable and the others, is also much smaller.
[00:21:15.520 --> 00:21:16.960] It doesn't really matter.
[00:21:14.840 --> 00:21:21.520] In other cases, the UI is super important, especially when you have lots of competition.
[00:21:21.840 --> 00:21:25.520] I don't think PodScan has too many competitors at this point.
[00:21:25.520 --> 00:21:27.360] And they look pretty bad as well.
[00:21:27.680 --> 00:21:36.160] All these tools, like the social monitoring tools and the podcast database tools, they were oldish or at least not very modernly designed.
[00:21:36.160 --> 00:21:36.880] I mean, that's the thing.
[00:21:36.880 --> 00:21:45.200] Like, if people have multiple options, at that point, the UI will matter more because then it's a matter of how easy is it to do something?
[00:21:45.200 --> 00:21:48.640] How easy is it to get your first alert set up in PodScan?
[00:21:48.640 --> 00:21:51.120] Like you said, like search is super important.
[00:21:51.120 --> 00:21:53.520] Like, can I find what I'm looking for?
[00:21:53.520 --> 00:21:55.120] If yes, yay, PodScan.
[00:21:55.200 --> 00:21:56.720] If no, this thing sucks.
[00:21:56.720 --> 00:21:58.800] Let's try out competitor A or B.
[00:21:59.120 --> 00:22:04.080] Especially in the beginning when you're on a trial plan, like you have no relationship with your product yet.
[00:22:04.080 --> 00:22:07.040] So it's very easy to just go somewhere else.
[00:22:07.040 --> 00:22:14.960] So there are so many different factors at play that really determine if UI or UX or design as a whole matters or how much it matters.
[00:22:14.960 --> 00:22:17.520] Because I think it always matters a little bit.
[00:22:17.520 --> 00:22:20.800] Yeah, I wonder, do you think that changes with the age of a company?
[00:22:20.800 --> 00:22:35.840] Like for a startup that just started, either Vibe coded or regular coded, whatever that might mean, how important is that kind of focus on consistency and uniformity and beauty of the interface?
[00:22:35.840 --> 00:22:37.040] What have you experienced?
[00:22:37.040 --> 00:22:41.600] Does age of the company and established market presence play into that?
[00:22:41.600 --> 00:22:42.240] I think so.
[00:22:42.240 --> 00:22:48.320] I think that if you've been around for a long time, people have a whole image of you and who you are.
[00:22:48.320 --> 00:22:49.840] Like, that's your branding.
[00:22:49.840 --> 00:22:53.520] Just look at the old Jaguar stuff, or now with Apple and Liquid Glass.
[00:22:53.560 --> 00:23:02.200] Like, there's more discussion between pro and against people of a certain happening, the more branding you have.
[00:23:02.200 --> 00:23:06.840] So, if you have lots of fans, you will also have lots of people who don't like you.
[00:23:06.840 --> 00:23:10.600] So, that's way stronger for an established brand.
[00:23:10.600 --> 00:23:23.960] When you're super new, it's way more important to get off the ground, survive, you know, revenue-wise, and see if you can establish something for yourself, you know, product market fit-wise.
[00:23:23.960 --> 00:23:26.360] And design plays a role there.
[00:23:26.360 --> 00:23:34.760] But I know just like with PodScan, like in the beginning, like when we first worked on PodScan, we really first had to figure out like one or two main user flows.
[00:23:34.760 --> 00:23:42.040] Now, like, if we have limited time, limited budget, if we change one thing or improve one thing design-wise, what should it be?
[00:23:42.040 --> 00:23:45.480] Like, that's been a whole discussion we had well over a year ago, I guess.
[00:23:45.480 --> 00:23:48.440] Because then you want to design a thing that has the biggest impact.
[00:23:48.440 --> 00:23:52.520] And we went with onboarding and setting up your first alert.
[00:23:52.520 --> 00:24:03.880] We wanted that to be as easy and wow moment, uh, as much wow moment as possible, as quickly as possible, because then it's way more likely for a trial user to extend.
[00:24:03.880 --> 00:24:09.320] Like, if they have had no benefit, they're like, Well, trial version is over, I'll go somewhere else.
[00:24:09.320 --> 00:24:14.760] But if they see, like, oh, this is useful, I will stay, I will give Arvid my money.
[00:24:14.760 --> 00:24:16.200] That's super important.
[00:24:16.200 --> 00:24:19.240] So, that's what I would do for a new company.
[00:24:19.240 --> 00:24:29.160] And that requires a lot of talking and brainstorming and having meetings together, which all happens outside of you know, lovable like tools.
[00:24:29.160 --> 00:24:39.320] There are still AI helpers, there, you know, transcribers, and throwing your transcription into a tool to summarize and make up next actions, that kind of stuff, like you said before.
[00:24:39.320 --> 00:24:43.320] Yeah, but that's the human side that will probably stay around.
[00:24:43.320 --> 00:24:44.600] Who knows?
[00:24:44.640 --> 00:24:50.000] Yeah, I do wonder how much of this can be automated away because I'm looking at transcription tools.
[00:24:50.000 --> 00:24:56.560] Obviously, like all I'm thinking about is transcription every single day, but even for a conversation like this, this is going to be transcribed, right?
[00:24:56.560 --> 00:25:03.200] And this episode of my podcast will land on PodScan, be transcribed, and searched for keyword terms.
[00:25:03.200 --> 00:25:07.840] So, if you want anybody to be notified at this point, just say all those phrases, right?
[00:25:07.840 --> 00:25:11.600] Like, everything now becomes observed and measured and tracked.
[00:25:11.600 --> 00:25:16.720] And I wonder, like, how much of this should flow back into the product design that we have?
[00:25:16.720 --> 00:25:30.320] Like, when I think about a feature that I recently built for PodScan internally for the company for sales enablement, is whenever somebody signs up, they go through the onboarding flow and we kind of encourage them to create an alert.
[00:25:30.320 --> 00:25:35.680] And we encourage them, if they don't, to at least try to search to just see the value prop of the business.
[00:25:35.680 --> 00:25:38.000] And then we track each of these things internally, right?
[00:25:38.000 --> 00:25:44.960] When somebody navigates to a podcast page, looks at a show or they look at a specific episode, we track this as an activity in our system.
[00:25:44.960 --> 00:25:59.760] So we can say, within the first 24 hours, this account, after signing up, went through onboarding, they created an alert, or maybe not, got a notification from this alert because something was found, like a mention was found, or they checked out the API documentation, right?
[00:25:59.760 --> 00:26:04.880] We look for intent, and then 24 hours later, we score each of these people's behavior.
[00:26:04.880 --> 00:26:09.520] Plus, we try to extract where they're from, what their name is, what the company is that they're working for.
[00:26:09.520 --> 00:26:15.280] And then we give them a score and how important it is for us to reach out to them and engage with them as customers.
[00:26:15.280 --> 00:26:24.880] There is a lot of value in doing all of this in an automated way: the tracking and the scoring and looking at which of these things is important and which is not.
[00:26:24.880 --> 00:26:35.960] But the one component that needs to be done by people, because otherwise, this would not be a business, it would just be a machine, is then reaching out to that person, having a conversation with them, booking a call for a demo.
[00:26:36.200 --> 00:26:38.120] That needs to be done by humans still.
[00:26:38.120 --> 00:26:39.400] Well, that's very true.
[00:26:39.640 --> 00:26:44.360] It also made me think about different projects I worked in in the past years.
[00:26:44.360 --> 00:26:46.680] Like my career is 10 years now in design.
[00:26:46.680 --> 00:26:52.440] So it's been quite a few projects and there's overlap, but they're hardly ever the same.
[00:26:52.440 --> 00:26:57.160] There's always a different founder with a different desire or goal.
[00:26:57.160 --> 00:26:59.160] They're all in different stages.
[00:26:59.160 --> 00:27:07.640] So I don't see too much automation happening there because I also feel like if we keep talking about onboarding, you have best practices.
[00:27:07.640 --> 00:27:16.840] For example, have that aha moment or that wow moment as quickly as possible because then it's more likely people will reach that point and it's more likely for them to stick around.
[00:27:16.840 --> 00:27:18.040] You know, that's a rule.
[00:27:18.040 --> 00:27:19.720] But how do you make it shorter?
[00:27:19.720 --> 00:27:20.920] That's the thing.
[00:27:20.920 --> 00:27:23.080] In practice, it's different each time.
[00:27:23.080 --> 00:27:29.000] And then there are also exceptions to the rule where you want to delay that aha moment on purpose.
[00:27:29.000 --> 00:27:33.160] It's not always the case to be as quickly as possible.
[00:27:33.160 --> 00:27:36.520] So I think that requires a bit of human interference.
[00:27:36.520 --> 00:27:39.720] There's a human who needs to be thinking about this stuff.
[00:27:39.720 --> 00:27:41.720] That's the coping I tell myself.
[00:27:41.720 --> 00:27:45.160] There needs to be a human for that part because it's not always the same.
[00:27:45.160 --> 00:27:46.840] Yeah, I have another example for this.
[00:27:46.840 --> 00:27:49.160] I think that's going to stress the point here.
[00:27:49.160 --> 00:28:05.400] In a couple of the demo calls that I had over the last weeks, I've shown the alerts feature and how to set up an alert, how to configure it to a lot of people in the PR agency world because that's the customer that we found has the strongest desire to actually pay for this product right now.
[00:28:05.400 --> 00:28:07.640] So, those are the people we'll be reaching out to.
[00:28:07.640 --> 00:28:13.000] And sales calls are still not easy for me because it's just not how I think.
[00:28:13.000 --> 00:28:21.840] I don't think in terms of what a salesperson would say, which is why it is nice to have Russell on board who's helping me learning this and actually doing this as well without my presence.
[00:28:22.000 --> 00:28:28.160] But in those calls, I showed them, you know, here's alerts, and here you can set up the filters for the alerts.
[00:28:28.160 --> 00:28:29.840] And we're like, okay, yeah, filter keywords.
[00:28:29.840 --> 00:28:30.640] Okay, I get this.
[00:28:30.640 --> 00:28:33.360] And then you can select, okay, what language do I want the alert to be?
[00:28:33.360 --> 00:28:35.520] And okay, English-speaking podcast from the US.
[00:28:35.520 --> 00:28:36.320] Okay, sure.
[00:28:36.320 --> 00:28:39.040] And then I showed them the context-aware question filter.
[00:28:39.040 --> 00:28:43.920] Okay, and here you can put in any question you could potentially have that has a yes or no answer.
[00:28:43.920 --> 00:28:46.240] And if it's true, then you get an alert.
[00:28:46.240 --> 00:28:48.400] And if it's not, then it gets dismissed.
[00:28:48.400 --> 00:28:54.800] And at that point, reliably, eyes just fly open, like chins drop, and eyes open up.
[00:28:54.800 --> 00:28:55.760] Wait, wait, wait, wait, wait.
[00:28:55.920 --> 00:29:00.640] You're telling me that I could ask one question of every single podcast out there.
[00:29:00.640 --> 00:29:06.800] And if it's true, then I get an alert, like any question, like to any detail about anything.
[00:29:06.800 --> 00:29:09.280] And then you see them literally thinking.
[00:29:09.280 --> 00:29:22.960] I've had conversations where there was like five people sitting in a just a meeting room somewhere, sterile business meeting room, and you would see them look at each other and like you could see them think about all the potential things you could do.
[00:29:22.960 --> 00:29:36.640] So one of the things that I immediately did was stress that part of the window or the form in which the alert is configured by putting examples there and like showing the variety of questions you can ask.
[00:29:36.640 --> 00:29:45.760] Before it was just a line where you could put a question in and nobody would think much of it, but now it's a whole segment with like, and here you can ask any question you want of the application.
[00:29:45.760 --> 00:29:49.840] And that has reliably created more alerts for more people.
[00:29:49.840 --> 00:29:54.000] Like now, I see people actually setting up alerts because they're like, oh, I really want to use this.
[00:29:54.000 --> 00:30:01.480] But obviously, that's also shot up my AI usage in the background because that is the one feature that needs to run an AI query on all of this.
[00:29:59.680 --> 00:30:07.320] But I'd rather pay a little bit more in AI fees if people use the product and find it worth paying a couple hundred bucks a month for it.
[00:30:07.480 --> 00:30:07.800] Right.
[00:30:07.800 --> 00:30:13.640] So that moment of people just looking at each other in disbelief, you can do this.
[00:30:13.640 --> 00:30:15.480] Nobody else does this.
[00:30:15.480 --> 00:30:16.760] They told me this.
[00:30:16.760 --> 00:30:20.040] This is a wow moment happening in front of my eyes.
[00:30:20.040 --> 00:30:21.720] This I can facilitate, right?
[00:30:21.720 --> 00:30:23.720] This I can push people towards.
[00:30:23.720 --> 00:30:34.680] And I would never have known if I hadn't showed them, if I hadn't had this conversation with them, and then recorded the conversation so I could figure out stuff after the fact and see these things actually happen in real time.
[00:30:34.680 --> 00:30:39.240] That's the pattern recognition that we as human beings are very proficient in.
[00:30:39.560 --> 00:30:42.840] At some point you were like, hey, wait a minute, this thing's happening.
[00:30:42.840 --> 00:30:46.280] And then you have to translate it to a way to show people.
[00:30:46.280 --> 00:30:51.240] And it's also something perhaps to put on a landing page, or maybe you should create a video where you show it.
[00:30:51.240 --> 00:30:57.080] Or maybe it deserves its own page because it's such a massive feature of PodScan.
[00:30:57.480 --> 00:31:01.000] So there are so many different things to potentially do about it.
[00:31:01.000 --> 00:31:03.800] And this is one thing that was super clear, super obvious.
[00:31:03.800 --> 00:31:07.560] And then you also have the more subtle reactions from people.
[00:31:07.560 --> 00:31:10.760] I mean, this one was, it was impossible to miss this one.
[00:31:10.760 --> 00:31:13.720] It was such a, like you said, they were all looking at each other.
[00:31:13.720 --> 00:31:15.320] Like, you know, it's hard to ignore.
[00:31:15.320 --> 00:31:17.560] But there are a few subtle ones as well.
[00:31:17.560 --> 00:31:32.200] I know back in the day when we did user research, you had the host and you had the note taker/slash observer who would look for the subtle cues within the test subject/slash human being in a user research or in a user test.
[00:31:32.200 --> 00:31:33.480] So that's super important.
[00:31:33.480 --> 00:31:40.920] I'm not sure how AI can do it if it's not text-based, unless you point a camera at someone and you can do facial recognition.
[00:31:40.920 --> 00:31:43.080] So there are all ways to assist.
[00:31:43.080 --> 00:33:18.280] They might find it in the voice pattern too you might find like slight agitation in the audio or something i i think ai can help like find just excitement or change in positivity and sentiment throughout a conversation that can also be detected right where they're grumpy in the beginning and they're using a lot of no's and babies and potentially and then at some point it shifts into oh yeah we should and we will and we have to right into other kinds of phrases i bet somebody's working on this i'm thinking about what is it clueli have you seen this have you seen this tool called cluly which apparently it just tracks what you say and what somebody else is saying in a call and then it gives you points at what you should be asking next or what you should be saying next i haven't used that it's been a dream of mine to have a tool like this for conversations like this where maybe i'm running out of steam at some point and i need another question to ask right as an interviewer this is useful as a salesperson it's useful to not forget what the next question should be that you ask them right or they have said so many things here's something you can recontextualize so having this kind of permanent guardian on your site that gives you ideas that gives you insights that tracks what's going on and gives you the next thing to do next thing to say that is the use of ai that i want to see more that's the augmented part right the augmentation it's not replacing anything but it augments what we already do to do it better i can see this for user interviews customer discovery customer service calls customer success calls or sales calls marketing stuff it's kind of nice to have somebody think about what you should be doing next at all times i agree but i also have a main concern here, and that's very philosophical.
[00:33:18.280 --> 00:33:20.520] Like, does it make us dumber?
[00:33:20.520 --> 00:33:22.440] But at the same time, does it matter?
[00:33:22.440 --> 00:33:24.920] Well, let's spend a couple of hours on this question.
[00:33:24.920 --> 00:33:28.200] Because honestly, I think all technology has this potential, right?
[00:33:28.200 --> 00:33:36.280] We've seen just in our lifetime, the ever-present mobile phone with access to the whole world's knowledge.
[00:33:36.280 --> 00:33:43.720] I mean, ask your grandparents if they're still around or your parents about dates, important, relevant historical dates.
[00:33:43.720 --> 00:33:51.960] And they can probably still regurgitate them from back when they were in school because they were just hammered into them and they had to learn them by heart.
[00:33:51.960 --> 00:33:56.280] And it still sticks around because that's the only way that knowledge was accessible.
[00:33:56.280 --> 00:33:57.880] If you had it in your mind, that was great.
[00:33:57.880 --> 00:33:59.400] And if not, you didn't have it, right?
[00:33:59.400 --> 00:34:04.440] You could go to the library, maybe check out a book and write it down, and then you had to look at the piece of paper.
[00:34:04.440 --> 00:34:08.760] 10 years ago, it was Wikipedia on your phone and you had all that data.
[00:34:08.760 --> 00:34:17.080] And now you literally have voice-enabled agents running at the same time that will show you in real time the knowledge that you need to know.
[00:34:17.080 --> 00:34:19.960] So I think the tendency is to outsource this.
[00:34:19.960 --> 00:34:32.440] I think that's called the body replacement thesis or something, that there's a philosophical theory that says that we are outsourcing capacity from our own body into tools, right?
[00:34:32.440 --> 00:34:34.680] The hand is strong, but a hammer is stronger.
[00:34:34.680 --> 00:34:37.720] The teeth are strong, but a saw is much stronger than those.
[00:34:37.720 --> 00:34:43.080] So the brain is strong, but a phone with connected access to the internet is stronger than the brain.
[00:34:43.080 --> 00:34:48.200] So we tend to outsource things if there is a possibility because they're more effective.
[00:34:48.200 --> 00:34:56.520] And I think for AI tools, we will take yet another thing, which is real-time reasoning, and outsource that to another tool.
[00:34:56.520 --> 00:35:06.040] That's probably not the best thing we can do to stay in full control of our cognitive capacity, but it will make things easier, which is also a human tendency, right?
[00:35:06.040 --> 00:35:07.720] To go for the easy way.
[00:35:07.720 --> 00:35:08.120] True.
[00:35:08.120 --> 00:35:20.640] We're animals and we are wired to spend the less amount of energy as possible just to survive because our animal brain doesn't know if the supermarket is open tomorrow or well, if you can find food tomorrow.
[00:35:20.960 --> 00:35:25.360] You know, but I know I know quite a few phone numbers still by heart.
[00:35:25.360 --> 00:35:31.920] But anyway, I'm thinking, you know, design-wise, I know like all the classics.
[00:35:31.920 --> 00:35:35.840] Like, I know gestalt and typography, color theory.
[00:35:35.840 --> 00:35:40.400] I had ethics, psychology, all that kind of stuff in school.
