[00:00:03–00:00:28]
Okay, good morning. Please sit down if there are more seats. Or else take a seat on the floor, maybe, or a table. Nice to see you all. My name is Oliver. I'm heading up 1508, and I'm here to set the stage, give you a warm welcome, and also introduce the two speakers of today.

[00:00:28–00:00:55]
Let's just get going, because I know that there are many slides and there is many things that we can talk about in terms of AI. First off, actually, just a question to you guys. If you just close your eyes and breathe in and then think of one word that actually describes how you feel about AI today.

[00:01:02–00:01:25]
Maybe putting the mic out there. Louise, I knew you were gonna ask me — fear. Fear, yeah. Over here. Yeah, maybe — insecure, insecure. Okay. I'm thankful. Thankful, yeah.

[00:01:25–00:01:49]
Nice. Okay. There's many, many feelings you can have, and all these feelings they are legitimate. You can be excited, you can be worried, you can be anxious, you can be overwhelmed, you can be flabbergasted, you can be so many things. And I meet these feelings inside myself every day, and I also meet them with my colleagues every day.

[00:01:49–00:02:14]
But one thing that we decided was that to be able to actually take a stance on AI, we need to be curious about everything. So before we can actually go out loud and say this is how we actually feel about things, and this is certainly our guidelines and our red lines and so forth, we need to stay curious for a bit. So this is how we are right now — super curious.

[00:02:15–00:02:36]
Still a lot of guidelines, a lot of principles, and a small introduction to some of those. So we are in this every day. We are building agents, we are co-designing with the AI, we are illustrating, we are building co-work agents, and all these things. Some say — and some of the designers say —

[00:02:36–00:02:59]
I have become two to three times faster in my job. Some of the developers say — I've heard one saying — I'm a hundred times faster than I was before. So there is something here that is a completely different thing. So our velocity and our speed has changed a lot. We still sell hours, so we just need to sell three to four times more projects to actually keep this going. That's the bad thing.

[00:03:02–00:03:29]
But what you can say also — a few observations for the past couple of months, because you can't say more than months right now, because things are changing as fast as they actually are. How AI is influencing our ways of working: one of the big things that we are doing right now is that when we are doing a new digital experience, whether it's a website or an app or something completely different, we need to design two different versions of this website.

[00:03:29–00:03:55]
We need to do one for agents and machines, and we need to do one for humans. We call the agents and machines one the highway. This is where you can find all the data within the LLMs and so forth, and we need to construct and structure the data and all the content of the website so that it could be found by the LLMs. Then there's the scenic route, the one for the humans, where you actually need an elevated experience, where you kind of can brand-validate what it is that you are shown in the LLM.

[00:03:56–00:04:16]
So whether you are buying something or you're finding information, you need to go to a browser and actually have this brand validation. This is the reason why we are designing two versions of everything right now. Also, super important here is that AI is a sparring partner. It's not a replacement for what we actually do.

[00:04:16–00:04:46]
We can also see a tendency to have this AI-fried brains, where people are losing their creativity. I just saw this new thing coming out — the people that work in marketing, 25% of them feel that their brains are completely fried by AI. And we don't want to do this. So I also saw someone in here last week doing a new brand design, standing with the bricks and having ink on the bricks and putting it

[00:04:46–00:05:10]
down on paper — and you can actually see this out here in the hallway when you walk through it. But that is creating brand design, not replacing it with AI. The other thing is that what we have seen is that people that have a deep craft also use AI in a different way. And if you are skilled with a deep craft, you can actually have the outputs of the AI become way better than if it's just me.

[00:05:10–00:05:31]
I'm educated in law. When I design something with AI, it looks bad — I can just say that. This is why I have two good people with me. Ha ha ha. Yes, also a different thing that we are seeing right now is that how we are designing design systems and how the design systems are being used are completely different as well.

[00:05:32–00:05:55]
So we still — if we are thinking about atomic design — we still do atoms, molecules, organisms. But when it comes to having a canvas and using these things for templates and actually designing something, I think that will actually change in a couple of years. So we will still do good brand design, we will still put it into design systems, but what we output is not in a canvas in Figma —

[00:05:55–00:06:18]
it's a different thing. It's probably way more coded than it used to be. Also, quite often we hear this: "We vibe-coded something, but can you please review it?" Because a lot of the organisations they are already doing these things. They are vibe-coding, they are outputting different tools, optimising personal workflows and so forth.

[00:06:18–00:06:48]
But a lot of people — they are afraid. Are we — are we — are we using data that we are not allowed to? And are we actually on brand with this? Is this actually fit for production? Can we actually put this out live? So quite often we are getting this call: "Can you please review what we have done, if we are breaking down any barriers?" And a lens that we have really talked a lot about —

[00:06:48–00:07:15]
and I think everyone is talking about — and this is what creates fear, this is what creates all the other negative feelings about AI, is that we need to be responsible in how we use it. So super aware of bias, yeah, energy, data, honesty — how is this actually outputted. So we have a full Morgenbooster on this from October '25, so not gonna touch on this, but it's filmed and you can view it on 1508.dk.

[00:07:17–00:07:37]
So just so that everyone is on the same page with the things that we're talking about today — what is vibe coding, actually? So what you can do is that you can now just describe what you want — as a tool, as a product, as an experience, as a service. Then you get AI to build it. It generates all the code.

