Watch the recording of the talk here
[00:00] Hey. I was going to start with a hand-scaling. How many people here are using AI every day? Quite a lot. How many people here feel they really master AI? Yeah, that wasn't that many. How many people here can't do their job without AI? Yeah, there's actually someone. And then to the big question: how many can see a direct effect on the bottom line because of AI? Yeah, that's a couple of. Interesting. My name is Magne, and I am talking too much at the moment with Claude – at least if you believe the Evening Post. I'm not here to teach you about online commerce. You can. I'm here to talk about what AI can mean for online commerce. And I'll try to do it without contributing to the hype.
[01:20] I'm a potato myself. I'm an artist and I've been at a cook school in Paris, but I've been running The Case for 15 years, and there I had every time a big team of designers, project managers and a whole bunch of developers. Until daily I build a lot of AI, explore and try to keep up with AI and help companies understand this AI drive we're on. But now I work alone. I don't have 60 employees to help me. I have to do everything myself. And it's almost become a kind of obsession for me to figure out how I can do things myself, while maintaining a certain kind of quality, and with a little more pragmatism than I had to have before.
[02:21] So I spend most of my time building with AI, code, design and writing. In a relatively short time I have built a full application, two websites and a mobile app. It has built a CRM system for me and an analysis tool. And the best thing is that these talk together, and Claude has access to everyone. It's a really great clock when it works across the systems. And then I write a lot, and along with AI, it's become both more fun and more effective, I think. But it doesn't write for me. The most common error I see when people use AI is that they really just use AI as a text generator: "write this for me" -- and then it hammers out the cross-section of the entire Internet.
[03:26] It becomes unidentifiable, and it's pretty easy to see when something is directly out of AI. Think of it as a writing process. When I'm working on an idea for an article, for example, I'm talking to AI, and it's very good at helping me dig in my head, giving me resistance and challenge my perspectives – which then turns into the next step, a brief with knowledge of my target group and my context. With such a process, I might get to 80-90% in some cases. And then I'm the one who completes and ensures that it's the way I want it to be. It's becoming more of a editorial process, and it's still mine. It had not managed to do it without the input from me – and that's the key here. Or this presentation.
[04:34] It's made with AI. Claude Design is coming Friday – it's completely king. Or this voice synced voice prompt app that we put yesterday. The fact that it is possible to build small, custom tools for my workflow is very cool and liberating – and not least cheaper than buying all possible applications. Before, it would be completely unrealistic for me to prioritize doing those things there. So, Al alone is not magic. I still have to take responsibility for the direction, quality and context I give it. But if one person with AI can do more than before -- what does that mean for an e-commerce team? What does that mean for management? For IT? Marketed? Product? Or the way you interact with an AI agent, maybe in the long run? A lot of what specialists do today, AI is going to be able to do for us.
[05:59] It can code, calculate, write, design, analyze – and the list only gets longer and longer. And it gets much better very fast. It's not more than a few months since most developers would be quite sceptical that AI took over part of the code writing. Now they say that approx. 80 % of Claude is written by Claude. It's all coke. And it's happened now. So I think we're going to go to a world where you don't necessarily need an expert -- you need conductors, someone who can express needs, understand customers, see patterns across and connect things. Last year there was a lot of talk about an AI bubble that hasn't yet come. And a lot of talk about getting started with AI, but didn't see any value on the bottom line. But is it really so nice?
[07:20] AI is not a new app that changes things from one day to another. It's not heavy Photoshop that's being replaced by Canvas. It's a new way of working, a new way of thinking, and a new muscle that needs to be trained and practiced. Right now it feels like old working methods may not work completely as they did, while new working methods are still a little too new. And then you see that everyone else gets to things with AI, and there's a lot of AI-drit published, and it's not that easy to navigate. But it's not necessarily a sign that AI doesn't work -- it's a sign that AI is new, and it takes time to master.
[08:13] It won't be productive on the day you buy the tools. It doesn't keep anyone saying "please, now you've got access to ChatGPT". It becomes productive when the organisation learns this together and has a relationship with how it is to be used in the enterprise. As far as AI and web commerce are concerned, I think it's about having two thoughts in my head at the same time. My LinkedIn feed is full of AI magic. I did a YouTube search on AI and online commerce to see what's happening there -- and it also looked like a magic show. But AI is rarely magic, especially if you're doing it well. In the short term there are lots of possibilities: AI can do routine tasks much faster, produce more product texts, SEO campaigns and so on.
[09:22] But in the long term, the danger is that the short-term space of opportunity shadows the long-term threats. My hypothesis is that it's going to get harder – not because we don't have good and smart tools, but because everyone else is going to have the good and smart tools. I think the Internet is going to be overloaded by noise. I think it will be harder for me as a consumer to separate noise from what is relevant. And I think it's getting harder for online shops and brands to stand out – unless you do like more of the other speakers here: focus on community, quality, history -- what makes a unique. So what's experienced as simple today can be harder tomorrow.
