Today’s episode will be about AI. But not the ChatGPT kind of AI, but the building agents kind of AI. We’ll talk about what Alfred has been cooking over the last months, what FAFO really means (not just what the acronym stands for), and about this new kid on the AI block: OpenClaw.

If you want to understand where AI-powered marketing is actually heading, not the LinkedIn hype, but the messy, exciting reality of building stuff every single day, this is the conversation.

Check out these helpful links

Repo: https://github.com/Pitcocy/niklas-podcast
The OpenClaw guy if you want to get inspired: https://x.com/AlexFinn
YouTube video with some extra context about setup: https://www.youtube.com/watch?v=3GrG-dOmrLU
Use cases: https://www.youtube.com/watch?v=Q7r–i9lLck
OpenClaw website: https://openclaw.ai/
Their link to the repo, that you can use to feed to Claude during setup: https://github.com/openclaw/openclaw
Alfred on X: https://x.com/AlfredSimon
Alfred on YouTube: https://www.youtube.com/@alfredsimonai

Check out the full episode

Check out the episode on YouTube, Spotify, or Apple Podcasts.

Auto-generated transcript

Niklas Buschner (00:02.104)
Today’s episode will be about AI. But not the chat GPT kind of AI, but the building agents kind of AI. We’ll talk about what Alfred has been cooking over the last month, what FAFO really means, not just what the acronym stands for, and about this new kid on the AI block that’s called OpenClaw. If you want to understand where AI powered marketing is actually heading, not the LinkedIn hype, but the messy, exciting reality of

building stuff every single day. This is the conversation. That being said, welcome to the podcast Alfred.

Alfred Simon (00:39.016)
Thank you very much, Niklas, for the invitation and happy to be here and talk about one of the topics that nobody talks about, AI.

Niklas Buschner (00:47.566)
Yeah, I also feel like we have uncovered a gap here in especially LinkedIn content with AI, but Yeah jokes aside. Thanks so much for taking the time for people to better understand why you are a perfect person to listen to when it’s about this topic. Can you share a little bit of context about your role and what you do basically day by day?

Alfred Simon (00:55.603)
Yeah.

Alfred Simon (01:14.526)
Yeah, of course. my job position on paper is stated AI architect. That’s the official and nice name that you can also use in a more normal ways. But I call myself a FAFO engineer. That’s F around and find out. Mostly my job is at the agency. So I work at the Dutch agency called Advice. It’s a digital marketing agency. And my role is to help our clients integrate AI in a way that…

makes sense and it’s also useful and possible. And also at the same way, other 50 % is to help the agency itself. So our teams and services, how we can make something faster, how we can do it with AI. it’s a lot of experimenting, implementing things and not just playing around, but we also have a lot of already running systems in production.

either at us or either at some of our clients. So yeah, basically that in a nutshell, but if there’s anything else you want to know, just feel free to ask.

Niklas Buschner (02:22.626)
Of I always loved these videos on YouTube where it’s like day in the life of a, I remember, software engineer at Google. Day in the life of even, I don’t know, investment banker because just you sometimes you generally understand what these people do, but you don’t have like a clear, like visual manifestation of how their day actually looks like. So can you take us through a typical day, like one day like AI architect at Edwise?

Alfred Simon (02:34.301)
Mm-hmm.

Alfred Simon (02:52.475)
Yeah, I can. I start really early. So my wake up is around 4.45, 4.50. Depends a bit on when the dog wakes up. And then I just quickly go out, take the dog to a tour and then basically after breakfast, I go straight away to work. What’s exciting, but also at the same time challenging about this role is that I really don’t have to.

days that are the same. it’s hard to find like a, I have obviously, I try to have a written where I work in the morning. I have two or three hours focused work, but the topics projects I work on, it’s constantly changing. it’s, it’s, it’s really, sometimes I don’t even know like Monday, what’s going to come up on Wednesday. So one day can be, for example, if I take a typical day, for example, yesterday, I was three hours at the client talking about their

new plans for how we can integrate AI. We already made a proof of concept for them and we tested it. And now we had to present that proof of concept with the client himself to his team and to his management. And we had to prove basically, okay, this is what we’ve been building in the past couple of weeks. It’s working. We made a demo for them and then yeah, we just had to present it. And now we are waiting for approval. So he is waiting for approval, the person that’s

building that inside the company so we can take it further because now it’s just a proof of concept. So that was in the morning. And then in the afternoon, I had to finish up something. We are working on a reporting agent inside the agency. I took the agent to a certain level and we had to take it to deploy it to a Docker server so we can run certain things also on the server side. My colleague, which is a

lot more technical than I am, finished up part of it and then I finished that in the afternoon. And then the rest of the day, they was just looking for some issues and bugs because one thing nobody talks about AI system that are in production that they’re going to break a lot of times. And the more system you have running, the more systems you need to maintain and fix. since we have at the agency right now,

Alfred Simon (05:14.557)
14 active systems that are running in certain ways. Some are running in NA 10, some are running outside NA 10. There’s always something that’s broken. Somebody always sends a slack with an image that’s not working. then, yeah, usually I spend like one hour a day to try to fix those as much as I can. And then if it’s a sport day, I go to sport and after that I do some work in the evening and basically that’s it.

Yeah, mostly it’s like 10, 13 hours of work a day about, and then learning about AI and then trying projects and things like that.

Niklas Buschner (05:49.795)
And what would you say, why do AI agents break so often? And do you feel like they break more often than traditional software products pre-AI era?

Alfred Simon (06:00.702)
Not really. So software wise, they don’t really break. It’s more like the response they give. It’s not always what’s expected. What I see from our examples is like 40 % is user are not using it the way it was intended to be used, which is with every software, I guess. no, but people need to get used to how to prompt. So most of the times what I see is that

colleagues put something into it and then they get something, the way they asked for it or the way the data that they gave to the AI, was that. AI had no chance to give a correct answer. And the problem is even if you set it into the system from these systems try to help in any way possible. So they try to come up with an answer. Sometimes it’s wrong and then that frustrates people. Other things are just yeah, timeout errors, something breaking in production. 30 % of the errors is probably just…

me making a mistake when I created it because yeah, these are, most of them are, I would say, wipe coded, written in tension. So I, I’m wipe coding in a way I know exactly each file, what needs to do in the project and what the folders are and database. So I understand the project is not completely wipe coded, but obviously the code, 90 % of the code is written by us. So something we break here there. So yeah, you also have to fix that.

Niklas Buschner (07:27.342)
And can you share an example of a product or an agent etc you build maybe some that is best for people also to to understand what’s possible so probably not the most complicated thing you’ve built but maybe something that’s just Yeah, very very easy to imagine also how it’s working out in a in a day-to-day either at the agency or at clients

Alfred Simon (07:53.969)
Yeah. Yeah, I can share an example from our agency. I assume, yeah, you listeners are also digital marketers, so it’s going to be easy to understand. The social team needed a quick tool where they can A-B test certain creatives. And just making small changes to the image itself. So when the design team from the client or our design team creates a creative, sometimes you want to test with like, I don’t know, the…

button on the image or something on the image changes from green to blue and then just small changes and those yeah instead of asking someone from the design team or going yourself into Canva or making the changes Nano Banana Pro is exceptionally good at those things so I created a small tool where they can just upload an image

Give context about what they want to change. Select the client if it’s available also in the database. So we already have some context about what they are doing and the brand colors and things like that. And then the user just has to say, okay, this image, I want to change that button from green to red. And then they get the output, they can download it. They have immediately a new version of almost the same creative.