[00:07:37–00:08:01]
You can actually just launch it, and then you iterate as well by talking to the LLMs, to the AI models. So you don't need to know how to code anymore. You can simply just output things. And this is what our two speakers are gonna talk about today. So an intro to Daniel and Nils, and a warm welcome.

[00:08:01–00:08:23]
Oh, sorry. Well, just gonna do the next one. So Daniel is working at 1508 as a designer, and I saw the first things co-designed with AI two and a half, maybe three years ago. So you were super excited back then.

[00:08:23–00:08:52]
So I guess how you feel today is super duper duper excited about things. And I'm curious to hear what you have to say. So after Daniel, Nils will join us. And Nils used to work at 1508, not working at 1508 anymore. But Nils is probably one of the most — yeah, concept-development-curious persons I've ever seen and experienced in real life.

[00:08:52–00:09:14]
You are here right now, and you are bringing a lot of examples of how to write code. So welcome to Daniel and Nils. Just gonna grab this from Oliver, and hopefully I can do this myself. Always a problem. Hi, yeah.

[00:09:14–00:09:43]
So — oh, it went backwards. Let's try that again, one more time. Sorry about that. Yeah, so I'm Daniel. I'm a senior designer here at 1508. I've been here for a while, but I've always been curious about digital design and design in general, and always been sort of tech-curious as well, mainly on sort of the coding and technology sides and that sort of thing.

[00:09:43–00:10:14]
But now also, as Oliver mentioned, a couple of years with AI, and trying to use that as a tool and not be too overwhelmed by it. But today I want to talk about something that's genuinely changed how I've begun doing design work. It's not just a tool that sort of saves me 20 minutes here and there and makes Oliver have to sell more hours to clients, but it's profoundly changed more of the way I think about design work and what I can actually achieve.

[00:10:14–00:10:43]
And that's what I'm here to share today, hopefully. So there is a bit of hype, but there's also maybe some skepticism and a slight bit of worry. I don't know about you guys, but my social media feeds are full of these two takes. One I've called the doom-and-gloom take — AI is coming for your job, it's obviously destroying the planet, and if you can't code you're replaceable, design as we know it as a craft is over.

[00:10:43–00:11:14]
That's one take on it. The other one is more the saviour type, right. We've got this — AI is optimising everything I do, I'm three times faster, I'm four times faster, I can output a lot more, I found the perfect AI flow, tech stack, and if you drop me a like or a comment on my post I'll send you a 97-page guide to how to do the same as I do and start your multi-million design business tomorrow.

[00:11:14–00:11:38]
I'm not really in either of those camps. I think both of them are sort of just filled with noise. And as Oliver mentioned, you sort of have to take your own stance on AI and be curious about it to actually form an opinion. And that's hopefully what I'm gonna share with you today. So today, more specifically and more personally —

[00:11:38–00:11:59]
something that's also a bit more useful for you guys. So one of the reasons I'm here is I want to start something, or start from a place that's more fundamental to being a designer, before we talk about tools and processes and all that stuff. Nils will definitely get into that.

[00:11:59–00:12:29]
But I guess you're in this room a bit like me here, that you design stuff because you can't help yourself. I — if I wasn't here at 1508 getting paid, I would probably, you know, be at home trying to make something by myself, for myself, for my friends, anyways. So I have this internal drive of coming up with ideas, and there's a certain discomfort inside of me if ideas don't get manifested or put into real life.

[00:12:29–00:12:50]
If you haven't done something about them. Oh. It it sort of urges me the wrong way. Last time — back in 2024 — almost two years ago, I also had a small primer talk about AI. That was like — in AI years, that's, I don't know, 300 years ago.

[00:12:50–00:13:13]
But I had this reference from Rick Rubin, the music producer. He has this great book about creativity and idea manifestation in general. And he has this quote that I used back then also: an idea needs to be felt, tasted, seen, or heard in order to become real. And he said it way better than I did. And I still feel — and I felt that way back then —

[00:13:13–00:13:36]
that this is where AI has some genuine power for me, at least. Making these ideas that we talk about tangible. He also said: an idea that lives only in your head is invisible to everyone else, but also honestly to yourself. You can't evaluate it, you can't share it, you can't build on it, you can't do much with it.

[00:13:36–00:13:57]
So ideas need to be visible. They need to be out there in the world somehow. But there is a gap. We have this great idea manifested in our head. It's full of potential, and we can see it clearly in our own little slideshow running in our minds.

[00:13:57–00:14:19]
But there is a gap to making it tangible — something that you can show to others, or you can talk about, or even present to a client. There's this gap in between: approvals, development budgets, politics, resources, queues, timelines, all that good stuff — scopes — and this is where ideas die.

[00:14:20–00:14:41]
They don't die because they're necessarily bad ideas. We don't even know that. But they're just exhausting enough to make real that we don't make them real. They get dropped. And then we don't know if they deserve to have a life of their own. And that's something I truly want to address today with the use of AI.

[00:14:41–00:15:02]
So I was trying to think of metaphors — oh — what is AI and how can I visualise it in my workflow. And I think you're all nerds like me, so obviously you know this is the Starship Enterprise from Star Trek. It has this fantastic warp drive that doesn't make things faster — it folds space, right. I mean, this all makes sense to you guys, I'm sure.