[00:35:40.400 --> 00:35:42.880] Nothing I do design-wise is random.
[00:35:42.880 --> 00:35:46.480] Like, everything on the page is there for a reason.
[00:35:46.480 --> 00:36:02.320] When I look at the new generation, which makes me sound very old, even though I'm only 32, but the new generation of designers who are using lots of vibe designing and that kind of stuff, I see that they know less about it.
[00:36:02.320 --> 00:36:08.320] But they have projects and they sell stuff and they make products, which makes me think: does it matter?
[00:36:08.320 --> 00:36:12.880] Do I need to be as proficient with the gestalt and all this stuff like I am?
[00:36:12.880 --> 00:36:19.200] I think yes, I am more proficient and better than what an AI can generate UI-wise.
[00:36:19.200 --> 00:36:27.200] But you can see other people who are more AI-proficient and less knowledgeable theory-wise, they manage as well.
[00:36:27.200 --> 00:36:28.800] So, that's something I struggle with.
[00:36:28.800 --> 00:36:38.880] Like, I think craft is very important, knowledge, having, you know, being able to stand firm when the internet is out, being able to do stuff still.
[00:36:39.200 --> 00:36:44.880] I think that's very important because you wouldn't go to a restaurant where the chef is like, Well, it's warm.
[00:36:44.880 --> 00:36:47.360] This person wants something that's cool.
[00:36:47.520 --> 00:36:50.800] He wants something that's sweet, but not bitter.
[00:36:50.800 --> 00:36:53.280] Hey, black box thingy, make me a dish.
[00:36:53.360 --> 00:36:59.720] No, you go there because the chef has a certain image, like it's a three-star chef.
[00:36:59.280 --> 00:37:05.080] I think it's very important to have skills, old school skills, still.
[00:37:05.720 --> 00:37:12.760] There's something to be said for expertise embodied in a human that can work without tools as well.
[00:37:13.000 --> 00:37:14.520] That's the thing with the hammer.
[00:37:14.520 --> 00:37:16.680] Somebody still has to guide the hammer.
[00:37:16.680 --> 00:37:21.400] Somebody still has to know where to put the saw to actually cut down the tree.
[00:37:21.640 --> 00:37:27.000] Like every Christmas, I'm painfully reminded of the fact that I'm not very experienced with that.
[00:37:27.000 --> 00:37:37.720] But that expertise and the experience that leads to it is something that I think, like you do, is something that you still need to have to be able to judge the output of these systems.
[00:37:37.720 --> 00:37:48.520] The thing that I see with my whole AI-based coding efforts that I do all the time, every single day, is I still need to be very good at explaining exactly what I want.
[00:37:48.520 --> 00:37:53.160] And if I know that it's hard to build, I need to give it some hints on how to build it.
[00:37:53.160 --> 00:37:57.000] Otherwise, it would probably find another weird way that may or may not work.
[00:37:57.000 --> 00:38:02.280] But if I already kind of know what good looks like, and I think you, you know this from the design side of things.
[00:38:02.280 --> 00:38:04.840] You know, if a design is good, you see it.
[00:38:04.840 --> 00:38:06.680] And if it's not good, you see that.
[00:38:06.680 --> 00:38:08.520] And I think I do the same thing for code.
[00:38:08.520 --> 00:38:14.920] When I look at code that gets created by an AI system, I very quickly can tell you if it's good code or not.
[00:38:14.920 --> 00:38:21.800] I can't necessarily tell you immediately if it's going to work exactly the way I want it to work, but I can tell you if it's going to be shit.
[00:38:21.800 --> 00:38:27.560] So to be able to have this capacity, I still needed to do this for the last 20 years, right?
[00:38:27.560 --> 00:38:34.520] Like to be able to understand what good code looks like, I needed to very painfully build up good code as a capacity.
[00:38:34.520 --> 00:38:37.000] And now I can judge it without having to do it.
[00:38:37.000 --> 00:38:39.720] But I needed to do it to get to the point of being able to judge it.
[00:38:39.720 --> 00:38:40.520] That's true.
[00:38:40.520 --> 00:38:47.840] And for me, it really helps that my father started to teach me design when I was six, you know, so there's almost second nature for me.
[00:38:48.160 --> 00:38:56.480] And also, the same thing applies, you know, just this week where I was working on a Laravel project with Blade and that kind of stuff.
[00:38:56.480 --> 00:38:58.640] Very familiar for you, I assume.
[00:38:58.960 --> 00:39:06.000] I made a design, client agrees, and they asked me to also put the design into a Blade template.
[00:39:06.000 --> 00:39:12.720] And I wanted something like a table of contents to be sticky on scroll, like that it stayed on a page with a few custom things.
[00:39:12.720 --> 00:39:16.320] I did a bit of work, then I asked the AI to do that for me.
[00:39:16.320 --> 00:39:17.280] It looked all fine.
[00:39:17.280 --> 00:39:19.440] I load the page, not sticky.
[00:39:19.440 --> 00:39:21.200] So I asked, like, hey, what's going on?
[00:39:21.200 --> 00:39:27.280] And it turns out I spent, I think, 30 or 60 minutes together with the AI debugging with console logs.
[00:39:27.280 --> 00:39:36.480] And in the end, we ended up doing a JavaScript version of calculating where the sticky thing should be on the page.
[00:39:36.480 --> 00:39:37.840] It was messy.
[00:39:37.840 --> 00:39:39.680] It was terrible code.
[00:39:39.680 --> 00:39:49.280] And it turned out the only problem was that at the top of the page, like the styling was in the head section, and we had the push instead of the section code there.
[00:39:49.280 --> 00:39:54.000] So all the code the AI made wasn't on the actual main layout.
[00:39:54.000 --> 00:39:56.960] So it's literally one word that needed to be changed.
[00:39:56.960 --> 00:40:03.840] And the people I work with, also with tons of experience, they were able to point it out to me, like, oh, it has to be this thing here.
[00:40:03.840 --> 00:40:06.160] And I replaced it and everything worked.
[00:40:06.160 --> 00:40:09.200] You know, so that's that 20-year experience you're talking about.
[00:40:09.200 --> 00:40:10.160] I didn't have it.
[00:40:10.160 --> 00:40:15.920] I explained it as good as I was able to with my semi-professional coding knowledge.
[00:40:15.920 --> 00:40:18.720] And the AI had to run with it, but it wasn't good enough.
[00:40:18.720 --> 00:40:21.200] And because of it, it took way longer than needed.
[00:40:21.200 --> 00:40:23.520] And the quality was way lower than needed.
[00:40:23.520 --> 00:40:25.600] Yeah, so that's an example of what you're saying.
[00:40:26.000 --> 00:40:30.520] I see most people who are using AI primarily to be quicker.
[00:40:29.920 --> 00:40:32.840] I use AI to become better.
[00:40:33.160 --> 00:40:39.080] I learn Laravel, I learn React, good enough for me to ship my design in code.
[00:40:39.080 --> 00:40:40.920] And I become quicker because of it.
[00:40:40.920 --> 00:40:46.520] But my main benefit from AI is that I know much more than I knew a year ago.
[00:40:46.520 --> 00:40:49.000] And I think that's a mindset I don't see too often.
[00:40:49.000 --> 00:40:56.760] Like on social media, at least, most people focus like 99% of their AI usage on doing more with less.
[00:40:56.760 --> 00:40:59.240] And that hurts the quality too.
[00:40:59.240 --> 00:41:04.280] Like if you go fast and break things, quality and security might suffer.
[00:41:04.280 --> 00:41:05.960] Well, I feel very similarly to this.
[00:41:05.960 --> 00:41:18.120] I think like if people exclusively use AI as a generative tool and then use that generated stuff as the product, they're missing the crucial component that is the AI should always be wrapped by a human, right?
[00:41:18.120 --> 00:41:21.720] You start with a human that has an idea that tells the AI what to do.
[00:41:21.720 --> 00:41:23.560] The AI then does the thing.
[00:41:23.560 --> 00:41:32.440] And then instead of just taking that and putting it somewhere and selling it or taking it into production, there still needs to be this kind of observer process of a human at the end.
[00:41:32.440 --> 00:41:40.280] There still needs to be 20% of the work done afterward to make sure that it fits, to make sure that it's actually what you wanted it to be.
[00:41:40.280 --> 00:41:51.240] The push thing that you said, I had the exact same problem two days ago on a very different project, but it was like adding something to a component and it was like, oh yeah, and then this script runs and this script runs, but it didn't.
[00:41:51.240 --> 00:41:53.160] And because again, it missed that one thing.
[00:41:53.160 --> 00:41:55.800] It's apparently not very good at push, you know?
[00:41:55.800 --> 00:41:59.880] And that's something that you can only understand if you know how it should work in the first place.
[00:41:59.880 --> 00:42:01.480] So there needs to be like oversight.
[00:42:01.480 --> 00:42:03.560] That's the managerial part I'm talking about.
[00:42:03.560 --> 00:42:04.760] You're an AI wrangler.
[00:42:04.760 --> 00:42:05.640] You're a shepherd.
[00:42:05.720 --> 00:42:13.560] You wouldn't just let your sheep go wherever they want or your dog, your shepherd dog for that matter, it still needs to be trained, right?
[00:42:13.560 --> 00:42:20.080] Even though it's good at herding sheep, it's still a dog and will still do whatever it wants if you don't train it to do its job well.
[00:42:14.840 --> 00:42:20.880] No, that's very true.
[00:42:21.040 --> 00:42:22.720] And we'll probably get better at that.
[00:42:22.720 --> 00:42:28.000] But having the gatekeeping human at the end is probably a smart thing to do.
[00:42:28.000 --> 00:42:29.440] That sounds about right to me.
[00:42:29.440 --> 00:42:32.480] You know, that is the thing we will have to deal with over the next years.
[00:42:32.480 --> 00:42:47.440] Like next 10 years or so is going to be developing this internal skill and expertise at working with AI systems without having AI systems go rogue or rampage around in our code base or create things that we didn't want them to create.
[00:42:47.440 --> 00:42:50.720] The skill here is going to be judgment, like, is this good?
[00:42:50.720 --> 00:42:51.760] And discernment.
[00:42:51.760 --> 00:42:52.960] Is this what I want?
[00:42:52.960 --> 00:43:06.400] Those are new skills that also I think can't really exist already because they are so bound to the technology on which they need to be developed that all you can do at this point is to keep prompting and keep checking.
[00:43:06.400 --> 00:43:08.400] That's the dynamic that I feel.
[00:43:08.400 --> 00:43:19.360] I just need to keep telling the thing to do stuff and then learn how my scoped argument of what it should be doing is translated or not into functional code.
[00:43:19.360 --> 00:43:21.200] I'm learning a new language here.
[00:43:21.200 --> 00:43:28.400] I recently had a conversation with somebody around the idea of, well, what are programming languages going to look like in the future?
[00:43:28.400 --> 00:43:31.040] And his point was, it's probably going to be English.
[00:43:31.040 --> 00:43:35.120] And my argument was yes, but probably with a different level of precision, right?
[00:43:35.120 --> 00:43:38.160] We're probably not going to say, hey, build me an UI for this.
[00:43:38.320 --> 00:43:42.720] Probably going to say we need, like in the old days, the hamburger menu, right?
[00:43:42.720 --> 00:43:46.000] It's a weird phrase, but at least it describes exactly what you want.
[00:43:46.000 --> 00:43:57.680] And I think we're going to have components that are language-based that AI systems can reliably implement for us at some point, much more than we're talking about actual code and actual programming languages.
[00:43:57.680 --> 00:43:59.360] That might not matter as much anymore.
[00:43:59.360 --> 00:44:05.320] Same for design, design systems might be expressed completely differently, not in pixels, but in vibes.
[00:44:05.480 --> 00:44:09.240] I don't know, vibe is a loaded word, but how does this feel, right?
[00:44:09.400 --> 00:44:11.320] Like, what does this represent?
[00:44:11.640 --> 00:44:16.600] I just know that this is going to be something that will change and we better stay on top of it.
[00:44:16.600 --> 00:44:17.480] Well, that's true.
[00:44:17.480 --> 00:44:23.320] I mean, that's the main thing to do, just to be open and to learn and to listen to people with different opinions too.
[00:44:23.320 --> 00:44:26.840] As a side note, it's like talking to a two-year-old.
[00:44:26.840 --> 00:44:33.560] When I tell my daughter in the evening, she's two, by the way, we are going to the zoo tomorrow.
[00:44:33.560 --> 00:44:37.320] She runs towards the door, grabs her shoes, and she's ready to go.
[00:44:37.320 --> 00:44:40.200] She stops listening at we are going to the zoo.
[00:44:40.200 --> 00:44:42.440] She doesn't listen to tomorrow.
[00:44:42.440 --> 00:44:46.440] And that's also with the coding and the AI-assisted coding that I'm doing.
[00:44:46.440 --> 00:44:48.760] Like, I cannot say I'm using Laravel and Bootstrap.
[00:44:48.760 --> 00:44:51.240] I have to say Laravel Blade and Bootstrap 5.
[00:44:51.240 --> 00:44:53.080] I have to be extremely specific.
[00:44:53.080 --> 00:44:54.040] And that's not enough.
[00:44:54.040 --> 00:45:00.680] I also have to say again, just to be certain, use as much Bootstrap 5 native classes as you can.
[00:45:00.680 --> 00:45:01.880] And I repeat it all the time.
[00:45:01.880 --> 00:45:03.880] Like, it's like talking to a two-year-old.
[00:45:03.880 --> 00:45:09.240] Like, people say it's like talking to an intern, but I think it's more like talking to a two-year-old.
[00:45:09.240 --> 00:45:15.320] That is super interesting because that is probably exactly how old this technology is at this point, too.
[00:45:15.640 --> 00:45:22.760] You know, if you think about Lovable is like, what, four or five months old, and Claude Code is just a bit younger than that.
[00:45:22.760 --> 00:45:26.200] Like Claude Anthropic, like 3.7 was out a year ago.
[00:45:26.200 --> 00:45:33.480] Like all these tools that we now use on this daily basis, they are younger than your child in some way.
[00:45:33.480 --> 00:45:37.640] And it's probably fine to talk to them like one in a development stage.
[00:45:37.640 --> 00:45:38.680] Very interesting point.
[00:45:38.680 --> 00:45:39.160] I agree.
[00:45:39.160 --> 00:45:42.200] When I do code stuff, I repeat myself all the time.
[00:45:42.200 --> 00:45:48.160] I give like explicit things like scan the whole code base for all functionality that has to do with this.
[00:45:48.480 --> 00:45:56.240] Like, just go through the code base into every PHP file and look for this term and make sure that if you change code, you also change it there.
[00:45:56.240 --> 00:45:56.560] Okay?
[00:45:56.880 --> 00:45:57.120] Right?
[00:45:57.120 --> 00:45:59.920] That's kind of how I talk to it at this point.
[00:45:59.920 --> 00:46:03.360] Well, that makes me think about more two-year-old anecdotes.
[00:46:03.360 --> 00:46:05.040] Like, she likes fruit.
[00:46:05.200 --> 00:46:06.080] She loves mangoes.
[00:46:06.080 --> 00:46:09.840] So then we go to the store, like 500 meters down the road.
[00:46:09.840 --> 00:46:18.640] Whenever she gets distracted by a flower or a cat or anything, basically, I keep telling her, Hey, we're going to the store to get mangoes.
[00:46:18.640 --> 00:46:20.080] And then she's like, oh, yeah.
[00:46:20.080 --> 00:46:21.440] And then she keeps walking.
[00:46:21.440 --> 00:46:24.640] Like, I have to, like, she's zigzagging the whole time.
[00:46:24.640 --> 00:46:29.440] And I keep saying the same thing because I know that's the initial goal.
[00:46:29.440 --> 00:46:31.120] It feels like that's exactly the same.
[00:46:31.120 --> 00:46:33.040] But no, no, no, it's Bootstrap 5.
[00:46:33.040 --> 00:46:34.240] Look at this file.
[00:46:34.240 --> 00:46:35.760] Oh, that's funny.
[00:46:36.080 --> 00:46:37.520] I think you cracked it though.
[00:46:37.520 --> 00:46:43.840] And I think all parenting advice for how to talk to young kids is probably how you should be prompting at this point.
[00:46:43.840 --> 00:46:44.160] Yeah.
[00:46:44.480 --> 00:46:45.840] Work with Claude for a week.
[00:46:45.840 --> 00:46:54.560] When your wife is like 36 weeks plus pregnant, you should do a crash course managing Claude just as a baby prep.
[00:46:54.560 --> 00:46:55.920] That is so funny.
[00:46:55.920 --> 00:47:01.840] In some ways, talking to a child of that age is talking to somebody who's just easily distracted.
[00:47:01.840 --> 00:47:04.240] They have a hard time grasping complicated things.
[00:47:04.560 --> 00:47:06.400] They go back to what they know, right?
[00:47:06.400 --> 00:47:16.240] It's pretty much the attention span of an AI that is kind of summoned into being for your request and then vanishes into the void after.
[00:47:16.240 --> 00:47:18.160] Like it's pretty much that.
[00:47:18.160 --> 00:47:22.480] So unsurprising that you have to talk to like a two-year-old man.
[00:47:22.800 --> 00:47:32.040] Just a great opportunity to pull it all together because we're at a stage where we're just like rambling on about the future of AI.
[00:47:32.360 --> 00:47:40.920] But for anybody who wants to see you share your insights about how to talk to two-year-olds in public, where do you want them to go?
[00:47:40.920 --> 00:47:42.440] Where do you want them to follow you?
[00:47:42.440 --> 00:47:47.960] Well, if they want to talk about design, then it's still on Twitter slash X.
[00:47:47.960 --> 00:47:52.520] I'm still using the old handle for my very first website called Toolbox of Design.
[00:47:52.680 --> 00:47:58.040] I also have the askadesigner.io website, which I'm currently redesigning.
[00:47:58.040 --> 00:48:05.560] If you go to the website today, you can see it's in this in-between state where you can see a bit of a red color and a bit of green.
[00:48:05.560 --> 00:48:08.280] And the header is updated, but the images aren't.
[00:48:08.280 --> 00:48:12.280] So you can see me build that thing in public.
[00:48:12.280 --> 00:48:25.720] If you want the design plus AI plus baby slash young children mixture, you know, then the incredible Tyler White and myself, we started a design table podcast based on design having a seat at the table.
[00:48:25.720 --> 00:48:27.400] Like that's where the name comes from.
[00:48:27.400 --> 00:48:29.640] He has a two-year-old as well.
[00:48:29.640 --> 00:48:42.120] So about 50% of our episodes end just like this episode is ending with strange AI slash two-year-old metaphors and overlapping behavior.
[00:48:42.120 --> 00:48:44.840] So that's designtablepodcast.com.
[00:48:44.840 --> 00:48:46.360] Man, thanks so much for being on.
[00:48:46.360 --> 00:48:51.960] I appreciate all your insights and knowledge about how to talk to prompting systems.
[00:48:52.120 --> 00:49:02.120] Honestly, I'm going to dive a bit deeper into this and just try to embody this two-year-old mentality and see if that actually improves the code that gets generated by the tool.
[00:49:02.120 --> 00:49:02.760] Cool, Nick.
[00:49:02.760 --> 00:49:04.280] Thanks so much for being on the show.
[00:49:04.280 --> 00:49:05.080] Really appreciate it.
[00:49:05.080 --> 00:49:05.800] Well, thank you.
[00:49:05.800 --> 00:49:07.480] Thank you for having me again.
[00:49:07.480 --> 00:49:08.680] And that's it for today.
[00:49:08.680 --> 00:49:10.920] Thank you so much for listening to the Bootstrap Founder.
[00:49:10.920 --> 00:49:14.360] You can find me on Twitter at Avid Kahl, A-R-V-I-D, K-A-H-L.
[00:49:14.360 --> 00:49:27.120] And if you want to support me in this show, please share podscan.fm, my SaaS business, with your professional peers and those who you think will benefit from tracking mentions of their brands, their businesses, and names on podcasts out there.
[00:49:27.120 --> 00:49:31.600] PodScan is a near real-time podcast database with a stellar API.
[00:49:31.600 --> 00:49:34.800] We have 32 million podcast episodes in the back now.
[00:49:34.800 --> 00:49:36.640] The database is humongous.
[00:49:36.640 --> 00:49:40.960] Please share the word with those who need to stay on top of the podcast ecosystem.
[00:49:40.960 --> 00:49:42.400] Thank you so much for listening.
[00:49:42.400 --> 00:49:44.880] Have a wonderful day and bye-bye.