Niklas Buschner (09:08.27)
And what would you say to people that say, OK, but I can also just do this in Gemini and then select Nano Banana Pro. So why should I build a tool for this use case?

Alfred Simon (09:20.752)
Yeah. One thing is that in Gemini, yeah, you can create a gem where you can also upload the brand colors and things like that, but that’s nice for one client. But when you need to create gems for each of your clients, then it’s going to take always a bit longer. So with this tool, we have it pre-coded. So they just need to get the selector. I’m now working with this client.

They already knows most of the colors that he can use and brand tone of voice and things like that. Because it’s not just color changing with the A-B test. You can also write a new text and then you don’t have to explain what words we want to use. Some clients like this type of text or not. And then that’s already over there and accessible. yeah, it’s in this way, it’s better. The other thing in NanoBanana, you can get rate limited. With this we can’t because we use straight the API.

If 10 colleagues are working on it at the same time and everyone sends a request, we just scale with that. And Gemini, at some point you run into some rate limits as well.

Niklas Buschner (10:27.789)
Hmm. And how do you decide if, for example, someone from the team or from a client comes with a request for, Hey, couldn’t we do this in a better way? Or couldn’t we maybe use AI for that? How do you, how do you decide between the different options? Like, okay, we can just use like chat, GBT, Gemini, Claude with maybe a little bit better prompt, but we use

the normal interface, and then we can use a Cloud project, custom GPT or a Gemini Jam. So create a dedicated instance, so to say, in the tool for it. And okay, we have to build a standalone agent or standalone tool. So what’s your decision-making checklist, so to say, for these requests?

Alfred Simon (11:19.632)
Yeah. Yeah. It always starts with the request and what they actually want as output. when I start this and I have like a talk or a small meeting with the client or team members that they want something to do with AI. My first question is always, what would you like to be the output and how often do you need that? so when you talk with this tool agent, whatever, what is that you want to see?

as a response and that most of time that really helps deciding on which way we need to go. So if they say, usually I want to have a conversation about something. So let’s say I want to talk to my data, then building a full agent is not always the best option. Then you just need to find a way for that person so they can get the data in a safe way to the…

already existing platform. So to like cloud or Gemini or, or, or, or chat GPT or whatever. And then they can talk there. The reason for example, for this example is that inside all these tools, there are a lot of other tools that the agents can call. So if you talk with chat GPT, can search internet, things like that. And people are used to that. If I recreate that as an agent, standalone agent, I need to recreate all those tools. And that takes a lot of time and a lot of experimenting and they won’t work as good as what OpenAI created.

So when someone wants to talk to data and things like that, then usually I recommend, okay, we find a way for you to get that actual data from BigQuery, from Google Analytics, from whatever to that tool. And then you can talk over there in a safe way. If someone wants to have like an automated process that runs every day, then those platforms automatically are not an option.

So then you need to look into, for example, if it’s not hard to make, then NA10 is a nice option. Depends. So when they say, okay, I need this every day at certain hour delivered to a certain place, then the next question what comes is, okay, where is your data? How much reasoning the agent has to do when it needs the, it gives you the answer.

Alfred Simon (13:32.794)
Based on that, I can already see if it’s like a multi-agent process, so we need like two or three or four agents to get the answer, then probably N810 won’t be a good fit because of timeouts and crashes and it’s not always reliable. Then you need to think about another framework. But if it’s just like one agent and certain API calls, and then you send it as an email, then you dialed it exactly. Okay, then we go with N810. It’s easy to build and we fire it up. So yeah, mostly…

This is my checklist. I try to find out what they want as an answer. Do they need it timed? Do they need it in a certain place or they just want to talk with it? And that decides from which direction we go from there.

Niklas Buschner (14:12.111)
When did you become an AI architect or when did like your role change? Maybe also if officially?

Alfred Simon (14:20.059)
Officially changed in July last year. so obviously people don’t know me, but I have no AI background at all. So I am not even a computer scientist or something like that. I started out as a regular normal Google S specialist. I have nine years experience in that field. That’s what I did at the agency before I started working on AI projects. And this is came pure from…

Niklas Buschner (14:22.863)
Mm-hmm.

Alfred Simon (14:49.307)
I loved it and I saw an opportunity here when chat GPT got released in 2023. think I saw it on the day of the release on Twitter. Someone was posting about it and I also immediately tried it and I was instantly hooked. I was the first guy who said the agency, because the next day was Monday. went in, I showed to everyone, Hey, this is nice too. You can talk and you get answers and he can do a lot of things. Nobody understood at that point what I’m talking about.

Two weeks later, everyone was talking about ShadGPT. And I just loved it. I saw the opportunity, things that you can create, things, how many chances you have. And then I spent every morning before work, after work, weekends, everything, building certain things, trying it, playing around with the tools that are available. When a new model came out, I tried it instantly. When I saw a new tool, I tried it instantly. And I…

I shared all these processes inside the agency. So I talked about it. shared certain tools that I made and obviously the leadership at the agency saw it. Yeah. Maybe we should give him some more time to actually have dedicated time for this. And then that increased from 20 % to 30, 40, 50. And then in July we decided, okay, I should go then probably full on this.

Niklas Buschner (16:09.826)
And do you think or do you maybe even see this role also popping up at other companies? Because I feel like it’s still fairly new.

Alfred Simon (16:21.209)
Yeah, yeah, I probably, can dare to say we were the first agency in the Netherlands. I don’t know about the other parts of the world, but I probably, we were the first agency in the Netherlands that had a full-time someone that was working with AI related things. I’m not sure if this is going to be a standard or not. I am definitely sure you will need someone that focuses on AI more than other things. So if I take the agency model.

I really see the benefit of having someone like this at the company. The thing is that if you have someone full time doing things like this, then they can have all kinds of time to experiment new tools, find out what’s working, what’s not working, see patterns, what you need to try, what you need to ignore. Because when other team members come from their industries verticals with a question, you know exactly which tool can help them because you’ve tried so many already.

And you know, do you need to build something or you don’t need to build something? And you can help them a lot easier. So if you expect like a dedicated SEO specialist or someone that’s working with social ads to also create a tool while they are managing clients and things like that, it’s going to take a lot longer. So doing it in tandem with someone that’s focusing on creating the tool and getting feedback from the actual specialist can help a lot. And yeah, you need a team or one person at least that understands

All the things that comes with AI. still don’t understand a lot of it, but I do understand, think a lot compared to colleagues that don’t spend a lot of time with playing around with these tools.

Niklas Buschner (18:02.754)
And what helps you keeping up with all the news? Because I feel like every week probably there’s a new release, either Claude releases a new model or there’s, I don’t know, a completely new set of features coming out, then there’s like new agent platforms popping up, et cetera. So I feel like when I talk to people that are maybe still, so they are way earlier in their AI journey than you, for example.

One of the things that I hear the most is that they struggle with keeping up and then you already feel left behind if you’re not like on the cutting edge. probably that’s not true, but how do you handle this maybe even more like mindset wise or mentally?

Alfred Simon (18:53.583)
Yeah. The number one thing you need to accept is that you can’t keep up. It’s impossible. So if you accept to keep up with all the tools, it’s my job and I can’t keep up with it. While doing all the things and creating projects, there’s no chance keeping up. I have an AI agent running for myself that sends me sometimes news. I don’t even have sometimes time to read what he sends. So let alone keep up with the news. You can. So I think we need to accept that.

some point you won’t be able to keep up with all the tools. My thing is how I do it is once you get into it, so at the beginning it’s really hard. You’re going to see a lot of tools, you’re going to see a lot of ways to do it and everything like that. As soon as you get more into it, you will see that most of tools are the same and just the behind what they do, it’s a bit different. So if you understand like, for example, if you understand context window and how that works.