[00:15:02–00:15:26]
It folds space in front of the spacecraft and expands it behind it, so it can go really fast, and so it doesn't really go faster — it minimises the gap. I thought this was a perfect metaphor, but I I sort of realise now that it's probably only for me. But for me it was a nice visual, and I got to have both Rick Rubin and now some sci-fi Star Trek reference in my presentation.

[00:15:26–00:15:53]
So what I'm here to say is: don't go faster — fold space. In the sense that Oliver was talking about, we can output more. In my mind, it's not about outputting more — it's what you can output. It is what tasks you as a designer can now undertake. As a designer, AI is sort of collapsing that space between the idea that you have in your head and making it tangible.

[00:15:54–00:16:17]
And I think that might be a really big shift for us. And that's something we should gravitate to, if we want to be curious about AI and use it for good. Related to that — and this is genuinely new for me, at least within the last, I don't know, 18 months — the actual cost and effort of building software.

[00:16:17–00:16:40]
And I say software maybe a bit loosely, but smaller tools, stuff you can use personally, stuff that's maybe not too complex — as that bottleneck has almost completely disappeared. You don't — in terms of cost — you don't need a big development budget. The barriers to building are almost gone.

[00:16:41–00:17:03]
Speed, as we talked about. What used to take long development sprints or design sprints, you know, you can almost do iterations on that from day to day. And then there's the accessibility in it — it opens up new possibilities for people who might not have been able to perform or create stuff in that space.

[00:17:03–00:17:25]
And I think that's really exciting. One thing I sort of want to tie this on to is Kevin Finn. He's a former brand design director at Saatchi. And he's done a bunch of YouTube album covers — I don't know if you know them. But he has this great talk about what brands are in the 21st century.

[00:17:25–00:17:45]
And one thing he discusses about being a designer today is we can't just be these problem solvers, necessarily, for client briefs and problems. We have to be out there looking for problems. As designers, we have to take on new stuff in the world.

[00:17:46–00:18:07]
Even though that's what we've been trained to do — you know, wait for the brief, wait for the client, exploring that space — I think we have to flip it on its head and maybe go out there and explore. And I think AI is a great vessel for exploring. In summary, he said: the current climate demands we expand our value and thinking beyond only problem-solving for clients.

[00:18:07–00:18:31]
We need to explore our own initiatives and their problems too. I think that resonates really well with me. So as a designer, is also suited to serving needs and desires, to inventing things no one has thought of before, and going back to that — having that creative urge inside of me — that's something I think I've always done, but him putting it into words sort of resonated with me.

[00:18:31–00:18:54]
We've always been used to this sort of the brief in, and we output our work. There's the client's problem, there's the designer's solution. But AI makes those alternatives possible. I'm not saying one thing is better than the other — being a problem solver versus being somebody who's a proactive designer, so to say.

[00:18:54–00:19:15]
They both exist. We still need to do client work, we still need to listen to our clients and their briefs. But there's something about being proactive in that space that really resonates with me. So traditionally, the problem solver — we're waiting for the brief, waiting for the client, the budgets, estimates, and so on. Being a proactive creator — or open creator — it means go finding the problem.

[00:19:15–00:19:41]
It's something different. You've built something before somebody asked for it. Maybe you come up with your own ideas and then pitch it to a client or a team, or whatever that is. But it shouldn't be the only mode you operate in. You obviously have to switch back and forth between these two. I sort of needed to come up with a model, because that is what designers do, right.

[00:19:41–00:20:04]
We have our way of working here at 1508 — we have this great complex model. So I came up with my own model. It's a sort of double diamond. I don't know if you've seen it before. I feel it's completely original. I call it the "F around and find out" model. And basically what it does — it's sort of does all the good stuff we know, but just — yeah —

[00:20:05–00:20:30]
follow your curiosity. It opens up to following more of your own curiosity, maybe making things that you didn't need — you know — or that your client needed, or your team needed, or your organisation needed. And then being proactive and exploring without necessarily having a set destination or a set output. And then when you've sort of manifested it and you're finding out — see what clicked, what worked, what —

[00:20:30–00:20:51]
what is — you've actually made the thing tangible. So now you can evaluate. Evaluate it, and don't be afraid to kill it again. You were spending a lot less time on developing these prototypes or these ideas, so maybe killing them off isn't such a big risk anymore. And then refine it. If something works, then try shipping it and learn from that. Yeah.

[00:20:51–00:21:13]
So that's something that's been rattling around in my head on this. Actually goes back to the way I've always been working, but this is a more formalised way of working with AI and creativity, for me at least. So one of the things I did when I was getting into AI and what can I use this for —

[00:21:13–00:21:39]
I was looking at all these social media posts on this is how you connect your Figma to your cloud, and this is what you can use Midjourney for, or nano banana Pro hype something making UI interfaces. I sort of had to chunk it down and say: what is the smallest problem I can solve? So I didn't start with a big, complex client brief or a full product that I knew I wanted to do.

[00:21:39–00:22:00]
I found some small, tiny real-world problems and then I sort of built from there. And that gave me a lot of learnings. The goal wasn't to ship something perfect. It usually never is, but you try to get there. But the goal was to ship something — something again that was tangible, and that we could evaluate and I could learn from.

[00:22:01–00:22:21]
So that has sort of been my — when I test AI tools, I try to start with something really small and then build from there. And that also sort of feels the most natural, to be honest. One of the first experiences I had into vibe designing — if you can call it that — was this wonderful project we did with our client Dragon Heart.