Prompt 2: Key Takeaways

Now please extract the key takeaways from the transcript content I provided.

Extract the most important key takeaways from this part of the conversation. Use a single sentence statement (the key takeaway) rather than milquetoast descriptions like "the hosts discuss...". 

Limit the key takeaways to a maximum of 3. The key takeaways should be insightful and knowledge-additive. 

IMPORTANT: Return ONLY valid JSON, no explanations or markdown. Ensure:
- All strings are properly quoted and escaped
- No trailing commas
- All braces and brackets are balanced
Format: {"key_takeaways": ["takeaway 1", "takeaway 2"]}

Prompt 3: Segments

Now identify 2-4 distinct topical segments from this part of the conversation.

For each segment, identify:
- Descriptive title (3-6 words)  
- START timestamp when this topic begins (HH:MM:SS format)
- Double check that the timestamp is accurate - a timestamp will NEVER be greater than the total length of the audio
- Most important Key takeaway from that segment. Key takeaway must be specific and knowledge-additive.
- Brief summary of the discussion

IMPORTANT: The timestamp should mark when the topic/segment STARTS, not a range. Look for topic transitions and conversation shifts.

Return ONLY valid JSON. Ensure all strings are properly quoted, no trailing commas:
{
  "segments": [
    {
      "segment_title": "Topic Discussion", 
      "timestamp": "01:15:30",
      "key_takeaway": "main point from this segment",
      "segment_summary": "brief description of what was discussed"
    }
  ]
}

Timestamp format: HH:MM:SS (e.g., 00:05:30, 01:22:45) marking the START of each segment.