Doesn’t matter which coding agent you are using. You are using Cursor, Cloud Code, Gemini, Codex. It works the same way. One has a bit bigger context window, one has a bit lower context window, but the logic behind it is the same. And as soon as you understand these things, the basics, with everything, getting up on speed with a new model or a new tool, it’s not really difficult after that.

Usually what you can expect from a new model is that it can do more, has a better context window, understands things. So what I usually do is when I’m doing something and I run into an issue, for example, with coding projects and AI can solve it. And I see that we’re just running around in circles. I make a note of it. I write down what I tried and why it failed. And then when a new model comes out, usually I go back to those notes and I try a few of those issues again with the new model to see what changed, does the thinking logic changes of the model or not.

to see what works or what doesn’t work with the new model. yeah, this obviously you need this first to start using it. So I think you don’t have to be like scared or something. think the best is pick one of the bigger players, Gemini, Claude, whatever, and start actually like making something, do a coding project. Everyone has like ideas, everyone has a tool that they want or something. Try to make that.

Alfred Simon (21:14.264)
see where that takes you to understand how you work with these tools. Because as soon as sooner you understand, it’s the easier it be to move to a new tool when it comes out. So yeah, I think that helps a bit. I don’t know if this makes sense.

Niklas Buschner (21:28.878)
Yeah, makes sense, totally. there is always a new kid on the block, as I also said in the introduction, and now it’s like this open claw thing. and I mean, you said the first thing you have to accept is that you can’t keep up. But as far as I know, you’re also someone that immediately jumps into it. So if like, so.

Alfred Simon (21:51.033)
Yeah. Yeah, but this is, forgot to say, if you are on Twitter on LinkedIn, I recommend if you are really want to be up to date with AI stuff, go to Twitter because what I see on LinkedIn sometimes is two or three days later. So people are sharing things on LinkedIn. I’m like, that’s all news that was posted yesterday on X or two days ago. when something big comes out, because there’s so much noise.

Niklas Buschner (22:01.476)
Mm-hmm.

Alfred Simon (22:17.388)
Everyone is testing everything. When something big comes out, you will see it everywhere. So you instantly know, I need to try this. So every small tool release, you won’t be able to try it. But when, for example, OpenClaw is now the name, but when it was launched, was Claude. But when it came out, my Twitter feed instantly got flooded with multiple people talking about it in like the first two or three hours. Then I knew instantly, okay, this is something tomorrow I need to book.

two hours in the morning, I need to test it immediately. If like 10 or 12 other people talk about it on X, I know, okay, this probably is serious. So I need to look into it. And yeah, this is what helps. So people are helping actually each other with keeping up with the news because there’s so much tool come releases every day and something that’s mediocre doesn’t worth the time. People won’t talk about it that much. When something big jumps, everyone talks about it. So then you need to.

test.

Niklas Buschner (23:15.44)
So please take us with you to these two hours that you blocked after open claw or back then called clawed bot came out. What did you see? What is this? Why is this substantial? Why should people care?

Alfred Simon (23:30.458)
Yeah. So first of all, what OpenClaw is, is one of the most capable AI assistants at the moment that’s on the market. You have to think about it as a regular chat GPT or Claude. It can do most of the things that those platforms can do for you, but it has a lot of extras to it. So first of all,

You can install OpenClaw on your device. I don’t recommend doing that. Definitely not on your main device that you are using daily. Because the biggest difference between OpenClaw and the chat GPT-like platforms is that OpenClaw can do with your device, everything that you can do with your device. So it can open a browser, it can delete files, it can search for files, it can edit files. And it does this. So if you have experience with cloud code.

then you can understand OpenClaw quite fast because OpenClaw can run terminal commands in your device and with terminal commands basically you can do almost everything on your laptop. So this is one of the biggest differences. I made some notes for myself to not forget to mention the most important things. So let me just make sure that I covered all these.

Yeah. So this is one of the biggest that it can run the PC actually that you can as a user. So that gives so much freedom with it that it’s insane. The other part is that it’s really autonomous. So you don’t have to explain to it a lot of things. For example, when I was setting up…

OpenClaw, usually you can set it up by just talking to it. So once it’s running on a server on your device, you can just start chatting with it. And then you say, for example, I said to him, yeah, I want to activate NanoBanana Pro Skill in your, in your tool base. So you can generate images for me. And then I saw in, I use it in Discord to talk with the tool. And I saw that he’s typing, typing, typing, thinking, and then he came back with the answer. Yeah, NanoBanana is activated. All I need from you is to put on the server the…

Alfred Simon (25:45.932)
API credentials. Cool. Okay. He made it. I put the API credential in and it was working. Honestly, I have no idea what he did in the background, but it’s working and it’s working since. And so it’s really autonomous. So you don’t have to explain everything, what he needs to do and things like that. And also the other part, the third thing that makes it super interesting is that it has built in wake up moments.

So the two things that they are called and we can go into detail later on, obviously, but one is called the heartbeat and the others are Chrome jobs. So heartbeat, the standard setting for the heartbeat is that it wakes up every 30 minutes and it sends you like a DM message or he checks on certain things that is configurable by the user. So what you want the agent or the AI assistant to do every 30 minutes, you can make it every hour, every day, depends on your settings.

And the Chrome jobs are like tasks that you can schedule just like for example, with any tenant every day at eight o’clock, send me a report about this and then it will do it and it will come into your Discord or to your email or whatever place you choose. I think these are the three biggest, biggest, biggest differences compared to just the chat GPT account.

Niklas Buschner (27:01.209)
Okay, cool. And why do people now talk to Cloudbot via WhatsApp? Because this is something that felt fairly new compared to using ChatGPT or compared to Cloud. So like, what is the backstory behind that or why is this happening? And is this also something where you feel like this is part of the reason why it got so popular? Because

It feels so natural, like I can just message this bot on WhatsApp and it will actually do things.

Alfred Simon (27:33.144)
Yeah.

Alfred Simon (27:37.697)
I think that helped to become like, it made it more personal. So I, for example, don’t use it with WhatsApp because the recommendation, if you use it with WhatsApp is that you get for the agent, a mobile phone number, a separate number. So you don’t write it from your own number. I use it via Discord, but that’s also the same. It’s just, it’s just fun because you talk with an AI on a platform that is already familiar for you. So like WhatsApp, everyone uses it to talk with someone and then you just, you are on

on your way to work and then you just message your AI assistant to look up something for you before you get to the meeting. It can do it for you. So once you get there, you already have the message on your WhatsApp. It really feels like, I think for people, it feels like for me that you are talking to a virtual assistant that’s there for you 24 seven, and it can do basically anything for you that you ask. And that probably helped a lot to make it like more, yeah, that we can get attached to it a bit faster than other chatbots.

Niklas Buschner (28:37.393)
Mm-hmm and can you check some of the use cases that you already explored? Like things maybe even things where you felt like okay. This is not working out so well yet So we are we doing a little bit of expectation management here also with people but then also things where you felt like a this is working really really great and I’m Even surprised by how well this is going

Alfred Simon (28:54.339)
Mm-hmm.

Alfred Simon (29:03.159)
Yeah. So at the moment I call her Liara. I gave, you have to give, gave the assistant a name when you are setting it up, you gave it a name. Then they have what’s called the soul.md. So like how they need to act. Yeah. It’s, a, it’s, it’s interesting. And it’s what’s really fascinating.