[00:22:21–00:22:47]
We've been pretty good at exposing it to you guys, so I won't talk too much about it. But it was sort of the first place where I realised this potential of doing proactive work for the client before the client actually knew it. So we had these all these explorations in terms of illustration styles. We had some real-world references from the castle that we were working at, and we wanted to develop that into illustration styles.

[00:22:47–00:23:09]
And we were looking into how broad could this go, and what could it be and what couldn't it be. And that was really good. We were using Midjourney at this point to sort of just have something to evaluate. But as we were doing this and coming up with all these illustration styles or directions, we also sort of figured maybe we can maybe we can do some other elements.

[00:23:09–00:23:38]
So one of the things we were looking into was this drop cap alphabet illustration style, and we thought — oh — that could work really well with some of the products and some of the branding we're doing. But that's a big undertaking, doing all these illustrations for the complete alphabet. But you know, a vibe design out there, and pretty quickly we could have a full alphabet or typography system we could show the client — and they didn't know, didn't know that we were doing it.

[00:23:38–00:24:02]
They didn't know they wanted it, but it turned out really well. They were really happy about it. And it just sort of gave us another element to play with in terms of branding, but also in a sense that it sort of gave us new ideas that we could use it for. So these customised posters, for example. And that is where I sort of got into this proactive AI designer mode —

[00:24:02–00:24:25]
oh, I can make this sort of complete, realised idea and show it to a client, and show it to a team, and actually get some sign-off. Instead of me showing mood boards or explaining, or trying to do illustrations by hand that don't quite fit the whole thing. So that is sort of where I learned a lot, I feel, initially about vibe designing.

[00:24:26–00:24:53]
So from that I also have a sort of these three ways of using AI that I like to do things by when I get into vibe designing. The first is just plainly building for myself. It can be small tools, small tests for current client projects, small stuff I use in my daily design workflows.

[00:24:53–00:25:14]
It can also be building for something or someone that you know — maybe it lives outside of your organisation, or outside of your team, or outside of your agency, but maybe it's something you're you know something about, but you've maybe always felt there's a space here that you can really do something with as a designer.

[00:25:14–00:25:35]
And the third one is simply — you know — try and build for your dreams. If you have a specific client, or you wanna use dragons in your next design system, or anything — yeah, why not? Why not try that approach? So a few examples on how these routes can unfold. So — building for myself.

[00:25:35–00:25:59]
I always start with being really scared of the blank page when doing a new design. I always start looking at grids and typography, and specifically typography systems — how they scale and stuff like that. So I built this little tool. Instead of sitting in Figma, you know, moving a pixel here or there or moving a value here or there, I — together with Claude —

[00:25:59–00:26:23]
built this little tool where I can test out different typography scaling systems, line spacing, tracking, that sort of thing. And it also had a little grid system built into it, so I can test both things at the same time. And it started again with me just asking Claude: can we build this little type test system, and then add it slowly —

[00:26:23–00:26:46]
oh, what if we did grids, what if we did a dark mode, what if we did different viewports, so you have a desktop or tablet or mobile. And stuff like that. And I won't say I use it every day, but it's a good jumping-off point for me now. And it feels faster than again sitting around in Figma or some place else, and playing around with the —

[00:26:46–00:27:17]
I don't know about you guys, but when I like mood boards — my Figma files always end up looking like this. And it can be really hard to convey to other team members or others in the projects and clients — what is this thing and what is that thing. So I built again a little tool that is simply — you know — I can export some images from Figma, dump them into this tool, it can generate these different themes for me, come up with some keywords,

[00:27:17–00:27:39]
some analysis on them — very basic stuff — and then basically just output a little slide deck that is organised, that is sort of coherent, and I can then go back and edit and expand with my own thoughts and what I'd like to do. And it's got like target themes and a small few small tools, and I can export it as a PDF.

[00:27:39–00:28:03]
That's also really useful for me when you're sitting there looking at your big Figma file and you don't know what to do next. Also, client stuff. I think Oliver mentioned we were doing some brick experimentations on our — if you're here, she's our designer here, doing some actual physical stuff with some prints based on bricks.

[00:28:04–00:28:25]
But we're also doing digital design for the client, and I thought: okay, how can we take some of that tactileness and convert it into something digital? So I built this little tool where I'm testing different distressed lines and how they could be performed or exemplified in HTML or in code.

[00:28:25–00:28:48]
So there's like this little tool you can edit small stuff like the colours, the line width, the amount of noise, and all that good stuff. And the good thing about this — this is code — I can almost present it to a developer or — yeah — whoever else is connected to the problem, and say: this could actually be something. Instead of me just sitting in Figma or Photoshop and saying —

[00:28:48–00:29:14]
I've made this line, can you make it into code? It's actually almost there. So that is — you know — finding these small tools, these small opportunities for myself. What can I learn from this and what can I build? Another direction — build for something you know, someone you know. This is my kid, Ico. He's autistic and nonverbal. So really glad about these sunflower tags, where there's this card with personalised information.

[00:29:14–00:29:37]
It gives us — his parents — some peace of mind in case something happens to us. It's got contact information, but it's really not that practical when you're a 17-year-old, a 17-year-old boy who wants to climb trees and go to the beach and play football and all that good stuff. So I always had this thing about — could we make it into a bracelet, maybe with some sort of code or something? And you know, it's a good idea.