Prompt 4: Media Mentions

Now scan the transcript content I provided for ACTUAL mentions of specific media titles:

Find explicit mentions of:
- Books (with specific titles)
- Movies (with specific titles)  
- TV Shows (with specific titles)
- Music/Songs (with specific titles)

DO NOT include:
- Websites, URLs, or web services
- Other podcasts or podcast names

IMPORTANT: 
- Only include items explicitly mentioned by name. Do not invent titles.
- Valid categories are: "Book", "Movie", "TV Show", "Music"
- Include the exact phrase where each item was mentioned
- Find the nearest proximate timestamp where it appears in the conversation
- THE TIMESTAMP OF THE MEDIA MENTION IS IMPORTANT - DO NOT INVENT TIMESTAMPS AND DO NOT MISATTRIBUTE TIMESTAMPS
- Double check that the timestamp is accurate - a timestamp will NEVER be greater than the total length of the audio
- Timestamps are given as ranges, e.g. 01:13:42.520 --> 01:13:46.720. Use the EARLIER of the 2 timestamps in the range.

Return ONLY valid JSON. Ensure all strings are properly quoted and escaped, no trailing commas:
{
  "media_mentions": [
    {
      "title": "Exact Title as Mentioned",
      "category": "Book",
      "author_artist": "N/A",
      "context": "Brief context of why it was mentioned",
      "context_phrase": "The exact sentence or phrase where it was mentioned",
      "timestamp": "estimated time like 01:15:30"
    }
  ]
}