Niklas Buschner (29:21.969)
We have to quickly explain here for people. Soul.md, because people might hear it. So it’s a file that is called Soul. And MD stands for Markdown. So it’s basically the soul of this bot written down how it is shaped, so to say. And I laughed because it’s so human.

Alfred Simon (29:28.887)
Yeah, it’s a file. Yeah. Exactly.

Alfred Simon (29:40.365)
Yeah, exactly. So in that sole file, standard comes out. So when you install it, it’s already there. has like a bit of instructions. And one of the instructions over there is that, I think literally says that this is your sole, keep it maintained, add things to it, remove things from it as you see fit. So the creator of OpenCloud…

deliberately created it in a way that it’s like it can change its soul. And what’s fascinating is that it’s really working. So as soon as you start talking to it, it adapts. it, for example, now it talks with me exactly how I want Liara to talk with me. I get short messages to the point explaining what I really want. No fluff, no nothing. That’s how I communicate as well. And it’s so nice to see that it changes over time and it remembers things because it also has a memory.

Markdown file. And in the system prompt of the, of this agent is stated that it needs to log for, for itself, the memory file. So when he thinks that something is important, it will make an entry automatically in the memory file. And when you talk with it and based on the question, the model already understands, okay, I might need to look up the memory because he’s asking something about the past. Then he searches the memory file quickly and then

automatically remembers, okay, we talked about that. We made these certain things. And these two things makes it really personal. But the original question for why I use it. It’s important to explain the Sol.md because I use it. One of the things I use it for is it’s like Liara is just now at the moment is a coworker at Backdoor Boost. That’s an app I’m working on.

And I don’t have too much time to work on the app. So I realized, okay, this is a nice use case for Open Cloud. So she can pick up things for me that I don’t have time for. And then we move on from there. So at the moment we are using it for creating blog entries for us to generate content. Creating images that we can use on those. That was one of my biggest time killers. So when I created the blog, even with AI, because most of it was already automated.

Alfred Simon (32:01.655)
The problem was I always had to think about the prompt. So what image I can use in that blog article, because you always need an image. And since I’m creating an image creator app, I always had to create images as examples. So you can create from this product image, this product image using our app. And that took a lot of time to think about the prompts. So what examples I can use. Now that’s, I just send a Discord message. Hey, write this blog and then create these example images. Everything is uploaded directly to our CDN.

everything is over there, it’s done. It’s awesome. So this is one of the things that she’s really good at. Writing blogs, it’s a hit and miss. It’s getting better due to the sole MD because I always try to give feedback on what’s working, what’s not working. So I scanned the memory and the sole file a few days ago and it’s really noting down things that I gave as feedback.

Niklas Buschner (32:41.648)
hehe

Alfred Simon (32:58.903)
And the blogs I see that are getting better. So if you install this for content writing, expect at the first week’s days, terrible output. And then you need to give it feedback just like for your employee. And then it gets better. It really gets better. So now the blogs that she’s writing are really good. So I don’t really have to give any feedback on them.

These are the two use cases that are the most heavily used. And the third one is reviewing our current code base. So she has access to full code of the app and she goes through it and comes with suggestions. So I have a, a crone job set up. So she wakes up every day at nine o’clock. She checks the whole code base and checks the metrics that we have. So how many people signed up in the past seven days and yesterday, did we make any revenue or not? And then she tries to help us with some features.

what we can add, what we can remove from the app. Because I said to her, your goal is to get us to 400 MRR at the end of February. since then we added 25 euros MRR. So it’s working. I don’t think we will make the 400, but hey.

Niklas Buschner (34:06.637)
Interesting. And what is something that maybe didn’t work out as well as you would have liked?

Alfred Simon (34:14.786)
code changes.

Niklas Buschner (34:16.333)
Mm-hmm, okay. That’s why you move to just having viewing rights on the code base and just making suggestions.

Alfred Simon (34:24.12)
She can make changes because she made a clone of the repository. She can send pull requests. So, hey, you should add this. And she actually writes the code and then we can decide, okay, do we add that or not? We can give feedback and things like that. Problem is that the code base that we have is already so big that even when you are coding with Claude, you have to direct it in a certain way so AI doesn’t get lost because it’s so much code and so much…

Niklas Buschner (34:28.529)
Mm.

okay.

Alfred Simon (34:53.144)
context there that the context window gets full immediately if you don’t manage it correctly. And that’s something I see with OpenClaw that I use that yeah, it just messes up the context window and it creates really weird things. Sometimes it writes a code and then I’m like, I’m testing on this code. Did you actually test what you wrote? Yeah, I tested it. I was like, are you sure? Because I just tested it and it crashed and then comes back with a message. Oh yeah, sorry. Yeah, I tested it. Yeah, it doesn’t work. I need to figure this out. I was like, okay, figure it out.

So code changes are like really a hit and miss up until now in total, I think she created 13 pull requests and I was able to use four of those actually to push through production something. So that saved me a lot of time actually. And the other ones are good ideas, just the execution was bad. So I saved them and I will give it another shot.

Might be my mistake as well. don’t use the most expensive model with OpenClaw because I’m still not a millionaire to spend thousands and thousands of euros on API calls because it makes some costs. So I didn’t dare hook it up to Opus 4.6, for example, because probably I will go bankrupt in a week. So I’m using it right now in two ways. Using with my OpenAI account, you can log in and then…

Liara uses the same account as I do with OpenAI. And then she’s using Codex 5.3, which is a really good model. Only thing that, yeah, it gets quite fast rate limited because we are using the same account. And then I use with API the Gemini 3.0 flash model because it’s cheap, it’s fast, it’s quite good, but for coding is not the best choice. So I still need to try it probably with the 4.6 Opus.

Niklas Buschner (36:38.866)
Hmm.

Alfred Simon (36:44.566)
Yeah, I didn’t dare at the moment to hook it off to 4.6.

Niklas Buschner (36:47.634)
Okay, if anyone listening has a million spare, Alfred would be delighted if he gets a donation to do some… Okay, 10,000 is enough. So people, let’s make it work. So Alfred can actually test Opus 4.6 with Liara. But I think here’s one critical point. And you mentioned that from the 13 pull requests, you could use four.

Alfred Simon (36:58.262)
I’m already happy with 10,000!

Alfred Simon (37:07.84)
Exactly.

Niklas Buschner (37:17.582)
Now I can imagine people listening to this and saying, this is very disappointing. So I would stop using it then. I could imagine that you think about this differently.

Alfred Simon (37:30.784)
Yeah, I know that from the 13, at least six was my fault, giving bad instructions and setting up Liara to fail basically. The model use immediately explains why some of those fails failed. So when she ran out of usage on the open AI model, the fallback model is the flash. So I assume she was halfway through the task.

Then the rate limit came in, then she fall back to 3.0 flash. And if you ever coded with 3.0 flash, you know that’s rubbish. So when she switched to 3.0 flash, it immediately became really bad quality code. So I am sure if I really like give it a good model and can run on a good model, I’m sure that that four will jump immediately to six or eight. But even with the four, I’m already satisfied because

Those were also bad instructions, but those were things that I didn’t have to care about at all. if you code with cloud code, it’s fast, it’s good, but you need to be there. You need to type things or talk into microphone and then transcribe it and then accept things and check code and things like that. But with this, all I did, I sent a Discord message and I went to work and then I came back and I had a response, hey, it’s done. And I just had to check the pull request and that’s…

If you bump it up with a better model, I think that’s only a win in this situation.