[10]:23] Slowly, but certainly the floor is raised for everyone. I think it'll be easier to run an online store, but harder to convert. And even if it becomes easier to produce more, the winners are going to be those who use AI to become more valuable to choose. AI competence – we might be a bit behind here in Norway – becomes a basic competence that is expected when businesses employ new people. But to get there, we have to go through a slightly chronous learning trip. Generic content and misguided experiments are part of that learning trip, because most of us are actually just beginners on this. So we have to learn -- and luckily, it's never been more accessible and easier to learn these things.
[11]:35] There are plenty of content that is actually good on the Internet. But we'd have to do that. And that's the two-part responsibility: as an employee, you have to learn, and then companies have a responsibility to facilitate that learning. The danger of not doing so – neither as an employee nor as a business – is that both of us are potentially irrelevant in the shift we are going through. It's coming faster than we think. AI doesn't mean the same for everyone in this room; you have different roles and tasks. So figure out what that means to me and my role. How does it fit into the company? And for the leaders -- I think a leader has an extra big and complicated responsibility right now.
[12] On the one hand, you have to create a room for learning in a rather busy everyday life. On the other hand, there are many leaders who see that the business models are potentially turned around a little bit – so you should have quite a lot of thoughts in your head at once. I talk a lot about strategic context that an AI can work with. It's a leadership responsibility to build it. Here's an example of when I've made it available to Claude: then it knows what I'm working on, what leads I have, what I'm going to do tomorrow – and it's available to it at any time. There's something quite different than starting a conversation with an AI from scratch. IT's job is not to say no to AI.
[13]:41] It's making safe rooms where AI can actually be used. And then I think it's super important that you make your systems available for an AI to consume. One example: a large group saying "with us only Microsoft is applicable, here only Copilot is involved". Maybe there's someone here? Unfortunately, I do not think Copilot is something very good – or at least I have not. Many employees get a bad experience if there is the first meeting with AI: "What was this for something?" It's not just a shame – it prevents the muscle we have to build, that exercise. So the fact that government or IT stops people from learning is potentially the most expensive decision IT can make now. AI makes it easy to make a lot of things – therefore quality, taste and perhaps even restraint are important.
[14:58] For those of us who are engaged in marketing, content and design, AI is already a mind-in-the-sickly useful tool. I have an example from an agency we're working with that's really gone into understanding AI. They had a customer who got 16 types of trampolines from China. All they had was a bad picture. This was supposed to be a campaign – but there are snow and snow and winter in Norway. How do you get these pictures? Before they had to wait for spring, build the location and so on. Now they can do it with AI. All they needed was that product image. And this can also change – you can switch models, switch scenes and so on.
[16:01] It starts to approach production quality — and if one, two, three months, it is definitely production quality in these things. And then to what I might burn most for: an AI model takes what you give it and enhances it. But you can't multiply anything by zero. And that's what most of us do -- we start blank when we talk to an AI. It looks like 70-80% of what we're actually putting into a prompt is about explaining the AI what we want and who we are. The best AI tools in the world are just cross-sections of the Internet, and they don't do anything good without getting perspective, context and input from you.
[16:56] Or what's even worse: you get something that looks good, but that's completely paddeflatten. And it's not AI that doesn't work -- it's the absence of identity. After a lot of testing and error, I have split it into these five categories: the strategic foundation – the context – must be in place. Your perspective: what do you mean? What does their world look like? What signals do you pick up on customer service, in the market, on return and sales? What experience have you done that is not on a dashboard? And the intention: what is this going to do, for whom, right now? If the precedent is that AI has come to be – how do we make sure to capture these things and re-feed it as context? If I'm going to share one tip today: Start making yourself understandable to AI.
[18] Because when AI gets these inputs and the context that's about you, it stops being a bad-ass Google search engine that's mostly used to get dinner recipes – and becomes something you can actually use as a real sparring partner. For those who are curious and willing to spend some time, I have uploaded seven-eight workflows that you can download and dump into their AI model, so it will interview you and document it in MD files that you can reuse when you are about to talk to an AI. A hot tip: Use something that Whisper Flow or anything else you can talk to. It's a little unusual at first, but it's insanely useful.
[19:13] For those who are extra interested, you can take a look at the instructions contained in those files – there is also a lot to learn. I don't work with online commerce for daily, but if I were to start a online store today, how would I attack it? I would start by building my strategic context: who are we? Who are our customers? What are we doing here? Then I would do a pretty big job of providing the products metadata that are available and readable for an AI. This product has largely a title, metadata and maybe a SKU - but who is it? What does that solve? How does it look?
[20:08] At the webshop itself, it might be good to use Shopify and so on. But my marketing surfaces – the marketing website, the story – would I build with Claude Code. Claude Code can have access to everything. I wanted to build a writing system for marketing from day one – not a prompt, but a system. And then I wanted to change the way I work with design and photography. Those who haven't tried Claude Design: it's quite exciting. And -- very important, and maybe a little bit heavy because I recognize that many people here sit on older systems -- that all systems are available to the AI models. To read the analyses, have access to the order system. Not just use it to generate, but to understand. There's a lot to go into, I also recognize that.
[21:21] I want you to have room to learn this in a busy everyday life. But what's certain is that there are now new online shops that work this way. It is the competitive image you must relate to – or become part of. Thank you for me.