[00:29:37–00:30:01]
But I also thought: okay, what if we could build a prototype to show that idea? So that's sort of what I did. And just enough design to make it feel more real, feel more tangible again. There are two sections to it. There's one where you sort of set up the ID, gives you a brief explanation of what the bracelet actually is.

[00:30:01–00:30:22]
AI hasn't solved that for me yet — what the physical bracelet should actually be. But at least I've got this digital flow part to it, where you can set up your person. And there's even a small demo of — so it simulates somebody tagging or reading the tag inside the bracelet with information.

[00:30:22–00:30:42]
And again, it just gives us something to evaluate, something to talk about. And again, I don't have a client for this, but I think there might be somebody out there who could see the value in this. Final thing — you also have to build for your dreams. I like skateboarding.

[00:30:42–00:31:03]
Obviously I'm a very old man, so I resort to watching videos on YouTube. This is one of the most inspiring videos I've seen. This is Christian Floris, who has this stair set he's trying to flip down. I think it took him almost three years and like four hospital attempts.

[00:31:04–00:31:33]
And it's just like — oh man — that's super inspiring, that's really human perseverance and all that good stuff. But if you're an old man, or maybe even a young kid, trying to learn to skateboard, maybe there's a better way. And also, I had this idea — maybe there's a client here, potentially somebody like Nike, where you can find the old Nike Running Plus sensor that fit in your shoes, with a learning app and a sensor that sort of tells you —

[00:31:33–00:31:55]
you're doing this wrong with your board, or you're not jumping high enough, or you should concentrate on this trick before you move on to that trick. And if you've worked with me before, I've been talking about this for nine years. I have the name, I have the logo, I have all that good stuff. But I never had a sort of prototype of an app. So I built an app — very, very simple.

[00:31:55–00:32:18]
And it started with the principle of — I just want to do the first thing — try to simulate a recording of a skateboard flipping. So obviously it has to have these sensors on it. You can choose the trick you're trying to land, simulate starting the sensor, and — did I — I think maybe I reset it. Let's try that again — oh, it's just gonna play.

[00:32:18–00:32:40]
And it does that perfectly fine. And then I realised — I can build these other things onto it. So maybe there's a library of tricks, maybe there's your favourite skate spots, maybe it has a sort of session log of how good you did with that trick, or where you most hang out to do tricks. And there's all these neat ideas you can build onto it.

[00:32:40–00:33:02]
And I think that's a really powerful thing — starting off with something small and then building onto it. So those are sort of my three approaches to vibe designing. Obviously somebody has already gone and built the darn damn thing. I don't know — I should have waited these eight years. But I did, and so somebody has already built this sensor and an app.

[00:33:02–00:33:22]
And I feel not disheartened, but maybe there's a collaboration. I don't know. But so — don't let your ideas sit in the back of your mind or your Notion board and die. Go out and build them with AI before it's too late. So a few final things. AI generates, you curate.

[00:33:24–00:33:44]
And I think this both gets lost in all the doomsayers and all the saviours from — we're talking AI up. Your aesthetic sensibilities, your sense of what is right, what is off as a designer, what's missing, what's too much — that's not something a model has yet. I don't know if it's coming.

[00:33:44–00:34:05]
But I still see across all the tools I've been in touch with that there's still a need for designer sensibilities. And that's the stuff you've been developing your whole career, your whole life potentially — like the concerts you go to, the movies you see, the TV shows, the — during — when you go on holidays — all this stuff helps influence you.

[00:34:05–00:34:28]
And they can't influence an AI. So be mindful of that. AI doesn't replace that. It hopefully extends your talent, your sensibilities, your reach. And I think that's potentially a good thing. Yeah. I don't — I hear this a lot — that designers are cooked, and you know, everybody has the potential to be a designer now.

[00:34:28–00:34:50]
I think it's the other way around. I think we as designers have the potential to take over a lot of industries, a lot of fields. We know all the hard stuff — we know the sensitivities and aesthetics and what works and what doesn't work. All the hard stuff about technicalities, or — you know — I don't know — Oliver, how to run a business, or law — all that stuff, you know — LLMs are really good at that.

[00:34:50–00:35:12]
So let's, you know, let's take over that space. That's what we can do. That's the superpower of AI. And it has, you know, lowered the cost and dependencies on other people. So go do what you think AI needs doing, or the design world, or the world in general needs. Out there, ideas deserve to become real.

[00:35:12–00:35:36]
For too long there's been this gap between the tangible and the idea. Hopefully your creative instincts haven't changed. But now you have a tool to keep up with your imagination. With all that being said, let's look at some real-world tools. I think I'll hand it over to Nils. So you can — please.

[00:35:38–00:36:00]
Repeat the sentence, strictly lower case, without punctuation and text formatting. OK, hi everyone. My name is Nils. I'm a partner at Heyra, where we do AI orchestration systems. Sounds very advanced, but basically we just automate engineering work.

[00:36:01–00:36:21]
I've been building services, products for the last 13 years. I see some old former colleagues sitting here as well. And I'm gonna tell you a little bit about all the stuff I do, how I do it, and how you might also do it —

[00:36:21–00:36:46]
not just you as an individual, but also something you can take back and discuss in your organisations. One thing about me is I'm not a very patient person. So for the last 10 years — maybe 12 years — I've been really I have this anxious thing in me.