Niklas Buschner (39:04.368)
And can you share something about the speed of execution? Because I tried Claude for Chrome. I haven’t tried Claude co-work yet, but for Claude for Chrome, I felt like it’s really fascinating and generally working well, but it takes really long. So it takes forever to complete a relatively simple task. So if you, for example, ask Liara to, I don’t know, publish a new blog post and create an image for,

the app or the blog for the app’s website. How long does it actually take to complete the job?

Alfred Simon (39:41.912)
Um, honestly, I’m not sure because the blog creation is a scheduled job for her. So she has to write one blog a week. I don’t want to spam too much content on our website. Uh, and it runs every Monday morning at eight o’clock and then I just get the message that’s done. So I’m not sure how long she takes, but the message usually arrives around like half, like eight 30, something like that.

But I’m not sure if actually needs half an hour for the whole process or just a message gets sent later. I assume it’s half an hour, let’s say half an hour. But if you just chat with the agent, depends a bit on the task, what you are asking, but usually it’s quite fast. So I remember two days ago, I needed something figured out based on our data. I was planning a meeting with my co-founder and I had no time to look up certain things in analytics.

And I was just stepping into the car from the office to come home. I, before I drove away with the car, I sent a message to Liara, Hey, look up this and this and this and this. And I remember that I barely drove away and I already got the message with the data. So it took like two or three minutes, not even to go to analytics, get all the data because she’s using mostly API. So it’s like quite fast. it’s getting the data fast. Depends a bit on the request, but it’s usually really, really fast. So I can’t complain about speed.

Niklas Buschner (41:07.164)
Okay, nice. And which other use cases have you planned to test? Because now you’re obviously settled already a little bit. You know how it works. It has adapted its soul so it meets your communication style. It feels like you’re through the early days of testing. So I could imagine that based on how I know you, that you’re already, at least in your brain, there are already a lot of ideas cooking what’s coming next.

Alfred Simon (41:35.573)
Yeah. So I upgraded the setup recently and at the first one and a half week, I was running her on a VPS on a virtual private server, which is not the best option. It’s probably the good option to start, but if you really want to have open Claw unleashed, you need to give it a device because then it can use browser on a way. Yeah. Yeah. That’s exactly what I have, a Mac mini. Good job.

Niklas Buschner (42:05.107)
I’m just holding a Mac Mini into the camera for everybody that’s just listening because obviously I also saw someone on LinkedIn posting about it, you should have a separate machine, blah blah blah, but yeah.

Alfred Simon (42:05.207)
Yeah.

Alfred Simon (42:14.648)
Yeah. Yeah. I totally agree with that take. A of people say, I don’t need a Mac mini, you need a VPS. So yeah, if you are like a nerd, VPS is probably good because you can figure out the server and everything. For me, the choice was to go with a Mac mini because I understood, okay, if I need to take this to the next level, I constantly needs to log into my server, make there certain changes. I have no idea how to run the server.

I barely done it a few times. I had to ask Cloud Code to make that for me. It’s a lot of work and I still feel that I’m limiting the potential. With the Mac mini, you also get the browser use case unlocked. So she can really use a browser like you do. Then you can say, hey, go to Verserl, create an account, put the project up and do something or check Twitter and get back to me about this and this. Cause otherwise you need to give in a VPS, all kinds of API access and things like that. Yeah.

It, it, it, it’s not, I don’t think it’s the ideal. And also if you have like a separate device, it’s, it’s a bit more secure because you don’t have to mess around with securing the VPS. So you know what you put on that separate device and you know that that’s the only thing that open clock and access. So if you put there an API key, you know that she has access to that API key. yeah, if you see like that’s something going wrong, you just take that API key away and you know that’s more or less isolated. On a VPS, you have to also make sure that you are.

securing the server itself. That’s also a lot of work. If you don’t know what you are doing, you can get in trouble. People can get into your server, get all your credentials, things like that. So that’s a risky move, which you are not technical. But what I’m planning right now, I have three plans for the upcoming period. I have now Sunday morning blocked off to set everything up because the Mac mini just arrived the beginning of this week and I just quickly set it up. So I need to really dive deeper into this.

But what I want is that I want her to create a blog for herself where she can write every day something. I will just leave it open. I really am curious what she will write about and will it get traction or not on the internet? The second one is that I want her to create a project that will create, I would say revenue, but at least get users. say that way. I want, I will say, Hey, this is come up with project ideas.

Alfred Simon (44:38.039)
start coding it and push it to virtual, create whatever you want, make a business. I’m curious if it can actually do that because I saw someone on Twitter that was able to do something like that. So I want to test that as well. And then the third one is just taking it further with the coding part for Backdrop Boost because if she’s running properly on a separate device, then if my research is not…

wrong, then I can use a cloud account where she can log in and then she can use cloud code via an account instead of the direct API, which is a lot cheaper. And then we can use Opus 4.6 and then we probably open Pandora’s box here in my small office.

Niklas Buschner (45:23.719)
Yeah, I was about to say the same. Pandora box. We’re approaching the end of the rainbow. Nice. A quick question before we dive into the setup. I know people want to know how to set this whole clod bot thing up because like it sounds cool, but then the question is, okay, but how do I get started? And we obviously want to deliver also on that promise. Why is it a Mac?

Alfred Simon (45:28.449)
Yeah

Niklas Buschner (45:50.535)
Why is it not a Windows or like a Microsoft PC? Why is everybody buying Mac Minis now?

Alfred Simon (45:58.251)
Because Mac mini is stupidly strong for the price that you have to pay. One of the reasons. And the other one, the main reason for me, for example, is I can use an iMessage to talk with her. And that’s even cooler. So that’s nice. And the other thing is I can share a of things from my Mac book to her. So Apple notes, things like that I can directly share because she will have like an iCloud account as well.

Niklas Buschner (46:11.187)
Hmm.

Alfred Simon (46:26.838)
So it unlocks a lot of things for me to do because I also have a Mac, so it makes it easier. But I think the main reason is that Mac minis are really cheap at the moment. I think Mac minis probably are a good investment for short term. So if you buy them now and everyone runs out of stock, you can resell it probably a few months later. I don’t take this as a financial advice, but yeah.

I think that’s one of the main reasons, because Mac Mini’s are cheap and strong.

Niklas Buschner (46:58.099)
Hmm. Got it. I guess there will be someone that buys like 100 Mac minis, then sets them all up and then provides like like a, EA as a service, like executive assistant as a service thing where you can basically have your own AI assistant.

Alfred Simon (47:05.824)
Probably.

Niklas Buschner (47:21.841)
Which in the back end is like a Mac is a cloud bot running on a Mac mini that will then be personalized to you and you can share what you want to share and they set up a couple of things already pre pre-made Are you doing it?

Alfred Simon (47:32.454)
Already people are already doing this. saw someone, yeah, I saw someone on Twitter. There’s a platform called Trust MRR. It’s a platform where you can, so a of people say a lot of things on internet and Mark Liu, the creator of the platform created this platform. You can connect your Stripe key to it and then you can actually see if people talking about their app actually make money or not. And I saw a lot of guys on

Trust MRR that they created an open clause something so you can use it or you can set it up there quickly. I saw one of the guys made 20,000 in like a week or something. I think if you go to the platform and you look up like open something or Claude bot, if you research with that, you can probably already see a of services that are there. And people making serious money already with the idea that you had. So yeah, that’s coming. Yeah. And it’s already here.

Niklas Buschner (48:09.844)
crazy.

Niklas Buschner (48:28.304)
Interesting. Very, very interesting. So now let’s get into how to set this up. I know you prepped a little something for people to also better understand it. Can you guide us through? if I feel like this is interesting, Alfred convinced me I want to get going. How can I get started tomorrow?