If no media is mentioned, return: {"media_mentions": []}

Full Transcript

[00:00:00.400 --> 00:00:04.560] Hey, it's Arvid, and this is the Bootstrap founder.
[00:00:09.040 --> 00:00:18.240] Today, I'm talking to Nick Groenevelt, a designer and user experience expert who has been a valued collaborator on PodScan, my own software product.
[00:00:18.240 --> 00:00:24.320] Nick works as a freelancer for a lot of companies, and all of them don't really have any in-house design knowledge.
[00:00:24.320 --> 00:00:32.320] So, he's been navigating the rapid changes that AI tools like Lovable and vZero are bringing to the design world for himself and his clients.
[00:00:32.320 --> 00:00:46.000] With 10 years of professional design experience and a deep understanding of design theory, and I can attest to that, he really knows his stuff, he brings a unique perspective on what happens when machines start creating interfaces that look surprisingly good.
[00:00:46.000 --> 00:01:00.160] We also tackle deeper questions about whether AI is making us dumber or why human judgment remains irreplaceable, and how you can use AI as a learning tool rather than just a productivity hack, both in design and beyond that.
[00:01:00.160 --> 00:01:06.160] A big shout out to the sponsor of today's episode, Paddle.com, my merchant of record, payment provider of choice.
[00:01:06.160 --> 00:01:13.840] They're taking care of all the things related to money in businesses so that founders like you and me can focus on building the things that only we can build.
[00:01:13.840 --> 00:01:17.920] Paddle handles all the rest, sales tags, credit cards failing, all of that.
[00:01:17.920 --> 00:01:19.360] They do it so you don't have to.
[00:01:19.360 --> 00:01:20.880] I highly recommend checking it out.
[00:01:20.880 --> 00:01:23.680] So, please go to paddle.com and look into it.
[00:01:23.680 --> 00:01:26.320] Let's dive right into our conversation with Nick.
[00:01:26.320 --> 00:01:27.200] Let's talk design.
[00:01:27.200 --> 00:01:28.320] Let's talk AI.
[00:01:28.320 --> 00:01:29.680] Here we go.
[00:01:31.280 --> 00:01:33.680] Nick, it's really nice to have you on the show again.
[00:01:33.680 --> 00:01:37.520] It's always nice to catch up with you, both on the record and off the record.
[00:01:37.520 --> 00:01:40.080] But today is one of those on-the-record days.
[00:01:40.080 --> 00:01:41.840] Thanks for coming on and thanks for chatting.
[00:01:41.840 --> 00:01:44.240] Oh, well, thanks for having me a second time.
[00:01:44.240 --> 00:01:57.360] Yeah, it's always nice to talk to you because you just give me perspective into the things that I'm working on and may not even be working on yet that I just could not get even from doing research or thinking about stuff.
[00:01:57.360 --> 00:02:01.000] I never get the same quality insight that I get from you.
[00:01:59.840 --> 00:02:01.880] So, thanks for that.
[00:02:02.120 --> 00:02:14.360] I'm really excited to talk to you about one particular thing that I've been running into the last couple of weeks because I've been playing around with stuff like Lovable and vZero, these extremely cool AI generation tools.
[00:02:14.360 --> 00:02:26.520] And one thing I noticed that I hadn't noticed about AI-assisted coding or building before is just how clean the interfaces look that these tools generate.
[00:02:26.520 --> 00:02:31.160] And as somebody who's struggling with white space, let's just call it that.
[00:02:31.160 --> 00:02:35.240] Like, I don't know what the right margin is, the right padding, and all of that.
[00:02:35.240 --> 00:02:44.040] Whenever I have an AI tool generate something for me, it looks really, really clean, stylish in some way, something that I could never accomplish.
[00:02:44.040 --> 00:02:45.960] How do you feel about this kind of stuff?
[00:02:45.960 --> 00:02:53.960] Like, I always felt in our conversations over the last months and years, really, you have this innate intuitive understanding of design.
[00:02:53.960 --> 00:03:00.120] How do you feel when a machine designs and people like me tell you that it looks good?
[00:03:00.120 --> 00:03:02.520] What's your take on this whole development?
[00:03:02.520 --> 00:03:08.280] Well, I have mixed feelings, or maybe a better way to put it is the feelings I have change from day to day.
[00:03:08.280 --> 00:03:16.200] At some point, I feel like I think everyone will feel at some point is, oh, help, it will replace me.
[00:03:16.200 --> 00:03:18.200] You know, that's a feeling I have sometimes.
[00:03:18.200 --> 00:03:25.960] And other times, I feel like I have to up my game, which is a good thing because then I will learn and it will force me to be better.
[00:03:25.960 --> 00:03:31.800] And then the third category of feelings that I have is that it's not that good.
[00:03:31.800 --> 00:03:35.480] It's about to hit the ceiling and we're all safe.
[00:03:35.480 --> 00:03:37.400] So it's those three.
[00:03:37.400 --> 00:03:40.360] And on top of it, there's also some sort of excitement.
[00:03:40.360 --> 00:03:45.600] This morning, I had the first time where I'm like, huh, that upcoming update sounds interesting.
[00:03:45.600 --> 00:03:47.280] I can't wait for it to be here.
[00:03:47.280 --> 00:03:48.160] Okay, what is that?
[00:03:44.920 --> 00:03:48.960] It's about Lovable.
[00:03:49.200 --> 00:03:54.880] They have something coming up in terms of using a design system instead of it being random.
[00:03:54.880 --> 00:03:59.840] Ooh, that's one of the things that I always wondered about because consistency is still a problem, right?
[00:03:59.840 --> 00:04:07.440] Like for any AI tool to be generally consistent in its output over time is always an issue because of the context that it lacks.
[00:04:07.680 --> 00:04:10.560] A designer has this in their mind, but the AI might not.
[00:04:10.560 --> 00:04:13.600] But you're telling me that they're now implementing this into Lovable?
[00:04:13.600 --> 00:04:16.080] Yeah, there have been a few teasers coming up this week.
[00:04:16.080 --> 00:04:19.200] And I think the end of last week, I saw it on LinkedIn and Twitter.
[00:04:19.200 --> 00:04:28.160] They're teasing something coming up around design systems, which is quite interesting because, as you say, it's still a bit random at times.
[00:04:28.160 --> 00:04:34.960] So it's for me personally, it's not ready yet to be used like on a production place, like in a project.
[00:04:34.960 --> 00:04:37.760] But one of the scenarios is that it will be at some point.
[00:04:37.920 --> 00:04:40.720] It's quite likely that it will be good enough at some point.
[00:04:40.720 --> 00:04:47.920] And this feels like it's, if it works well, you know, as well as they tease, then it feels like it's a step into that direction.
[00:04:47.920 --> 00:04:49.120] Yeah, for sure.
[00:04:49.120 --> 00:05:00.480] I mean, from the perspective of somebody who does not feel like I have a design system at all after a year and a half of building this, I know that you have in the things that you have designed.
[00:05:00.480 --> 00:05:04.640] It's just on a technical level, it's never been clearly expressed anywhere, right?
[00:05:04.640 --> 00:05:13.680] It lives in the collective knowledge of the Figma file that exists out there and your brain, but it's never been codified into the system.
[00:05:13.680 --> 00:05:20.880] It would be interesting to think about this almost as a configuration step in AI-assisted coding, right?
[00:05:20.880 --> 00:05:30.760] Because all these tools, like the clawed codes and the cursors and stuff, they have these markdown-based configuration files that are almost like system prompts in a way.
[00:05:29.680 --> 00:05:41.800] Like every single prompt draws them in, so they know, okay, I'm actually building a Laravel project or a Next.js project, and this is how we deal with dependency injection or whatever, these specifics, right?
[00:05:41.800 --> 00:05:51.720] These technical specifics, they're codified in there, but it usually doesn't extend to anything further than, oh, we kind of use Tailwind most of the time for design, right?
[00:05:51.720 --> 00:05:56.040] That's kind of the core maximum of what design is laid out as.
[00:05:56.040 --> 00:05:58.920] So that could change into something much more elaborate?
[00:05:58.920 --> 00:06:01.960] It will be better, I think, is the safe way to put it.
[00:06:01.960 --> 00:06:17.080] Because at the same time, I know from my experience as a designer that let's say you have a super clean design system, and there are always edge cases where a designer will say like detach instance and then they will make something custom and then that's not consistent anymore.
[00:06:17.080 --> 00:06:25.800] Just this morning, I worked on a bit of tailwind code and it has like the H2 and the H3 like for you know to structure your content.
[00:06:25.800 --> 00:06:36.440] And I used my default design system for it, but then I ran into a situation where the H2 followed directly after a styled list where each item was a card.
[00:06:36.440 --> 00:06:43.240] But then I felt with my design eye that the H2 following that final card was way too close to the card above.
[00:06:43.240 --> 00:06:46.440] So I had to add some custom margin there.
[00:06:46.440 --> 00:06:48.840] And sure, that's something that could happen on any page.
[00:06:49.000 --> 00:06:55.640] You can define it as like whenever you follow a card, this should be your margin instead of the default one.
[00:06:55.640 --> 00:07:02.760] But that's just one of hundreds of examples where things might be slightly different compared to the default design system.
[00:07:02.760 --> 00:07:05.560] So that's, I think, a challenge that will remain.
[00:07:05.560 --> 00:07:09.960] And then you also have existing code bases for other projects.
[00:07:09.960 --> 00:07:15.000] I work on a code base plus the design for something that's been around for five plus years.
[00:07:15.520 --> 00:07:25.680] And there are all sorts of quick fixes and inline code and stuff people have done before me that I have no clue about, like why they did something, like why they did A instead of B.
[00:07:26.000 --> 00:07:30.320] It's hard to be consistent there because there are so many little details.
[00:07:30.320 --> 00:07:39.520] And I'm not sure how to make that super simple or how to have an AI that's smart enough to be like a super designer in each of those situations.
[00:07:39.520 --> 00:07:45.520] But I'd love to see them try and get as close as they can because I'm just curious about how it would work.
[00:07:45.520 --> 00:07:50.880] I would like to see that too from the other side of things, from the technical implementation side of things.
[00:07:50.880 --> 00:08:09.520] Because if the people at Lovable or vZero and whatever other tools might be out there that I'm not even aware of, if they come up with a way to codify design as something that is as specific as you were just saying, like it has all these rules, and then there's also the exceptions that are also kind of rules, right?
[00:08:09.520 --> 00:08:15.360] There are meta rules, rules of composition, not necessarily of individual dimension, right?
[00:08:15.360 --> 00:08:18.480] Like if this and this exists, it needs to have white space.
[00:08:18.480 --> 00:08:20.240] That's not about the A or the B.
[00:08:20.240 --> 00:08:22.160] It's about A and B in combination.
[00:08:22.160 --> 00:08:24.320] So it's a rule of rules.
[00:08:24.320 --> 00:08:44.560] And if we can codify this in some way, or even if we can just express it in some way, and then the machine internally codifies it into whatever tailwind code or HTML that comes out of it, then if we can have this, we can also give an agent the opportunity and the task to just look into a whole code base.
[00:08:44.560 --> 00:08:45.360] That's my dream.
[00:08:45.360 --> 00:09:00.440] Like, go through the whole code base, find the inconsistencies, and with a system that we have well-defined that you can actually implement, re-implement this stuff and re-implement it in a way that still passes all the tests and that still is visually pleasing and so on.
[00:09:00.760 --> 00:09:12.520] I think the first agent that does this like 95% well, that's going to have a massive impact on existing code bases and the tools out there in a very visually present way.
[00:09:12.520 --> 00:09:16.200] Like you can clearly observe this because it's going to change a lot of tools.
[00:09:16.200 --> 00:09:17.080] Yeah, I think so too.
[00:09:17.080 --> 00:09:20.680] And you see today that they're all trying the same thing.
[00:09:20.680 --> 00:09:25.880] And then in the end, I think one or two like main players will survive.
[00:09:25.880 --> 00:09:32.520] Yeah, that's kind of already happening in the platforms for AI inference world, like with OpenAI and Anthropic, right?
[00:09:32.520 --> 00:09:36.040] Maybe some DeepSeek and maybe some Gemini, and that's kind of it, right?
[00:09:36.040 --> 00:09:37.880] Mistrol is struggling and so on.
[00:09:37.880 --> 00:09:39.320] But yeah, I think so too.
[00:09:39.320 --> 00:09:41.160] And honestly, Lovable has been really cool.
[00:09:41.160 --> 00:09:47.560] Like I've been building two lovable projects for PodScan over the last couple of weeks, one of which is on the homepage.
[00:09:47.560 --> 00:09:52.600] It's the rotating world that shows wherever new podcast episodes are released.
[00:09:52.680 --> 00:09:54.360] Took me a couple of days to figure this out.
[00:09:54.360 --> 00:10:07.560] And I think I had to upgrade my lovable plan from $20 to $50 a month to be able to do all these little things, these little extra prompts with like, yeah, okay, this is nice, but it should kind of fade in maybe, you know, those things.
[00:10:07.640 --> 00:10:09.720] Took me a while to get that right.
[00:10:09.720 --> 00:10:24.120] But that was really fun to build, like integrating my own API, the PodScan API, the Firehose push API that throws 50,000 data packets right onto it every single day, and seeing these things come to life in the right locations.
[00:10:24.120 --> 00:10:33.880] Like the prompt that I gave it, it was so bizarre how simple and enjoyable this was, was just, I want a world map, and there will be country codes, two leather country codes.
[00:10:33.880 --> 00:10:38.040] Put the image on the right spot on the map.
[00:10:38.040 --> 00:10:39.480] And then it did.
[00:10:39.480 --> 00:10:40.760] How wild is that?
[00:10:40.760 --> 00:10:41.000] Right?
[00:10:41.000 --> 00:10:43.080] Like, that was one of those things where I was like, what?
[00:10:43.080 --> 00:10:44.960] I don't need to keep that list.
[00:10:44.440 --> 00:10:51.520] It's just going to generate a list over 190 countries with each having a central coordinate and it's going to put the image there.
[00:10:44.680 --> 00:10:51.760] Okay.
[00:10:52.400 --> 00:10:54.240] So that was one thing that I built.
[00:10:54.240 --> 00:11:01.440] And that is now running as a kind of landing page visualization of the speed and the intensity of the data that we have.
[00:11:01.440 --> 00:11:07.680] And the other thing that I built was for a prospective customer that I had a conversation with, a demo call.
[00:11:07.680 --> 00:11:09.520] They are in the entertainment industry.
[00:11:09.520 --> 00:11:11.040] They have an analytics company.
[00:11:11.040 --> 00:11:16.400] They just track listens and views and social media presence of artists.
[00:11:16.400 --> 00:11:19.280] And they were wondering, well, podcasts look interesting.
[00:11:19.280 --> 00:11:21.440] Lots of people are talking about artists on podcasts.
[00:11:21.440 --> 00:11:22.960] Might we be interested in that?
[00:11:22.960 --> 00:11:26.560] So we had a whole call, an hour-long demo call.
[00:11:26.560 --> 00:11:46.800] And what I did at the end of it, and this might just be really interesting as a concept for building prototypes for customers, I took the transcript of the call, I threw that into Claude, and I told Claude, write me a prompt for Lovable so that Lovable can build a prototype of the application that this customer already has with our data included.
[00:11:46.800 --> 00:11:49.600] So it went through this hour-long conversation.
[00:11:49.600 --> 00:11:56.480] It turned that into a prompt, which I copied, pasted into Lovable, and had Lovable work on it for like half an hour or so.
[00:11:56.480 --> 00:12:11.760] And what came out of this was kind of a click dummy prototype version of a part of their website with their own branding of the company that I was talking to already baked in that was showing what our API could facilitate on their website.
[00:12:11.760 --> 00:12:19.840] So I just took that link, sent it over the next morning to those people, and have been in contact with them ever since about building something together.
[00:12:19.840 --> 00:12:34.760] And for both of these purposes, I didn't spend any coding effort at all on this, but out came two really, really strong visual things, well designed from my perspective, that communicate very clearly what the business can offer.
[00:12:29.680 --> 00:12:36.280] It's quite magical.
[00:12:36.600 --> 00:12:37.480] It is magical.
[00:12:37.480 --> 00:12:45.800] That's exactly what I wanted to say: like magical, because I feel like old school coding that's really like ones and zeros, and it works or it doesn't.
[00:12:45.800 --> 00:12:47.720] And then if it works, you can optimize.
[00:12:47.720 --> 00:12:50.440] Like it feels much more scientific.
[00:12:50.440 --> 00:12:52.360] What you're describing is magical.