[00:36:46–00:37:06]
I hate meetings, I hate talking. I just wanna go out and just build the thing, right. But the thing is, with so in the last year or so, after LLMs got really good at coding — especially with Anthropic's Opus 4.5 —

[00:37:06–00:37:28]
things really started to shift for me. So all of a sudden I didn't have to sit in meetings, I didn't have to engage developers to build for me. I could just go out and just do stuff. So like all of these things I've had in my head for the last 10 years — I could just go out and build. So I built around 20 tools — all kinds of weird stuff.

[00:37:28–00:37:51]
Bedtime stories for adults and all kinds of weird things. This is just — excuse me — it's small, but I built it all in HTML, because I don't like working in PowerPoint or Keynote. This is how it was before, and maybe in some organisations it's still like this.

[00:37:51–00:38:14]
Building — I was going to say "was" — but it still is slow and it can be expensive if you do it like this. So often it would be like: you have an idea, you discuss it a little bit, then you sketch it out maybe on paper, you go into Figma, you start drawing boxes, and spend your entire work day in that tool.

[00:38:14–00:38:35]
You show it to some users, they click around — it's not real, but you know, kind of real but not really real — and they say cool. And you might decide: okay, let's do it — the users liked it, ish. Maybe we don't know because it wasn't really real. But let's try anyway. You go out, you need some developers. Oh, developers —

[00:38:35–00:38:57]
now we need a budget, now we need a budget. Oh, we also need a timeline now. Now it's getting advanced. Let's have a meeting. So we have a meeting. We discuss the budget allocations, time. Then we have another meeting, another meeting — and that really cool idea slowly dies. And I think that's very sad.

[00:38:58–00:39:28]
But it's just how it was. We accepted this because — yeah — it is what it is. Then with LLMs coming out — and especially Opus 4.5, Opus 4.6 — the cost of building suddenly collapsed, because the cost of doing code was reduced drastically, because the models got so good at coding that you could actually just start building.

[00:39:28–00:39:57]
You didn't need a developer anymore, you didn't need a timeline anymore. So this is how I see the world today, and this is the world I live in — where building is cheap and it's fast. And it is because we can take out a lot of steps of the old processes. So now — and this is the way I work — I have an idea, I describe it, and I start building it immediately with AI as my agents and my assistants and my sparring partners.

[00:39:58–00:40:20]
And then I ship it. And I've never written a line of code. I could do some HTML and CSS before, but now I can ship fully functional products. So that's awesome for me. So I just started building. And the way I build is I look at friction, I look at problems, and I don't like to complain about them.

[00:40:20–00:40:45]
I just take them and see if I can solve them. So what I'm gonna show you now is just me trying to fix my own problems. I haven't done any market research, I don't do competitor analysis. I just — this fucking thing annoys me so much, I'm gonna build something that solves it for me. And one of the very first things I did was this thing called Unsub.

[00:40:45–00:41:07]
And it was basically because I was so tired of having a red icon on my mail inbox which showed like 10,000 unread emails. And what I discovered was most of those emails — because I don't even read them — are newsletters, promo codes, products I signed up for, never unsubscribed from again. And then in Apple they have — oh —

[00:41:07–00:41:28]
you can unsubscribe from this specific thing. But I just wanted an overview of all the stuff I've subscribed to over the years, and just do like unsubscribe everything. So that was basically what I built. I had an idea in mind — it's like: connect to my Gmail, scan for these newsletters, and one big wipe.

[00:41:28–00:41:49]
And that was my initial kind of concept for it. And then I sat down and I built this. Also, I always do small videos, so you — you go in — this is the landing page I built. I log in, I connect to my Google. It's not verified yet, so Google comes up with red stuff.

[00:41:49–00:42:11]
But it's just for me, so I don't care. It scans, gets all of the newsletters I've signed up for. I can go in and unsubscribe over here. I unsubscribe from life, from Turbo Script. I can also select all and unsubscribe from them all in bulk.

[00:42:12–00:42:35]
So this was kind of the first — one of the first things — and it worked. I still have that red thing, so I just went in and — it doesn't show on my phone anymore, at least. And I don't get any more newsletters. But yeah, I couldn't do like "read all mails", because maybe there were some mails that were important that would — going that. And then when you start building, you can start adding on.

[00:42:35–00:43:02]
So I had a new feature in mind where I could — okay, can I connect this to my bank account and get it to scan my bank account for subscriptions I have, and how much I pay? So I made this small tool for that. This is my subscriptions.

[00:43:02–00:43:26]
I upload an Excel sheet from my bank statements, it analyses it, and gives me a full view of how much I'm paying. I'm paying around 10,000 a year. I can go in, I can go directly to my and manage my subscription in here and discontinue if I want to.

[00:43:26–00:43:48]
So this is actually — I start with something simple, then I build something extra onto that. I tried to connect it via APIs to my bank account, but the banks in Denmark are not very happy to let you own your data. So I have to go in and get an Excel sheet. I think I tried five different APIs for open banking services, but every time I had to be vetted by the financial authority.

[00:43:48–00:44:08]
So I would have made it smoother by having APIs, because I could do everything automatically. So this is kind of not one of the things I'm most proud about, because you have that manual step of going in, getting your Excel sheet. But because the banking system is built like it is, I couldn't do anything else. Then I did — no, wait —

[00:44:08–00:44:39]
maybe I'll skip this. I don't have so much time, but basically it was — I can see it — it was an overlay on Ekstra Bladet, politics and person. I did it because I'm tired of these stupid — so I put a layer on top to just make it a news feed. And basically what it does is — basically what it does is it takes the initial headline that makes you want to click, but I look at the article behind, and also take the manchet.