Alfred Simon (48:34.41)
Yeah. Yeah.

Alfred Simon (48:47.882)
Yeah. If you are rich, then buy a back mini. If you can afford it, do it straight away. If you are like a bit more cautious and you don’t want to spend 400, 500 euros straight away, then you can probably do it the way I did it. So that is go to AWS, that’s Amazon’s server stuff. They have a super generous free tier. Basically you get, if you are making a first time account, you get 200 euros.

Niklas Buschner (48:52.094)
Ha

Alfred Simon (49:17.046)
think as a budget, which is plenty enough. So you can run your open claw for free for at least a long period. So I ran it over there for more than a week and I didn’t even spend 10 euros of that free budget. You can create that and then on there you can start experimenting. Okay, is this something that I like? Is this something that I don’t like? Is it nice? You will probably have a lot of headache managing the server, but for that I created something for

listeners and I will share my screen if you don’t mind to walk everyone through this small thing.

Niklas Buschner (49:55.701)
For everybody that is only listening to this, we will also share what Alfred is showing us with a link in the show notes so you can check this out. Otherwise, highly recommended going to YouTube and checking it out there. I don’t know the timestamp exactly that we will be on, but probably somewhere around the 50 minute mark. Then you can check it out on YouTube also.

Alfred Simon (50:22.752)
Yeah. So what I’m sharing right now for the people that are not watching the screen is I created a small repository on GitHub with the Claude MD file. So if you are not familiar with Claude MD, it’s like, it’s a markdown file that if you are using Claude code for coding, it will be sent every time you send a message to Claude in the background. So it’s like a system. And what I did with the AEWS server, I set it up.

quickly on the interface. And then I use cloud code to configure the server in a secure way and everything. Because cloud code, because it can use a terminal, just like your open cloud wheel, it can log into the server via the terminal. You just need to give it the keys to the proper place. So what I created here, I opened out the cloud.md. It’s basically a template for the cloud.md that you can use in your project if you want to do it the way that I did it.

So this will explain for Cloud code, the SSH access, Cloud will help you to set up the PM file. So I don’t even explain that. You just can ask Cloud what’s in this file and he will explain it. And then Cloud will can also guide you through what server you need to select, what you need to install on that AWS server when you are setting it up. Just have a chat about basically what you want to do because in this cloud.md file, I mostly try to put everything that I already use and it’s important.

And the, one of the biggest things that I think it’s the most value I can share right now is the OpenClaw command map. What I did, I scraped the whole GitHub repository from OpenClaw and I scraped all the commands that you can run on the server to set up certain things. like changing modders, logging into certain models. So I have here, for example, OpenClaw channels list. That’s the command you can run in the server terminal window.

And then it will do something. But this is not for you. This is for Claude code. So if we go back quickly to the Claude.md, that’s the thing that every time is sent to Claude code, there is a line at the end of it, OpenClaudeCommands. See OpenClaudeCommands.md for the full CLI reference. So when you are talking with Claude and you say, for example, to Claude, Hey, I want to switch models in my OpenClaude agent.

Alfred Simon (52:48.959)
Can you do that for me? Yeah. Claude then will do a check that command the map. He knows exactly which command he needs to run on the server. And he will make the changes for you. So you don’t have to enter the server every time, write certain commands, because trust me, that’s a terrible thing if you are not familiar with using terminal. So yeah, that’s one of the things. I also have a setup guide. This is probably where you need to start. Just give this to your AI assistant that you like. I recommend using Claude code because you will need that as well to

set up the server and it’s one of the best things you can use at the moment. But it’s another conversation. But back to the set of file. This is, you just give this to Claude and he will walk you through the whole set of how you set this up on AWS on the free tier. Which storage you need to select, which instance you need to select, how many ramps and things like that. If you don’t have an idea.

It will guide you through. And if you don’t want to use AWS, I also added other options such as DigitalOcean and other providers. I also saw yesterday that Hostinger has a, actually they made a service for this. So if you search on Hostinger OpenClaw, they have a one-click setup for I think seven euros a month. So you just buy that and they create a server for you where OpenClaw is there. So you just need to configure it after that.

Niklas Buschner (54:05.318)
Mmm.

Alfred Simon (54:13.48)
It’s also secure and it’s a probably fast way to do it. But then you already need to pay for that. And this what I show you, it’s free if you do it with AWS. And the rest of the file is just for Claude explaining what he needs to do once he gets into the server. Basically the instructions are not even for you. And I have here in this file, I think I have a small part of the…

Yeah, it’s a small snippet of the sole.md that I was mentioning at the beginning. But this has like the core values, personality traits, and basically this is like what’s in the sole.md. It’s like explaining to the OpenClaw agent who she or he is, identity and things like that. And the nice thing about the sole.md, this is the core identity, but this is editable by the agent himself. So this will improve over time based on the conversations that you are doing. And the last thing in the…

Niklas Buschner (54:46.889)
Hehe

Niklas Buschner (55:09.823)
How long did it… sorry. No, no.

Alfred Simon (55:12.425)
Yeah. The last thing in the repository is a logger. That’s also explained in for Cloud itself. This is just that Cloud, when you are configuring the server using Cloud code, he will make here logs what things have been changed on the server. It’s like a memory for Cloud code. So what I did, I created a project on my device, which is called Liara. And that’s me using Cloud code to configuring the server every time. And I need to make a change on the server.

Once the server is up and running, it sounds really complicated, but trust me, you can do it in like, in like, I don’t know, I did it in like 15 minutes using cloud code. Once the server is up and running and the OpenClaw agent is live and it’s living and you can talk with it on Discord. After that, the whole setup happens with talking to it. So you just say, Hey, I want you to be able to do this and this and this, and then OpenClaw will just fix it. Maybe he will maybe ask a few things like.

Niklas Buschner (56:08.595)
Wow.

Alfred Simon (56:11.734)
You will say, Hey, I want you to be able to browse on the internet. And then he will come back like, okay, I need an API key for Brave and the rest I fix. then you, the API keys, you don’t directly give it to the OpenClaw. not just copy paste it. When you need to add the API key to the OpenClaw agent, then you go to your Claude code setup and then you tell Claude, Hey, I need to add to this server a new API key in the instructions. He knows how to do that.

Cloud will give you a command that you need to run yourself. Also don’t give your API key to Cloud himself. Don’t just copy paste it into the interface. Cloud will give you a command which you need to then run in your terminal. Ask Cloud how to do that, it will help you. And then you just run the terminal command and after the terminal command, you just paste the API key. And once you run that, the API key will be added to the ENV file in the server, which then Open Cloud can use. And that’s…

Niklas Buschner (57:09.427)
Hmm.

Alfred Simon (57:11.131)
secure more or less.

Niklas Buschner (57:13.115)
Nice. It’s very good that we have this repository that we can share with people. And thanks so much that you built it actually for us to share. I highly appreciate it. What I wanted to ask you, because I feel like, so I feel this sense of excitement you have about it and drawing back on your experience you had when you saw the launch of Chatchi BT and that feeling of

Monday morning, going to the office, talking to people, hey, have you already seen that blah, blah. How would you rate, and I know since this inception moment, you have been deeper and deeper and deeper with AI. How would you rate this moment now with open claw, like in comparison to that and everything that that happened before.

Alfred Simon (57:53.525)
Mm.

Alfred Simon (58:02.741)
Yeah, if that moment was a 10, I think it was a 10 back in 2023. So I was super excited about it. Then I would say this open claw, it’s probably a seven and a half, eight-ish on the whole scale. I think I will go quickly to nine once I have the whole Mac mini set up properly. So I will, I might change my score after that.