[00:12:52.360 --> 00:12:55.960] It's also what they call a 10-star experience.
[00:12:55.960 --> 00:12:59.080] They think of things before you think of it.
[00:12:59.080 --> 00:13:02.680] And then when they do it well, you're like, huh, that's nice.
[00:13:02.680 --> 00:13:03.960] I didn't even think of it yet.
[00:13:03.960 --> 00:13:05.160] Like, how cool is that they think?
[00:13:05.160 --> 00:13:12.200] Like, you go to a super expensive restaurant and then they remembered that it's your birthday and then they bring you a cake without you asking for it.
[00:13:12.200 --> 00:13:13.080] Like, oh, that's nice.
[00:13:13.080 --> 00:13:15.480] I will give them a big tip at the end.
[00:13:15.480 --> 00:13:18.520] So that's the UX of this whole thing.
[00:13:18.520 --> 00:13:20.600] Like, you feel like, wow, that's impressive.
[00:13:20.600 --> 00:13:22.040] And it's useful too.
[00:13:22.040 --> 00:13:25.880] At the same time, you know, you mentioned like this is design that's good enough for me.
[00:13:25.880 --> 00:13:30.200] And I noticed within myself when I saw the globe spinning, my head was spinning too.
[00:13:30.200 --> 00:13:31.720] First of all, because of how cool it was.
[00:13:31.720 --> 00:13:43.640] And then, secondly, like, oh, we can do so many things here to make it really flow within the whole landing page that I've had in mind for months for PodScan, but haven't been able to work on yet.
[00:13:43.640 --> 00:13:46.280] Like, I really want to take it to the next level.
[00:13:46.280 --> 00:13:54.200] And that's why I think that my job is safe, to give a bit of peace of mind to the people who are afraid that their job will go away.
[00:13:54.200 --> 00:13:58.840] Like the layer on top, that's what I'm eager to build and design now.
[00:13:58.840 --> 00:14:00.600] Yeah, it really is a shift.
[00:14:00.600 --> 00:14:07.640] And if you're noticing that already in what design is, because it's more like at this point to me, it's almost managerial.
[00:14:07.640 --> 00:14:11.160] Like you're managing the systems, you're managing processes.
[00:14:11.160 --> 00:14:20.160] You're not necessarily shifting pixels around anymore, but you're allowing others, humans or machines, to implement things better and more consistently.
[00:14:20.160 --> 00:14:21.840] Same for the tech side of things, right?
[00:14:21.840 --> 00:14:30.480] When I code, I don't know when I've written my last actual line of code, probably two or three days ago, but I have shipped four features since, right?
[00:14:30.640 --> 00:14:31.440] That's the difference.
[00:14:31.440 --> 00:14:42.080] And maybe I wrote a line of code yesterday, but it probably was just a log somewhere in a file for me to understand if the data at that point was right before I yelled at the machine to fix it in a different way, right?
[00:14:42.080 --> 00:14:47.280] Sometimes I just tell it, hey, write more logs and then read the logs and fix it yourself and then remove the logs.
[00:14:47.280 --> 00:14:58.400] I don't need to do it myself anymore at all, but it becomes almost like you're a shepherd of sometimes smart and sometimes very dumb tools, depending on what you tell them to do, really.
[00:14:58.720 --> 00:15:04.480] And I wonder if designing will also change in that way as a whole industry.
[00:15:04.480 --> 00:15:12.320] If it will always have an almost an automated component or if there will still be the human touch the same way that it was, I don't know, 10 years ago.
[00:15:12.320 --> 00:15:18.880] Because here's the thing: I remember wireframing and then implementing prototypes and maybe just drawing stuff on a piece of paper.
[00:15:18.880 --> 00:15:32.720] But if you have AI tools that can do that for you and that can create this 10-star experience, I wanted to say something about this too, because it reminded me that when, for my prospective customer, this app was built, it had three bars on the top.
[00:15:32.720 --> 00:15:37.040] And these were kind of these kind of standout things for you have to act on this now.
[00:15:37.040 --> 00:15:40.880] And in there were things that I personally would have never thought about.
[00:15:40.880 --> 00:15:45.280] It presented me like, here is a spike for this artist that happened recently.
[00:15:45.280 --> 00:15:50.080] Or here is a scatter plot of which artist is currently the most talked about.
[00:15:50.080 --> 00:15:56.080] And if they're going up or down, things that I personally did not think about could be done with the data.
[00:15:56.080 --> 00:16:06.200] But just from having this full conversation and the prompt in there, the lovable tool figured out that this might be useful and actionable data for my customer.
[00:16:06.200 --> 00:16:17.560] And that was like, okay, not only is this tool generating this for me within five minutes, which would have taken me two weeks to build, it also is giving me business insights that I would never have had.
[00:16:17.560 --> 00:16:20.920] Like, again, magical, but in a totally different, unexpected layer.
[00:16:20.920 --> 00:16:23.400] Well, you know, I want to be very realistic.
[00:16:23.400 --> 00:16:26.680] Sometimes people are very protective of their craft and their field.
[00:16:26.680 --> 00:16:32.280] But just my observation is that I do very little UI work for PodScan.
[00:16:32.280 --> 00:16:34.840] At the same time, PodScan is growing.
[00:16:34.840 --> 00:16:36.520] I get your notifications.
[00:16:36.520 --> 00:16:41.720] Like, I'm part of the team, so I can see the insights and how things go up, which is great.
[00:16:41.720 --> 00:16:47.080] Question is, would the lines go up quicker if I did more UI work?
[00:16:47.080 --> 00:16:52.360] Does the UI really matter to that potential client you talk about?
[00:16:52.680 --> 00:16:59.800] If you have like a zero to 100 level, like if you play a game, like you're a level one design, like it's not that good.
[00:16:59.800 --> 00:17:09.080] But if it's between like 40 and 70 out of 100, like that's good enough that most people will be like, yeah, I will sign up for this PodScan thing.
[00:17:09.080 --> 00:17:12.200] Like, does it need to be 95, which I can do?
[00:17:12.200 --> 00:17:13.720] Well, that's an honest question to you.
[00:17:13.720 --> 00:17:14.680] Like, what do you think?
[00:17:14.680 --> 00:17:19.000] Would a nicer looking UI have benefits for PodScan?
[00:17:19.000 --> 00:17:19.560] Yeah.
[00:17:19.560 --> 00:17:24.200] Well, obviously, yes, but is that worth the effort and focus?
[00:17:24.200 --> 00:17:28.760] Particularly with you being a fractional designer that also has other projects.
[00:17:28.760 --> 00:17:32.680] To me, it's always like, is it worth also your time to spend on this?
[00:17:32.680 --> 00:17:37.960] And does that have as much impact for PodScan as your work would have on somebody else at the same time?
[00:17:37.960 --> 00:17:39.000] That's a whole other level.
[00:17:39.000 --> 00:17:42.200] But let me answer the actual question that you asked me here.
[00:17:42.200 --> 00:17:46.640] I think our customers, and currently we're looking at like two main customer groups, right?
[00:17:46.640 --> 00:17:58.720] We're looking at the PR departments, the PR agencies, the marketing departments, marketing agencies, everybody who has a kind of a FOMO when it comes to people and brands and products being mentioned on podcasts somewhere.
[00:17:58.720 --> 00:18:02.400] Those are one user group and they use the UI quite heavily.
[00:18:02.400 --> 00:18:18.720] And then there's the other user group that's founders, entrepreneurial SaaS founders who need that data to pull into their own system because they're building a note-taking app, a podcast listening app, a podcast discovery app, or have part like the customer that I was talking about, have the need for the data inside their own existing infrastructure.
[00:18:18.720 --> 00:18:20.240] Like they don't care about UI at all.
[00:18:20.240 --> 00:18:21.920] Their UI is the API.
[00:18:21.920 --> 00:18:24.240] And that one is mostly well designed, right?
[00:18:24.240 --> 00:18:31.040] It's definitely documented well enough that you can throw it into an AI coding tool and it will implement it well because that's what Lovable did.
[00:18:31.040 --> 00:18:35.920] I literally took the docs for the fire hose, threw it in there, and it implemented it right away.
[00:18:35.920 --> 00:18:47.680] But the PR departments and the agencies that have to work with the interface every day, I get very little feedback on UI stuff for them where they express frustration at all.
[00:18:47.680 --> 00:18:52.720] If they express anything, it's we need this feature because we need this data.
[00:18:52.720 --> 00:19:05.840] So even those companies, I think they're just so used to working with clunky tools that a tool that is not necessarily clunky, but also not like the most beautiful it could be somewhere in the middle is just acceptable to them.
[00:19:05.840 --> 00:19:10.000] They have a horizon of too bad somewhere and good enough.
[00:19:10.000 --> 00:19:12.080] And we're somewhere there in the good enough stage.
[00:19:12.080 --> 00:19:13.040] That's what I'm saying.
[00:19:13.040 --> 00:19:15.840] Obviously, it's not the most consistent tool.
[00:19:15.840 --> 00:19:26.320] Consistency is a problem that I see because now that I have been building things with AI, occasionally things look very different from part of the system to another part of the system.
[00:19:26.320 --> 00:19:35.240] But the complaints that I get have caused me to work on it and actually implement stuff that makes people's usage experience easier.
[00:19:29.680 --> 00:19:36.760] That was mostly in the search interface.
[00:19:37.320 --> 00:19:40.760] It was mostly in search is weird, type ahead search is weird.
[00:19:40.760 --> 00:19:46.360] We want to be able to type a phrase, use and and or and that kind of stuff, then hit enter and get results.
[00:19:46.360 --> 00:19:49.800] And then we want to have one button where we can add all of the results to a list.
[00:19:49.800 --> 00:19:52.600] That was something that I didn't even expect anybody to do.
[00:19:52.600 --> 00:19:55.640] Like lists to me were an afterthought for the whole tool.
[00:19:55.640 --> 00:20:10.600] But it turned out that for a PR agency that has a customer that needs a list of podcasts, well, yeah, they need a list to put the podcasts in and then export that list into their own whatever system so they can present it to their customers or their clients.
[00:20:10.600 --> 00:20:19.800] So it was from these conversations with customers that I was like, okay, they need this and it needs to look good enough to know where to click to get it so they can get the CSV export.
[00:20:19.800 --> 00:20:24.200] So it's still a very functional, almost a very procedural approach.
[00:20:24.200 --> 00:20:26.600] Like they have a procedure that they need to follow.
[00:20:26.600 --> 00:20:29.960] Find thing, add to list, export list, present, get money.
[00:20:29.960 --> 00:20:31.880] That's their job to be done, right?
[00:20:31.880 --> 00:20:37.160] So if I can facilitate this, then what we currently have seems to be good enough.
[00:20:37.160 --> 00:20:41.720] Although, gotta say, like, I switched out the whole search systems a month ago.
[00:20:41.720 --> 00:20:50.120] Like, the search that is now currently running is way better and technically completely differently implemented than the one we had before.
[00:20:50.120 --> 00:20:53.480] And it shows because it just gives you better results.
[00:20:53.480 --> 00:20:55.320] So that's the UX stuff.
[00:20:55.320 --> 00:20:58.040] The data is better, not necessarily the interface.
[00:20:58.360 --> 00:20:59.480] Yeah, well, that's the thing.
[00:20:59.480 --> 00:21:04.200] Like, in some situations, like the UI doesn't matter as much.
[00:21:04.520 --> 00:21:15.280] And because of it, the impact of a certain tool, an AI-powered tool, you know, Lovable and the others, is also much smaller.
[00:21:15.520 --> 00:21:16.960] It doesn't really matter.
[00:21:14.840 --> 00:21:21.520] In other cases, the UI is super important, especially when you have lots of competition.
[00:21:21.840 --> 00:21:25.520] I don't think PodScan has too many competitors at this point.
[00:21:25.520 --> 00:21:27.360] And they look pretty bad as well.
[00:21:27.680 --> 00:21:36.160] All these tools, like the social monitoring tools and the podcast database tools, they were oldish or at least not very modernly designed.
[00:21:36.160 --> 00:21:36.880] I mean, that's the thing.
[00:21:36.880 --> 00:21:45.200] Like, if people have multiple options, at that point, the UI will matter more because then it's a matter of how easy is it to do something?
[00:21:45.200 --> 00:21:48.640] How easy is it to get your first alert set up in PodScan?
[00:21:48.640 --> 00:21:51.120] Like you said, like search is super important.
[00:21:51.120 --> 00:21:53.520] Like, can I find what I'm looking for?
[00:21:53.520 --> 00:21:55.120] If yes, yay, PodScan.
[00:21:55.200 --> 00:21:56.720] If no, this thing sucks.
[00:21:56.720 --> 00:21:58.800] Let's try out competitor A or B.
[00:21:59.120 --> 00:22:04.080] Especially in the beginning when you're on a trial plan, like you have no relationship with your product yet.
[00:22:04.080 --> 00:22:07.040] So it's very easy to just go somewhere else.
[00:22:07.040 --> 00:22:14.960] So there are so many different factors at play that really determine if UI or UX or design as a whole matters or how much it matters.
[00:22:14.960 --> 00:22:17.520] Because I think it always matters a little bit.
[00:22:17.520 --> 00:22:20.800] Yeah, I wonder, do you think that changes with the age of a company?
[00:22:20.800 --> 00:22:35.840] Like for a startup that just started, either Vibe coded or regular coded, whatever that might mean, how important is that kind of focus on consistency and uniformity and beauty of the interface?
[00:22:35.840 --> 00:22:37.040] What have you experienced?
[00:22:37.040 --> 00:22:41.600] Does age of the company and established market presence play into that?
[00:22:41.600 --> 00:22:42.240] I think so.
[00:22:42.240 --> 00:22:48.320] I think that if you've been around for a long time, people have a whole image of you and who you are.
[00:22:48.320 --> 00:22:49.840] Like, that's your branding.
[00:22:49.840 --> 00:22:53.520] Just look at the old Jaguar stuff, or now with Apple and Liquid Glass.
[00:22:53.560 --> 00:23:02.200] Like, there's more discussion between pro and against people of a certain happening, the more branding you have.
[00:23:02.200 --> 00:23:06.840] So, if you have lots of fans, you will also have lots of people who don't like you.
[00:23:06.840 --> 00:23:10.600] So, that's way stronger for an established brand.
[00:23:10.600 --> 00:23:23.960] When you're super new, it's way more important to get off the ground, survive, you know, revenue-wise, and see if you can establish something for yourself, you know, product market fit-wise.
[00:23:23.960 --> 00:23:26.360] And design plays a role there.
[00:23:26.360 --> 00:23:34.760] But I know just like with PodScan, like in the beginning, like when we first worked on PodScan, we really first had to figure out like one or two main user flows.
[00:23:34.760 --> 00:23:42.040] Now, like, if we have limited time, limited budget, if we change one thing or improve one thing design-wise, what should it be?
[00:23:42.040 --> 00:23:45.480] Like, that's been a whole discussion we had well over a year ago, I guess.
[00:23:45.480 --> 00:23:48.440] Because then you want to design a thing that has the biggest impact.
[00:23:48.440 --> 00:23:52.520] And we went with onboarding and setting up your first alert.
[00:23:52.520 --> 00:24:03.880] We wanted that to be as easy and wow moment, uh, as much wow moment as possible, as quickly as possible, because then it's way more likely for a trial user to extend.
[00:24:03.880 --> 00:24:09.320] Like, if they have had no benefit, they're like, Well, trial version is over, I'll go somewhere else.
[00:24:09.320 --> 00:24:14.760] But if they see, like, oh, this is useful, I will stay, I will give Arvid my money.
[00:24:14.760 --> 00:24:16.200] That's super important.
[00:24:16.200 --> 00:24:19.240] So, that's what I would do for a new company.
[00:24:19.240 --> 00:24:29.160] And that requires a lot of talking and brainstorming and having meetings together, which all happens outside of you know, lovable like tools.
[00:24:29.160 --> 00:24:39.320] There are still AI helpers, there, you know, transcribers, and throwing your transcription into a tool to summarize and make up next actions, that kind of stuff, like you said before.
[00:24:39.320 --> 00:24:43.320] Yeah, but that's the human side that will probably stay around.
[00:24:43.320 --> 00:24:44.600] Who knows?
[00:24:44.640 --> 00:24:50.000] Yeah, I do wonder how much of this can be automated away because I'm looking at transcription tools.
[00:24:50.000 --> 00:24:56.560] Obviously, like all I'm thinking about is transcription every single day, but even for a conversation like this, this is going to be transcribed, right?
[00:24:56.560 --> 00:25:03.200] And this episode of my podcast will land on PodScan, be transcribed, and searched for keyword terms.
[00:25:03.200 --> 00:25:07.840] So, if you want anybody to be notified at this point, just say all those phrases, right?
[00:25:07.840 --> 00:25:11.600] Like, everything now becomes observed and measured and tracked.
[00:25:11.600 --> 00:25:16.720] And I wonder, like, how much of this should flow back into the product design that we have?