[00:44:39–00:45:00]
And manchet often just describes what the article is about. So here it's — the former NFL star Matt Khalil has sued his ex-wife Haley for speaking about his penis. And Haley Khalil's lawyer says that his client is just using his freedom of speech. So okay — is this an article I wanna click on? No.

[00:45:00–00:45:26]
Yes. So that was just basically a layer I built on top of that. Then I did something about voicemails. I got so tired of getting a message from Trey — hey, you have a voicemail — doesn't say anything else, just "you have a voicemail". So I have to call up my voicemail, go through these old-school menus of "press 1 for this, press 2 for that, press 3" — and then I have to listen to the voicemail and figure out what is this about, what does this person want from me.

[00:45:26–00:45:49]
So easily I've spent five minutes — or several minutes at least — trying to digest a simple voice message. So I thought: okay, I'm gonna build my own voice messaging service, where people can put in a voice message. I'll send that message to a transcription service. I get OpenAI to look at it, and send me an SMS back with what this message is about.

[00:45:49–00:46:12]
It's pretty simple. This is how that worked. Hi, Bill here. What's up? Hi, what's up. This is Brian from Microsoft IT support, and it's very urgent that you call me back. You need to solve the hacker on your computer and I will help you remove him. Yeah. So this was like a Brian from Microsoft.

[00:46:12–00:46:32]
I get a message. The message says what it's about. It's actually warning me, because it sounded like a scam attempt from Microsoft. And because OpenAI is pretty smart, or — the LLMs are pretty smart at digesting and reading the message — it actually can put in warnings about that specific message.

[00:46:33–00:46:53]
As you can see here, here it says a potential scam attempt was detected — high risk. Unsolicited urgent request from an unknown sender. Request to perform remote desktop access, misidentification from Microsoft. So this is also things you can start doing and building into it.

[00:46:53–00:47:17]
So I actually just made a smart voice messaging service. And then my girlfriend found out that she could send me messages through that. So she started sending me shopping lists and stuff like that, because I didn't pick up the phone. Explain — this was something I did with Anja from Sundhed.dk, just to give you an example of how you can also utilise this to show concepts that are actually working.

[00:47:17–00:47:39]
So for Sundhed.dk, we built this small thing. And health journals in Sundhed.dk are written by doctors, medical staff — it's written in a weird language you don't really understand. And their idea was that we could make an AI tell me what my journal was about.

[00:47:39–00:48:03]
So we built this small feature in here at Sundhed.dk. Very simple AI explanation of what is actually being said here in this journal. And Anja could take this back to her boss, say: it works. We have to figure out how we can send data — maybe we need to have some local LLM so we don't send personal information into the system.

[00:48:04–00:48:28]
And what happened was — after we built this, it took maybe 30 minutes to build — Anja and her whole team at Sundhed.dk got access to Cursor, which we used to build this, and are now doing this kind of prototyping in Sundhed.dk — at least using it — before they mocked it all up in Figma and clicked and — oh, maybe we could do a drag and drop — and huge mess.

[00:48:30–00:48:50]
Then I built a Danish language tool that's really good at Danish, because me and Adam — my good friend and colleague — do lots of videos, and we also do videos in Danish. And we want a good Danish voice actor. Voice actors are expensive. It's a hassle — you send them a script, they send something back, it's not really good, you go back and forth.

[00:48:50–00:49:13]
It can be very expensive when you produce a film. So we made a tool that is pretty good at Danish. He told me to really put up the new — OK, last — Braunau bitte. Ein Stimme pro in Stimmenke war heiltet po.

[00:49:13–00:49:33]
This is all AI, by the way. Set at denisten feels for Kate som en tanke der ik var mening du sku høre. Den ka viske forsigtigt, langsomt, næsten kærligt. Eller den ka træk på et smil på dig, som om den vil noget du ik ved endnu.

[00:49:34–00:49:58]
He he, den ka lige lidt. Den ka os bli stille og tænksom, som lys der falder ind over et tomt gulv. Som noget der føles større end ordene egentlig er. Den ka få rum omkring sig, luft. Som andre står og ligner i en stor sal og laver et ord hængeligt.

[00:49:58–00:50:30]
Den ka komme ned og fra som geben vand, langsomt, tømt, som noget der på vej op til overfladen. Det ka lyde som et sinnet noget opfanget og sindt noget der knittere, noget der kæmper for, og noget igen. So this was a tool we built for ourselves. We go in, we dump our script, we have some different audio tags where we can kind of get the voice to whisper,

[00:50:30–00:50:57]
as it did a little bit, and so forth. This is a really powerful tool for us. And I also — there was no Morgenbooster here today, but I made a small commercial. This is another voice we have, with one of our male voices. Nogen dag begyndte den i duften af brød der endnu bærer åndens varme, især skoven, gi'r efter med den der lille sprødlige — du kender den.

[00:50:58–00:51:20]
Smør og smeltede langsom, kaffen dammen, lyset står lavt i vinduet. AB plus die Vögeln werden in Lillesmühlen Müller mehr Menschen. Frisch gebackt morgen früh. Boom, salt in the salt — hello there.