Niklas Buschner (58:07.946)
Mm-hmm.

Niklas Buschner (58:19.561)
Mm-hmm.

Alfred Simon (58:31.624)
But just based on the early experiments I made, I would say seven and half, eight, which is quite high because the last tool that made me give such high score was the launch of Nano Banana Pro. And that was a big one as well. so yeah, it’s, it’s, think this is probably a really, I wouldn’t say, probably it’s revolutionary in a way that I think the bigger,

Niklas Buschner (58:43.637)
Hmm.

Alfred Simon (59:00.954)
open bigger AI platforms such as OpenAI or Anthropic. They saw now that, okay, there is need for this. And if we make something like this in a secure way, then that’s a key to a lot of other new users. Because the platforms currently can do something like this, but they have the strongest models. So imagine running a Liara natively built by Anthropic on an Opus 4.6. That would be like absolutely bonkers.

Niklas Buschner (59:21.077)
Hmm.

Alfred Simon (59:31.315)
I’m not sure if they can make it because there are a lot of security problems with OpenCLO at the moment. So probably that’s why they didn’t do it because for them it’s like a big risk if they mess up something. But at some point they will figure it out how to do it properly. And then we have proper AI assistance that we can use by the native model creators. that’s probably the thing we will have again, that will be the biggest thing at that point.

Niklas Buschner (59:58.455)
Because that’s what I wanted to ask. mean, if I remember the Chet GPT, I think it was for second half of 2025 strategy document that got leaked in a suit, so in a legal process in the US with Google, they actually said the division for Chet GPT is to become this super assistant.

And I feel like so the more I listen to you talking about it, I feel like this is the closest we got to having the super system now. So you feel like security is the biggest problem. So this coming from the community basically as an open source project is due to an individual creator.

Alfred Simon (01:00:44.67)
The project was wipe coded. the guy who created OpenClaw basically, he said it in an interview. I have no idea what every file is in there. So I was just talking to, he created it with Codex and some of it with Cloud Code. He was just talking to his laptop and he was programming it and then it started to work and then from there it grew. yeah, you can see it as a wipe-coded project that is really good.

But yeah, there are some security stuff. You have to be careful how you use it. And I think that should be a totally different episode as well, but I’m not sure if you have time or I can go like quickly five minutes into like security things. Be careful what you give access to your open claw if you set it up at the beginning. So if you do it on a separate device, then you are more or less good in a way that you know what’s on that device and…

Niklas Buschner (01:01:25.974)
Of course, let’s do it, please, yeah.

Alfred Simon (01:01:41.616)
She has access only to that things. Don’t sign up your agent to platforms where other agents are. One of the platforms that got really popular is the Mold Book. That’s like Reddit for all these open-claw agents that people created. And that’s number one most unsecure place you can give your agent to access because people can do prompt injections from there.

So that’s, that’s, that’s, that then, then it’s bad, obviously don’t I, for example, I wouldn’t recommend allowing your agent emailing, people other than you or talking to people other than you at the beginning, because if someone knows what they are doing, they can also do prompt injections. So if someone sends now an email to Liara with a prompt injection, it will probably go through it and then she can leak my API keys and things like that. So I wouldn’t want to do that.

So at the beginning, just keep it on your device. If you are running it on a VPS or on a private server, make sure the server is secured. And after that, just give access to the tools that you first want to use. So I would say use it with Discord at the beginning, super easy to set up. Telegram is also super easy to set up. What I heard, I didn’t test it, but I shouldn’t be hard. And then just start talking to it. Make sure you have one.

clear use case in your mind what you want to test and just give access to the tools at the beginning that you want to use and nothing more. Be mindful how you give access via the API keys. Create separate API keys with clear budget. for example, if you use Nano Banana Pro, so you want your assistant just like I did to create images. Create a new project on Google console for that. Set a budget alert of 10 euros.

over there and an instant kill if the budget is depleted and create a separate API, obviously. So you can take precautions like this. If something goes wrong, she burns a lot of credits, the most you lose is a tenner and that nobody gonna have a headache about it. But if you don’t do it and she burns 1000, you might have a different conversation at home.

Niklas Buschner (01:03:58.027)
Maybe depending on which account is charged. Yeah. Okay. These, yeah, these are very important additions. Thanks so much for that. I guess. Are there any resources that you draw on when you think about this? So is there anything that like you read like a blog post or a post that we can also maybe put in the show notes that have some helpful security guidelines? Because I could imagine people thinking,

Alfred Simon (01:04:00.762)
Exactly, exactly, exactly. I know my wife wouldn’t be happy.

Niklas Buschner (01:04:27.442)
Okay, yeah, that sounds really good, but wait a second. What was it again? And I know how people are. They probably won’t really listen the whole the whole part. So is there anything that we might be able to share?

Alfred Simon (01:04:39.624)
Yeah, obviously their own website is really good. They have a lot of docs. So openclaw.ai I think it’s the domain or com. I’m not sure. Yeah. There is a lot of good stuff about setup and other things that can be helpful at the beginning and things like that. I recommend also checking a YouTuber called, I don’t want to…

Niklas Buschner (01:04:43.19)
Mm-hmm.

Niklas Buschner (01:04:49.226)
Yep. .ai.

Alfred Simon (01:05:08.862)
Let me just quickly look it up because I’m subscribed and he should be on the top because I watched the video yesterday evening. Matthew Baerman, he created a lot of good videos about how to set it up. So there is also lot of good content and he also has lot of links in those videos. I can share maybe a link with you, the video I watched and then that can help as well.

These are mostly like setting up security and things like that. If you want to get inspired on how to use OpenClaw and what you can do with OpenClaw, then follow Alex Finn on Twitter. The guy already bought two Mac studios. That’s a 10,000 euro device. He bought two of them to run OpenClaw. The guy is creating an army of OpenClaw agents. He’s doing a lot of things. I think…

50 % is hype to just generate content because he’s a content creator. So probably there’s a lot of things like that, but he is really inspiring to open your mind and to see certain use cases that you can do with your open clause. So I will share also the link to his profile so you can put it in the show notes.

Niklas Buschner (01:06:18.517)
Wow, nice. Thanks so much. Wrapping up the conversation a little bit. I always want to ensure that people will leave the episode with the most actionable advice. And I think we already delivered on that like three times. So thanks so much for that. But we touched on getting started with AI. We touched on

how to think about structuring the problems, which route to go. So is it a problem I can solve basically in the UI of ChatchBT, et cetera? Can I solve it with Cloud projects, custom GPT, et cetera? Do I need an agent? We talked about Cloudbot, setting out on VPS, Mac mini. So we covered a lot. Now, if there are people listening that are thinking a lot about AI, but are still early in their journey.

Like the early, okay, yeah, I have built a custom GPT, but other than that, I’m still just prompting to GPT, for example, for stuff. What would be your maybe two to three top pieces of advice, actionable advice for people that want to take their next step, at least towards something that, yeah, you’re building, for example.

Alfred Simon (01:07:17.992)
Mm.

Alfred Simon (01:07:37.427)
Yeah. Number one, I would say start a coding project where you use AI. You will learn so much about how these models work and what are the capabilities and everything you do in code, it can be translated later to other use cases. You don’t have to understand the code that you are writing. Just use AI for these things so you can understand how to talk with them. That’s what you will learn.

by creating, you don’t have to recreate the new Facebook. Do something small. So I can’t believe, I think everyone has like an idea or something for an app. Maybe you have a problem at home that you are frustrated about. don’t know. For how I started, actual example, my very first project that I coded with you, I was a gym tracker. Because all the gym trackers that were out there were rubbish, were not for my use case and for my program that I was following.