[00:25:16.720 --> 00:25:30.320] Like, when I think about a feature that I recently built for PodScan internally for the company for sales enablement, is whenever somebody signs up, they go through the onboarding flow and we kind of encourage them to create an alert.
[00:25:30.320 --> 00:25:35.680] And we encourage them, if they don't, to at least try to search to just see the value prop of the business.
[00:25:35.680 --> 00:25:38.000] And then we track each of these things internally, right?
[00:25:38.000 --> 00:25:44.960] When somebody navigates to a podcast page, looks at a show or they look at a specific episode, we track this as an activity in our system.
[00:25:44.960 --> 00:25:59.760] So we can say, within the first 24 hours, this account, after signing up, went through onboarding, they created an alert, or maybe not, got a notification from this alert because something was found, like a mention was found, or they checked out the API documentation, right?
[00:25:59.760 --> 00:26:04.880] We look for intent, and then 24 hours later, we score each of these people's behavior.
[00:26:04.880 --> 00:26:09.520] Plus, we try to extract where they're from, what their name is, what the company is that they're working for.
[00:26:09.520 --> 00:26:15.280] And then we give them a score and how important it is for us to reach out to them and engage with them as customers.
[00:26:15.280 --> 00:26:24.880] There is a lot of value in doing all of this in an automated way: the tracking and the scoring and looking at which of these things is important and which is not.
[00:26:24.880 --> 00:26:35.960] But the one component that needs to be done by people, because otherwise, this would not be a business, it would just be a machine, is then reaching out to that person, having a conversation with them, booking a call for a demo.
[00:26:36.200 --> 00:26:38.120] That needs to be done by humans still.
[00:26:38.120 --> 00:26:39.400] Well, that's very true.
[00:26:39.640 --> 00:26:44.360] It also made me think about different projects I worked in in the past years.
[00:26:44.360 --> 00:26:46.680] Like my career is 10 years now in design.
[00:26:46.680 --> 00:26:52.440] So it's been quite a few projects and there's overlap, but they're hardly ever the same.
[00:26:52.440 --> 00:26:57.160] There's always a different founder with a different desire or goal.
[00:26:57.160 --> 00:26:59.160] They're all in different stages.
[00:26:59.160 --> 00:27:07.640] So I don't see too much automation happening there because I also feel like if we keep talking about onboarding, you have best practices.
[00:27:07.640 --> 00:27:16.840] For example, have that aha moment or that wow moment as quickly as possible because then it's more likely people will reach that point and it's more likely for them to stick around.
[00:27:16.840 --> 00:27:18.040] You know, that's a rule.
[00:27:18.040 --> 00:27:19.720] But how do you make it shorter?
[00:27:19.720 --> 00:27:20.920] That's the thing.
[00:27:20.920 --> 00:27:23.080] In practice, it's different each time.
[00:27:23.080 --> 00:27:29.000] And then there are also exceptions to the rule where you want to delay that aha moment on purpose.
[00:27:29.000 --> 00:27:33.160] It's not always the case to be as quickly as possible.
[00:27:33.160 --> 00:27:36.520] So I think that requires a bit of human interference.
[00:27:36.520 --> 00:27:39.720] There's a human who needs to be thinking about this stuff.
[00:27:39.720 --> 00:27:41.720] That's the coping I tell myself.
[00:27:41.720 --> 00:27:45.160] There needs to be a human for that part because it's not always the same.
[00:27:45.160 --> 00:27:46.840] Yeah, I have another example for this.
[00:27:46.840 --> 00:27:49.160] I think that's going to stress the point here.
[00:27:49.160 --> 00:28:05.400] In a couple of the demo calls that I had over the last weeks, I've shown the alerts feature and how to set up an alert, how to configure it to a lot of people in the PR agency world because that's the customer that we found has the strongest desire to actually pay for this product right now.
[00:28:05.400 --> 00:28:07.640] So, those are the people we'll be reaching out to.
[00:28:07.640 --> 00:28:13.000] And sales calls are still not easy for me because it's just not how I think.
[00:28:13.000 --> 00:28:21.840] I don't think in terms of what a salesperson would say, which is why it is nice to have Russell on board who's helping me learning this and actually doing this as well without my presence.
[00:28:22.000 --> 00:28:28.160] But in those calls, I showed them, you know, here's alerts, and here you can set up the filters for the alerts.
[00:28:28.160 --> 00:28:29.840] And we're like, okay, yeah, filter keywords.
[00:28:29.840 --> 00:28:30.640] Okay, I get this.
[00:28:30.640 --> 00:28:33.360] And then you can select, okay, what language do I want the alert to be?
[00:28:33.360 --> 00:28:35.520] And okay, English-speaking podcast from the US.
[00:28:35.520 --> 00:28:36.320] Okay, sure.
[00:28:36.320 --> 00:28:39.040] And then I showed them the context-aware question filter.
[00:28:39.040 --> 00:28:43.920] Okay, and here you can put in any question you could potentially have that has a yes or no answer.
[00:28:43.920 --> 00:28:46.240] And if it's true, then you get an alert.
[00:28:46.240 --> 00:28:48.400] And if it's not, then it gets dismissed.
[00:28:48.400 --> 00:28:54.800] And at that point, reliably, eyes just fly open, like chins drop, and eyes open up.
[00:28:54.800 --> 00:28:55.760] Wait, wait, wait, wait, wait.
[00:28:55.920 --> 00:29:00.640] You're telling me that I could ask one question of every single podcast out there.
[00:29:00.640 --> 00:29:06.800] And if it's true, then I get an alert, like any question, like to any detail about anything.
[00:29:06.800 --> 00:29:09.280] And then you see them literally thinking.
[00:29:09.280 --> 00:29:22.960] I've had conversations where there was like five people sitting in a just a meeting room somewhere, sterile business meeting room, and you would see them look at each other and like you could see them think about all the potential things you could do.
[00:29:22.960 --> 00:29:36.640] So one of the things that I immediately did was stress that part of the window or the form in which the alert is configured by putting examples there and like showing the variety of questions you can ask.
[00:29:36.640 --> 00:29:45.760] Before it was just a line where you could put a question in and nobody would think much of it, but now it's a whole segment with like, and here you can ask any question you want of the application.
[00:29:45.760 --> 00:29:49.840] And that has reliably created more alerts for more people.
[00:29:49.840 --> 00:29:54.000] Like now, I see people actually setting up alerts because they're like, oh, I really want to use this.
[00:29:54.000 --> 00:30:01.480] But obviously, that's also shot up my AI usage in the background because that is the one feature that needs to run an AI query on all of this.
[00:29:59.680 --> 00:30:07.320] But I'd rather pay a little bit more in AI fees if people use the product and find it worth paying a couple hundred bucks a month for it.
[00:30:07.480 --> 00:30:07.800] Right.
[00:30:07.800 --> 00:30:13.640] So that moment of people just looking at each other in disbelief, you can do this.
[00:30:13.640 --> 00:30:15.480] Nobody else does this.
[00:30:15.480 --> 00:30:16.760] They told me this.
[00:30:16.760 --> 00:30:20.040] This is a wow moment happening in front of my eyes.
[00:30:20.040 --> 00:30:21.720] This I can facilitate, right?
[00:30:21.720 --> 00:30:23.720] This I can push people towards.
[00:30:23.720 --> 00:30:34.680] And I would never have known if I hadn't showed them, if I hadn't had this conversation with them, and then recorded the conversation so I could figure out stuff after the fact and see these things actually happen in real time.
[00:30:34.680 --> 00:30:39.240] That's the pattern recognition that we as human beings are very proficient in.
[00:30:39.560 --> 00:30:42.840] At some point you were like, hey, wait a minute, this thing's happening.
[00:30:42.840 --> 00:30:46.280] And then you have to translate it to a way to show people.
[00:30:46.280 --> 00:30:51.240] And it's also something perhaps to put on a landing page, or maybe you should create a video where you show it.
[00:30:51.240 --> 00:30:57.080] Or maybe it deserves its own page because it's such a massive feature of PodScan.
[00:30:57.480 --> 00:31:01.000] So there are so many different things to potentially do about it.
[00:31:01.000 --> 00:31:03.800] And this is one thing that was super clear, super obvious.
[00:31:03.800 --> 00:31:07.560] And then you also have the more subtle reactions from people.
[00:31:07.560 --> 00:31:10.760] I mean, this one was, it was impossible to miss this one.
[00:31:10.760 --> 00:31:13.720] It was such a, like you said, they were all looking at each other.
[00:31:13.720 --> 00:31:15.320] Like, you know, it's hard to ignore.
[00:31:15.320 --> 00:31:17.560] But there are a few subtle ones as well.
[00:31:17.560 --> 00:31:32.200] I know back in the day when we did user research, you had the host and you had the note taker/slash observer who would look for the subtle cues within the test subject/slash human being in a user research or in a user test.
[00:31:32.200 --> 00:31:33.480] So that's super important.
[00:31:33.480 --> 00:31:40.920] I'm not sure how AI can do it if it's not text-based, unless you point a camera at someone and you can do facial recognition.
[00:31:40.920 --> 00:31:43.080] So there are all ways to assist.
[00:31:43.080 --> 00:33:18.280] They might find it in the voice pattern too you might find like slight agitation in the audio or something i i think ai can help like find just excitement or change in positivity and sentiment throughout a conversation that can also be detected right where they're grumpy in the beginning and they're using a lot of no's and babies and potentially and then at some point it shifts into oh yeah we should and we will and we have to right into other kinds of phrases i bet somebody's working on this i'm thinking about what is it clueli have you seen this have you seen this tool called cluly which apparently it just tracks what you say and what somebody else is saying in a call and then it gives you points at what you should be asking next or what you should be saying next i haven't used that it's been a dream of mine to have a tool like this for conversations like this where maybe i'm running out of steam at some point and i need another question to ask right as an interviewer this is useful as a salesperson it's useful to not forget what the next question should be that you ask them right or they have said so many things here's something you can recontextualize so having this kind of permanent guardian on your site that gives you ideas that gives you insights that tracks what's going on and gives you the next thing to do next thing to say that is the use of ai that i want to see more that's the augmented part right the augmentation it's not replacing anything but it augments what we already do to do it better i can see this for user interviews customer discovery customer service calls customer success calls or sales calls marketing stuff it's kind of nice to have somebody think about what you should be doing next at all times i agree but i also have a main concern here, and that's very philosophical.
[00:33:18.280 --> 00:33:20.520] Like, does it make us dumber?
[00:33:20.520 --> 00:33:22.440] But at the same time, does it matter?
[00:33:22.440 --> 00:33:24.920] Well, let's spend a couple of hours on this question.
[00:33:24.920 --> 00:33:28.200] Because honestly, I think all technology has this potential, right?
[00:33:28.200 --> 00:33:36.280] We've seen just in our lifetime, the ever-present mobile phone with access to the whole world's knowledge.
[00:33:36.280 --> 00:33:43.720] I mean, ask your grandparents if they're still around or your parents about dates, important, relevant historical dates.
[00:33:43.720 --> 00:33:51.960] And they can probably still regurgitate them from back when they were in school because they were just hammered into them and they had to learn them by heart.
[00:33:51.960 --> 00:33:56.280] And it still sticks around because that's the only way that knowledge was accessible.
[00:33:56.280 --> 00:33:57.880] If you had it in your mind, that was great.
[00:33:57.880 --> 00:33:59.400] And if not, you didn't have it, right?
[00:33:59.400 --> 00:34:04.440] You could go to the library, maybe check out a book and write it down, and then you had to look at the piece of paper.
[00:34:04.440 --> 00:34:08.760] 10 years ago, it was Wikipedia on your phone and you had all that data.
[00:34:08.760 --> 00:34:17.080] And now you literally have voice-enabled agents running at the same time that will show you in real time the knowledge that you need to know.
[00:34:17.080 --> 00:34:19.960] So I think the tendency is to outsource this.
[00:34:19.960 --> 00:34:32.440] I think that's called the body replacement thesis or something, that there's a philosophical theory that says that we are outsourcing capacity from our own body into tools, right?
[00:34:32.440 --> 00:34:34.680] The hand is strong, but a hammer is stronger.
[00:34:34.680 --> 00:34:37.720] The teeth are strong, but a saw is much stronger than those.
[00:34:37.720 --> 00:34:43.080] So the brain is strong, but a phone with connected access to the internet is stronger than the brain.
[00:34:43.080 --> 00:34:48.200] So we tend to outsource things if there is a possibility because they're more effective.
[00:34:48.200 --> 00:34:56.520] And I think for AI tools, we will take yet another thing, which is real-time reasoning, and outsource that to another tool.
[00:34:56.520 --> 00:35:06.040] That's probably not the best thing we can do to stay in full control of our cognitive capacity, but it will make things easier, which is also a human tendency, right?
[00:35:06.040 --> 00:35:07.720] To go for the easy way.
[00:35:07.720 --> 00:35:08.120] True.
[00:35:08.120 --> 00:35:20.640] We're animals and we are wired to spend the less amount of energy as possible just to survive because our animal brain doesn't know if the supermarket is open tomorrow or well, if you can find food tomorrow.
[00:35:20.960 --> 00:35:25.360] You know, but I know I know quite a few phone numbers still by heart.
[00:35:25.360 --> 00:35:31.920] But anyway, I'm thinking, you know, design-wise, I know like all the classics.
[00:35:31.920 --> 00:35:35.840] Like, I know gestalt and typography, color theory.
[00:35:35.840 --> 00:35:40.400] I had ethics, psychology, all that kind of stuff in school.
[00:35:40.400 --> 00:35:42.880] Nothing I do design-wise is random.
[00:35:42.880 --> 00:35:46.480] Like, everything on the page is there for a reason.
[00:35:46.480 --> 00:36:02.320] When I look at the new generation, which makes me sound very old, even though I'm only 32, but the new generation of designers who are using lots of vibe designing and that kind of stuff, I see that they know less about it.
[00:36:02.320 --> 00:36:08.320] But they have projects and they sell stuff and they make products, which makes me think: does it matter?
[00:36:08.320 --> 00:36:12.880] Do I need to be as proficient with the gestalt and all this stuff like I am?
[00:36:12.880 --> 00:36:19.200] I think yes, I am more proficient and better than what an AI can generate UI-wise.
[00:36:19.200 --> 00:36:27.200] But you can see other people who are more AI-proficient and less knowledgeable theory-wise, they manage as well.
[00:36:27.200 --> 00:36:28.800] So, that's something I struggle with.
[00:36:28.800 --> 00:36:38.880] Like, I think craft is very important, knowledge, having, you know, being able to stand firm when the internet is out, being able to do stuff still.
[00:36:39.200 --> 00:36:44.880] I think that's very important because you wouldn't go to a restaurant where the chef is like, Well, it's warm.
[00:36:44.880 --> 00:36:47.360] This person wants something that's cool.
[00:36:47.520 --> 00:36:50.800] He wants something that's sweet, but not bitter.
[00:36:50.800 --> 00:36:53.280] Hey, black box thingy, make me a dish.
[00:36:53.360 --> 00:36:59.720] No, you go there because the chef has a certain image, like it's a three-star chef.
[00:36:59.280 --> 00:37:05.080] I think it's very important to have skills, old school skills, still.
[00:37:05.720 --> 00:37:12.760] There's something to be said for expertise embodied in a human that can work without tools as well.
[00:37:13.000 --> 00:37:14.520] That's the thing with the hammer.
[00:37:14.520 --> 00:37:16.680] Somebody still has to guide the hammer.
[00:37:16.680 --> 00:37:21.400] Somebody still has to know where to put the saw to actually cut down the tree.
[00:37:21.640 --> 00:37:27.000] Like every Christmas, I'm painfully reminded of the fact that I'm not very experienced with that.
[00:37:27.000 --> 00:37:37.720] But that expertise and the experience that leads to it is something that I think, like you do, is something that you still need to have to be able to judge the output of these systems.
[00:37:37.720 --> 00:37:48.520] The thing that I see with my whole AI-based coding efforts that I do all the time, every single day, is I still need to be very good at explaining exactly what I want.
[00:37:48.520 --> 00:37:53.160] And if I know that it's hard to build, I need to give it some hints on how to build it.