[00:51:22–00:51:54]
So this is just, you know, an example of it. I don't have so much time, so maybe I'll just skip. And gonna tell you a little bit about how I built. Skip this one — yeah, Slides. It's actually what I built all of these slides in, because Oliver told me — and then you sent me a Keynote — "hey, we have a shared Keynote, can you go" — and I was like: fuck this — let me just dump my storyline into I build — I build like the framework I call Slides.

[00:51:54–00:52:18]
It's a design system I made for slides specifically, and I just dump my storyline into it and it produces the slides for me, basically. So all of these slides are produced by an AI. All of this — it's also HTML code, so it's crisp, it's super sharp, easy to share, and yeah, it just works. This is basically how it works. And you get the presentation after.

[00:52:18–00:52:41]
If you wanna do a presentation like this, you can always just clone the GitHub repo into your cloud code or whatever tool you wanna use, and you can start building slides like me. And so — how the fuck did one person do all that? And I have a very structured way, actually, of working with it. This is kind of the old way of working.

[00:52:41–00:53:03]
We call it big design up front. It's basically waterfall — full requirements, full design, full Figma, everything is laid out, now we begin. It's very slow. But it's high quality when it ships — if it ever ships. And then you have the new vibe coding over here — no design up front. It's basically just you prompting and wishing for the best.

[00:53:03–00:53:27]
Then we have the middle ground, which I'm a big fan of. What we call "just enough design up front". It's actually a framework — you can go search on Google. But it's more like you think about what you wanna build before you build it. You use an agent as your sparring partner to make a plan, and then go from there. So it's: define must, should, could, won't — speak with your agent — build, iterate, iterate a lot, and then refine and ship.

[00:53:29–00:53:49]
And it looks like this when I do it. This is just me rebuilding on Sub. So it starts with me as a prompt. It's fairly long. It defines the user journey I wanna create. It defines how I wanna build it as a web app. I wanna use Tailwind CSS and Shadcn for UI components. I don't want it to focus on design. I just want it to work.

[00:53:49–00:54:10]
I focus on getting the system to work first, then I can make it pretty afterwards. Then we start making the plan. One very important thing is the plan you get — you put plan mode on. We also have plan mode in Lovable — most of these IDEs have plan mode, because that's how the agent works today.

[00:54:10–00:54:31]
It creates the plan, you read it. It also scopes everything out into to-dos or tickets. And so scaffold next year's project, set up Google authentication, build Gmail API. So it basically just makes the plan for you. You read it through — does it make sense? Yes — we build. And this is just enough design up front.

[00:54:31–00:54:54]
It's unstyled HTML. Times New Roman — it doesn't look pretty. But I don't care. What I care about is: when I press this button, I get a Google authentication and it scans my inbox. And I see my inbox and I can select the newsletter I wanna subscribe for. And it works. And when it works, I'm happy.

[00:54:54–00:55:14]
Then I do the design part. And then I start doing lots of refinement, lots of — okay, make it pretty, I wanna do it like simple, minimalist, whatever. And then you go on and put on more design. But when you start building, just focus on getting the thing to work. Just enough design up front.

[00:55:15–00:55:39]
So yeah, it's the one specification that drives it all. The more you build, the more accumulated knowledge you have in your GitHub repo, for example — the more the agents will start drawing from your code base, your history, standards, MD files, and also your taste. It knows me and my design systems, how I want things to be.

[00:55:39–00:56:12]
And then it can actually help me just plan it — agent drafts the build, I approve it, it executes the build with me iterating together with it. And it also helps me ship real, playable products in a few hours. And it scales to organisations too. That's what we do in Heyra, where we use these kind of — speaking with the agents, getting it out into Jira or Linear or whatever task management system you use, and then orchestrate that engineering layer.

[00:56:12–00:56:33]
We're doing it primarily with data engineers today, but we're also looking at doing it with front-end engineers and so forth in the future. So basically it is: we spec things, we plan it, the agent sends all the tickets in the plan into Jira, so it's very visible for everyone. It can help you — it can help you execute the ticket.

[00:56:33–00:56:56]
You can say: agent, go do this. I can do it myself, but I will always get an agent to do it. It builds, and then it creates a pull request to you. And you as a human look through it, and most of the times it's good. And then you deploy. So we have a pretty good case from Novo Nordisk Foundation, actually, where we increased developer output with a multiple of 7x.

[00:56:56–00:57:19]
So they had — before they had this agent system helping them — they could do roughly five use cases a week, each engineer. Now they do 35 a week, each engineer. So you can increase your productivity incredibly by implementing these kind of agent layers. It works in your stack.

[00:57:19–00:57:43]
Don't need new software to do this. It's just a layer on top. And then just the last part — I have one minute — shit. AI builds what you ask it for. So there's still a lot of — you know — you still have a lot of responsibility as the creator of something.

[00:57:43–00:58:05]
AI doesn't know your users. AI doesn't have taste. AI doesn't have feelings. AI can solve the wrong problem if you guide it towards that wrong problem. So that's all on you still. You still have to set the direction. You still have to make the final judgment. You still have to — you know —

[00:58:06–??:??:??]
feel if this feels the way you want. Because the AI doesn't feel anything. Oh, shit. It's not a tool I've built, this one. And so yeah, the tools change. This thing is still ours — still the human kind of — okay, does this do anything good for me? And so yeah, thanks.