[00:37:53.160 --> 00:37:57.000] Otherwise, it would probably find another weird way that may or may not work.
[00:37:57.000 --> 00:38:02.280] But if I already kind of know what good looks like, and I think you, you know this from the design side of things.
[00:38:02.280 --> 00:38:04.840] You know, if a design is good, you see it.
[00:38:04.840 --> 00:38:06.680] And if it's not good, you see that.
[00:38:06.680 --> 00:38:08.520] And I think I do the same thing for code.
[00:38:08.520 --> 00:38:14.920] When I look at code that gets created by an AI system, I very quickly can tell you if it's good code or not.
[00:38:14.920 --> 00:38:21.800] I can't necessarily tell you immediately if it's going to work exactly the way I want it to work, but I can tell you if it's going to be shit.
[00:38:21.800 --> 00:38:27.560] So to be able to have this capacity, I still needed to do this for the last 20 years, right?
[00:38:27.560 --> 00:38:34.520] Like to be able to understand what good code looks like, I needed to very painfully build up good code as a capacity.
[00:38:34.520 --> 00:38:37.000] And now I can judge it without having to do it.
[00:38:37.000 --> 00:38:39.720] But I needed to do it to get to the point of being able to judge it.
[00:38:39.720 --> 00:38:40.520] That's true.
[00:38:40.520 --> 00:38:47.840] And for me, it really helps that my father started to teach me design when I was six, you know, so there's almost second nature for me.
[00:38:48.160 --> 00:38:56.480] And also, the same thing applies, you know, just this week where I was working on a Laravel project with Blade and that kind of stuff.
[00:38:56.480 --> 00:38:58.640] Very familiar for you, I assume.
[00:38:58.960 --> 00:39:06.000] I made a design, client agrees, and they asked me to also put the design into a Blade template.
[00:39:06.000 --> 00:39:12.720] And I wanted something like a table of contents to be sticky on scroll, like that it stayed on a page with a few custom things.
[00:39:12.720 --> 00:39:16.320] I did a bit of work, then I asked the AI to do that for me.
[00:39:16.320 --> 00:39:17.280] It looked all fine.
[00:39:17.280 --> 00:39:19.440] I load the page, not sticky.
[00:39:19.440 --> 00:39:21.200] So I asked, like, hey, what's going on?
[00:39:21.200 --> 00:39:27.280] And it turns out I spent, I think, 30 or 60 minutes together with the AI debugging with console logs.
[00:39:27.280 --> 00:39:36.480] And in the end, we ended up doing a JavaScript version of calculating where the sticky thing should be on the page.
[00:39:36.480 --> 00:39:37.840] It was messy.
[00:39:37.840 --> 00:39:39.680] It was terrible code.
[00:39:39.680 --> 00:39:49.280] And it turned out the only problem was that at the top of the page, like the styling was in the head section, and we had the push instead of the section code there.
[00:39:49.280 --> 00:39:54.000] So all the code the AI made wasn't on the actual main layout.
[00:39:54.000 --> 00:39:56.960] So it's literally one word that needed to be changed.
[00:39:56.960 --> 00:40:03.840] And the people I work with, also with tons of experience, they were able to point it out to me, like, oh, it has to be this thing here.
[00:40:03.840 --> 00:40:06.160] And I replaced it and everything worked.
[00:40:06.160 --> 00:40:09.200] You know, so that's that 20-year experience you're talking about.
[00:40:09.200 --> 00:40:10.160] I didn't have it.
[00:40:10.160 --> 00:40:15.920] I explained it as good as I was able to with my semi-professional coding knowledge.
[00:40:15.920 --> 00:40:18.720] And the AI had to run with it, but it wasn't good enough.
[00:40:18.720 --> 00:40:21.200] And because of it, it took way longer than needed.
[00:40:21.200 --> 00:40:23.520] And the quality was way lower than needed.
[00:40:23.520 --> 00:40:25.600] Yeah, so that's an example of what you're saying.
[00:40:26.000 --> 00:40:30.520] I see most people who are using AI primarily to be quicker.
[00:40:29.920 --> 00:40:32.840] I use AI to become better.
[00:40:33.160 --> 00:40:39.080] I learn Laravel, I learn React, good enough for me to ship my design in code.
[00:40:39.080 --> 00:40:40.920] And I become quicker because of it.
[00:40:40.920 --> 00:40:46.520] But my main benefit from AI is that I know much more than I knew a year ago.
[00:40:46.520 --> 00:40:49.000] And I think that's a mindset I don't see too often.
[00:40:49.000 --> 00:40:56.760] Like on social media, at least, most people focus like 99% of their AI usage on doing more with less.
[00:40:56.760 --> 00:40:59.240] And that hurts the quality too.
[00:40:59.240 --> 00:41:04.280] Like if you go fast and break things, quality and security might suffer.
[00:41:04.280 --> 00:41:05.960] Well, I feel very similarly to this.
[00:41:05.960 --> 00:41:18.120] I think like if people exclusively use AI as a generative tool and then use that generated stuff as the product, they're missing the crucial component that is the AI should always be wrapped by a human, right?
[00:41:18.120 --> 00:41:21.720] You start with a human that has an idea that tells the AI what to do.
[00:41:21.720 --> 00:41:23.560] The AI then does the thing.
[00:41:23.560 --> 00:41:32.440] And then instead of just taking that and putting it somewhere and selling it or taking it into production, there still needs to be this kind of observer process of a human at the end.
[00:41:32.440 --> 00:41:40.280] There still needs to be 20% of the work done afterward to make sure that it fits, to make sure that it's actually what you wanted it to be.
[00:41:40.280 --> 00:41:51.240] The push thing that you said, I had the exact same problem two days ago on a very different project, but it was like adding something to a component and it was like, oh yeah, and then this script runs and this script runs, but it didn't.
[00:41:51.240 --> 00:41:53.160] And because again, it missed that one thing.
[00:41:53.160 --> 00:41:55.800] It's apparently not very good at push, you know?
[00:41:55.800 --> 00:41:59.880] And that's something that you can only understand if you know how it should work in the first place.
[00:41:59.880 --> 00:42:01.480] So there needs to be like oversight.
[00:42:01.480 --> 00:42:03.560] That's the managerial part I'm talking about.
[00:42:03.560 --> 00:42:04.760] You're an AI wrangler.
[00:42:04.760 --> 00:42:05.640] You're a shepherd.
[00:42:05.720 --> 00:42:13.560] You wouldn't just let your sheep go wherever they want or your dog, your shepherd dog for that matter, it still needs to be trained, right?
[00:42:13.560 --> 00:42:20.080] Even though it's good at herding sheep, it's still a dog and will still do whatever it wants if you don't train it to do its job well.
[00:42:14.840 --> 00:42:20.880] No, that's very true.
[00:42:21.040 --> 00:42:22.720] And we'll probably get better at that.
[00:42:22.720 --> 00:42:28.000] But having the gatekeeping human at the end is probably a smart thing to do.
[00:42:28.000 --> 00:42:29.440] That sounds about right to me.
[00:42:29.440 --> 00:42:32.480] You know, that is the thing we will have to deal with over the next years.
[00:42:32.480 --> 00:42:47.440] Like next 10 years or so is going to be developing this internal skill and expertise at working with AI systems without having AI systems go rogue or rampage around in our code base or create things that we didn't want them to create.
[00:42:47.440 --> 00:42:50.720] The skill here is going to be judgment, like, is this good?
[00:42:50.720 --> 00:42:51.760] And discernment.
[00:42:51.760 --> 00:42:52.960] Is this what I want?
[00:42:52.960 --> 00:43:06.400] Those are new skills that also I think can't really exist already because they are so bound to the technology on which they need to be developed that all you can do at this point is to keep prompting and keep checking.
[00:43:06.400 --> 00:43:08.400] That's the dynamic that I feel.
[00:43:08.400 --> 00:43:19.360] I just need to keep telling the thing to do stuff and then learn how my scoped argument of what it should be doing is translated or not into functional code.
[00:43:19.360 --> 00:43:21.200] I'm learning a new language here.
[00:43:21.200 --> 00:43:28.400] I recently had a conversation with somebody around the idea of, well, what are programming languages going to look like in the future?
[00:43:28.400 --> 00:43:31.040] And his point was, it's probably going to be English.
[00:43:31.040 --> 00:43:35.120] And my argument was yes, but probably with a different level of precision, right?
[00:43:35.120 --> 00:43:38.160] We're probably not going to say, hey, build me an UI for this.
[00:43:38.320 --> 00:43:42.720] Probably going to say we need, like in the old days, the hamburger menu, right?
[00:43:42.720 --> 00:43:46.000] It's a weird phrase, but at least it describes exactly what you want.
[00:43:46.000 --> 00:43:57.680] And I think we're going to have components that are language-based that AI systems can reliably implement for us at some point, much more than we're talking about actual code and actual programming languages.
[00:43:57.680 --> 00:43:59.360] That might not matter as much anymore.
[00:43:59.360 --> 00:44:05.320] Same for design, design systems might be expressed completely differently, not in pixels, but in vibes.
[00:44:05.480 --> 00:44:09.240] I don't know, vibe is a loaded word, but how does this feel, right?
[00:44:09.400 --> 00:44:11.320] Like, what does this represent?
[00:44:11.640 --> 00:44:16.600] I just know that this is going to be something that will change and we better stay on top of it.
[00:44:16.600 --> 00:44:17.480] Well, that's true.
[00:44:17.480 --> 00:44:23.320] I mean, that's the main thing to do, just to be open and to learn and to listen to people with different opinions too.
[00:44:23.320 --> 00:44:26.840] As a side note, it's like talking to a two-year-old.
[00:44:26.840 --> 00:44:33.560] When I tell my daughter in the evening, she's two, by the way, we are going to the zoo tomorrow.
[00:44:33.560 --> 00:44:37.320] She runs towards the door, grabs her shoes, and she's ready to go.
[00:44:37.320 --> 00:44:40.200] She stops listening at we are going to the zoo.
[00:44:40.200 --> 00:44:42.440] She doesn't listen to tomorrow.
[00:44:42.440 --> 00:44:46.440] And that's also with the coding and the AI-assisted coding that I'm doing.
[00:44:46.440 --> 00:44:48.760] Like, I cannot say I'm using Laravel and Bootstrap.
[00:44:48.760 --> 00:44:51.240] I have to say Laravel Blade and Bootstrap 5.
[00:44:51.240 --> 00:44:53.080] I have to be extremely specific.
[00:44:53.080 --> 00:44:54.040] And that's not enough.
[00:44:54.040 --> 00:45:00.680] I also have to say again, just to be certain, use as much Bootstrap 5 native classes as you can.
[00:45:00.680 --> 00:45:01.880] And I repeat it all the time.
[00:45:01.880 --> 00:45:03.880] Like, it's like talking to a two-year-old.
[00:45:03.880 --> 00:45:09.240] Like, people say it's like talking to an intern, but I think it's more like talking to a two-year-old.
[00:45:09.240 --> 00:45:15.320] That is super interesting because that is probably exactly how old this technology is at this point, too.
[00:45:15.640 --> 00:45:22.760] You know, if you think about Lovable is like, what, four or five months old, and Claude Code is just a bit younger than that.
[00:45:22.760 --> 00:45:26.200] Like Claude Anthropic, like 3.7 was out a year ago.
[00:45:26.200 --> 00:45:33.480] Like all these tools that we now use on this daily basis, they are younger than your child in some way.
[00:45:33.480 --> 00:45:37.640] And it's probably fine to talk to them like one in a development stage.
[00:45:37.640 --> 00:45:38.680] Very interesting point.
[00:45:38.680 --> 00:45:39.160] I agree.
[00:45:39.160 --> 00:45:42.200] When I do code stuff, I repeat myself all the time.
[00:45:42.200 --> 00:45:48.160] I give like explicit things like scan the whole code base for all functionality that has to do with this.
[00:45:48.480 --> 00:45:56.240] Like, just go through the code base into every PHP file and look for this term and make sure that if you change code, you also change it there.
[00:45:56.240 --> 00:45:56.560] Okay?
[00:45:56.880 --> 00:45:57.120] Right?
[00:45:57.120 --> 00:45:59.920] That's kind of how I talk to it at this point.
[00:45:59.920 --> 00:46:03.360] Well, that makes me think about more two-year-old anecdotes.
[00:46:03.360 --> 00:46:05.040] Like, she likes fruit.
[00:46:05.200 --> 00:46:06.080] She loves mangoes.
[00:46:06.080 --> 00:46:09.840] So then we go to the store, like 500 meters down the road.
[00:46:09.840 --> 00:46:18.640] Whenever she gets distracted by a flower or a cat or anything, basically, I keep telling her, Hey, we're going to the store to get mangoes.
[00:46:18.640 --> 00:46:20.080] And then she's like, oh, yeah.
[00:46:20.080 --> 00:46:21.440] And then she keeps walking.
[00:46:21.440 --> 00:46:24.640] Like, I have to, like, she's zigzagging the whole time.
[00:46:24.640 --> 00:46:29.440] And I keep saying the same thing because I know that's the initial goal.
[00:46:29.440 --> 00:46:31.120] It feels like that's exactly the same.
[00:46:31.120 --> 00:46:33.040] But no, no, no, it's Bootstrap 5.
[00:46:33.040 --> 00:46:34.240] Look at this file.
[00:46:34.240 --> 00:46:35.760] Oh, that's funny.
[00:46:36.080 --> 00:46:37.520] I think you cracked it though.
[00:46:37.520 --> 00:46:43.840] And I think all parenting advice for how to talk to young kids is probably how you should be prompting at this point.
[00:46:43.840 --> 00:46:44.160] Yeah.
[00:46:44.480 --> 00:46:45.840] Work with Claude for a week.
[00:46:45.840 --> 00:46:54.560] When your wife is like 36 weeks plus pregnant, you should do a crash course managing Claude just as a baby prep.
[00:46:54.560 --> 00:46:55.920] That is so funny.
[00:46:55.920 --> 00:47:01.840] In some ways, talking to a child of that age is talking to somebody who's just easily distracted.
[00:47:01.840 --> 00:47:04.240] They have a hard time grasping complicated things.
[00:47:04.560 --> 00:47:06.400] They go back to what they know, right?
[00:47:06.400 --> 00:47:16.240] It's pretty much the attention span of an AI that is kind of summoned into being for your request and then vanishes into the void after.
[00:47:16.240 --> 00:47:18.160] Like it's pretty much that.
[00:47:18.160 --> 00:47:22.480] So unsurprising that you have to talk to like a two-year-old man.
[00:47:22.800 --> 00:47:32.040] Just a great opportunity to pull it all together because we're at a stage where we're just like rambling on about the future of AI.
[00:47:32.360 --> 00:47:40.920] But for anybody who wants to see you share your insights about how to talk to two-year-olds in public, where do you want them to go?
[00:47:40.920 --> 00:47:42.440] Where do you want them to follow you?
[00:47:42.440 --> 00:47:47.960] Well, if they want to talk about design, then it's still on Twitter slash X.
[00:47:47.960 --> 00:47:52.520] I'm still using the old handle for my very first website called Toolbox of Design.
[00:47:52.680 --> 00:47:58.040] I also have the askadesigner.io website, which I'm currently redesigning.
[00:47:58.040 --> 00:48:05.560] If you go to the website today, you can see it's in this in-between state where you can see a bit of a red color and a bit of green.
[00:48:05.560 --> 00:48:08.280] And the header is updated, but the images aren't.
[00:48:08.280 --> 00:48:12.280] So you can see me build that thing in public.
[00:48:12.280 --> 00:48:25.720] If you want the design plus AI plus baby slash young children mixture, you know, then the incredible Tyler White and myself, we started a design table podcast based on design having a seat at the table.
[00:48:25.720 --> 00:48:27.400] Like that's where the name comes from.
[00:48:27.400 --> 00:48:29.640] He has a two-year-old as well.
[00:48:29.640 --> 00:48:42.120] So about 50% of our episodes end just like this episode is ending with strange AI slash two-year-old metaphors and overlapping behavior.
[00:48:42.120 --> 00:48:44.840] So that's designtablepodcast.com.
[00:48:44.840 --> 00:48:46.360] Man, thanks so much for being on.
[00:48:46.360 --> 00:48:51.960] I appreciate all your insights and knowledge about how to talk to prompting systems.
[00:48:52.120 --> 00:49:02.120] Honestly, I'm going to dive a bit deeper into this and just try to embody this two-year-old mentality and see if that actually improves the code that gets generated by the tool.
[00:49:02.120 --> 00:49:02.760] Cool, Nick.
[00:49:02.760 --> 00:49:04.280] Thanks so much for being on the show.
[00:49:04.280 --> 00:49:05.080] Really appreciate it.
[00:49:05.080 --> 00:49:05.800] Well, thank you.
[00:49:05.800 --> 00:49:07.480] Thank you for having me again.
[00:49:07.480 --> 00:49:08.680] And that's it for today.
[00:49:08.680 --> 00:49:10.920] Thank you so much for listening to the Bootstrap Founder.
[00:49:10.920 --> 00:49:14.360] You can find me on Twitter at Avid Kahl, A-R-V-I-D, K-A-H-L.
[00:49:14.360 --> 00:49:27.120] And if you want to support me in this show, please share podscan.fm, my SaaS business, with your professional peers and those who you think will benefit from tracking mentions of their brands, their businesses, and names on podcasts out there.
[00:49:27.120 --> 00:49:31.600] PodScan is a near real-time podcast database with a stellar API.
[00:49:31.600 --> 00:49:34.800] We have 32 million podcast episodes in the back now.
[00:49:34.800 --> 00:49:36.640] The database is humongous.
[00:49:36.640 --> 00:49:40.960] Please share the word with those who need to stay on top of the podcast ecosystem.
[00:49:40.960 --> 00:49:42.400] Thank you so much for listening.
[00:49:42.400 --> 00:49:44.880] Have a wonderful day and bye-bye.