So I created that in like an afternoon back then with Sonnet 3.5. It was like, feels like it was 10 years ago. And that teach me so much about how to talk with these, what are limitations, things like that. And it’s really something like simple like that you can set up quite fast with AI. And if you’re not sure if your idea is simple or not, ask AI. So take a microphone so you don’t have to type. Install handy.computer.

I can share a link that’s a free transcriber. So if you are familiar with Whisperflow, it’s awesome, but you have to pay for it. And if you only need a transcriber, then handy.computer is an app that you just install and it runs locally and it’s completely free. Then turn it on, talk for 10 minutes into your microphone about your idea, send it to Claude, Gemini, ChagePT, whatever. Ask the tool, okay, is this difficult or not to build?

And if you get like a plan or more or less an idea, okay, this is, it looks like manageable. Then install cloud code or cursor. I recommend if you are doing a coding project, don’t do it in a chat interface. It’s going to be a lot harder and you lose motivation. So install cursor or get cloud code and that probably would be the wiser option. And anti-gravity, it’s also a really good option and I think it’s free.

Alfred Simon (01:10:00.436)
So if you don’t want to spend money on this, then use anti-gravity from Google and you can use Gemini 3.0 Pro there for free, if I’m not mistaken, for a certain period of time. And you can create probably a nice project and you understand most of the things. You don’t have to understand like code and things like that, but you will understand a lot about how these works, context windows, things like that, because you will encounter a lot faster over there, these issues than in just a chat interface.

And those things are playing everywhere when you are working with AI. I think this is one of the biggest things that you can take as next step. And the other one, which you should already do and should be doing and should stay doing, write everything down. If it’s in your head, it doesn’t exist. So if you need help from an AI assistant, doesn’t matter which two, about your fitness goals, personal projects.

If the AI doesn’t know what’s your current status, what you are doing at the moment, how the heck you expect you get a proper tip from AI. So when I realized this, I started to track everything. I’m also a gym nerd, so I created the tool. I track daily what I do, what I eat, what’s my weight. So, and my AI is connected to that via API. So when I asked Claude, hey, can you get the data from my fitness and check it to my goals for 2026? Am I on track or not?

He knows exactly where I’m at at the moment and what I need to change to get better. But if you don’t have this context, you just get slopped. So write everything down, write projects down that you are working on, challenges that you face with that project, everything basically. So what I use is Apple Notes just because it’s accessible then from all my devices. And I have a folder for each project that I’m working on and also areas of my life. If you read the book,

build a second brain from Tiago Forte, then the listeners that read the book probably understand what I’m talking about. It’s like a system where you take notes in certain way. It helped a lot also with AI stuff. So I work on a project, I write down everything, what work, what doesn’t, what are the challenges. And when I asked Claude about it, he knows the whole context. I don’t have to explain things. So he knows what I already checked, already did. And it’s a lot easier to write these things down as you go.

Alfred Simon (01:12:20.211)
And you have a lot more context to work with and you get proper, proper stuff out of these AI tools. But you need to write things down because if it’s only in your head, then yeah, then it’s nowhere.

Niklas Buschner (01:12:32.937)
Have you always been like an explorer type person? Like there’s a new thing and I’m intrigued. I want to understand it better. So I will play around with it. Also pre AI.

Alfred Simon (01:12:40.892)
Yeah.

Yeah, yeah, pre-AI. My whole life is just farfawing. So just to give a small context, I moved to the Netherlands nine years ago. originally from, Hungarian. when we, so we had certain things at home and we wanted to try something new with my wife. So one day we just sold everything and we stepped into the car and then we drove to the Netherlands to start a new life.

Niklas Buschner (01:12:47.756)
Hmm.

Alfred Simon (01:13:09.212)
The only thing I knew where I was going was my cousin, because he was living here and he said, you can live one month at us. I had no idea where I’m going to work. And three days later I had to work. So yeah, my whole life was exploring and yeah, trying things.

Niklas Buschner (01:13:24.019)
Yeah, I feel like that’s a trait that a lot of people share that are on the frontier of like these technological things because they don’t wait for someone else to show them the way, but they actually find the way themselves. So you’re the personal story with you moving to the Netherlands. It’s like a perfect testament to that. I would say, couldn’t be a better example.

Alfred Simon (01:13:42.802)
Yeah.

Yeah, don’t, I’m not afraid making mistakes, breaking things. And you will do a lot when you are working with AI. So I see a lot of colleagues do something and then it breaks, doesn’t work and they come into panic and I don’t know, I made some mistake. like, I broke only 14 things today. That’s not bad. Yesterday I did 60.

Niklas Buschner (01:13:51.37)
Hmm.

Niklas Buschner (01:14:02.808)
Was it hard to get buy-in from the agency leadership that it will in the future be normal that things break?

Alfred Simon (01:14:13.195)
happily not. so they are really open, open-minded. So they understood what this brings. I explained it at the beginning. They already knew me, so I already broke a lot of things before that. so it was easy for them to accept, okay, this probably just will give more problems to us when after this unleashed. So, but it’s also gave a lot of benefits. also, they said it, so it’s not, not my words. They said it to me. So, yeah.

But happily it was easy to get them on board.

Niklas Buschner (01:14:42.178)
Awesome.

Very cool. I have a final question for today, which is what didn’t we talk about that we should have talked about?

Alfred Simon (01:14:58.034)
I think we covered most of it, but I would say one important thing, I saw a few days ago on Twitter, someone wrote an article that got like 70 something million impressions about that the new COVID is here and it’s AI and got a lot of attention. And what I hate at the moment on whole social media is the fear mongering about AI.

And I saw someone else reacting to that post with one line and it was brilliant. He said, AI doesn’t come for your job. Fear is coming for your job. So if you are just sitting still and you don’t want to do anything about it and you are close-minded and you don’t test these tools and you are not open to using them yet, then you probably will lose your job or changes will come to your life. But most of the people are not like that.

People are like me, the explorers, yeah, we are way ahead, but everyone will catch up at some point. And if you look around in the real world, how fast things get implemented, how open-minded are boards and things like that to implement AI. When you go into a real company with rules and corporate stuff, yeah, it’s not easy to get an AI project up and running. So you see all these guys on LinkedIn talking about AI, but those are like explorers. have…

This is their job. Most of them are also content creators, so they need to talk about something to generate impressions. But I would say, don’t be afraid. I would say I’m a positive person. would say really nice things are coming towards us. We will be able to solve so complex things that we were not able to solve before. And that will unlock a lot of things. Probably for some people will also bring negative things. But yeah, just don’t let fear disappoint you or…

Don’t get overwhelmed by the fear posts and things like that. And I’m already behind, I’m not going to try it. That’s the worst you can do. So just be open-minded, ignore everything, go try it yourself, see what it can do in your field, in your area for you. And you will understand it a lot better than reading on Twitter. When you see someone just on LinkedIn, you see someone just talking about fear and things like that. Yeah, that’s the sign to close LinkedIn and go try it yourself and see where that takes you.

Niklas Buschner (01:17:16.97)
very well put. Thanks so much. I very much enjoyed the conversation. It was very insightful. I like I literally have this Mac mini here standing in front of me on my desk and the

Alfred Simon (01:17:23.09)
Yeah, me too.

Alfred Simon (01:17:26.93)
You should end the podcast and install it.

Niklas Buschner (01:17:30.827)
Yeah, yeah, the more we talk, unfortunately I only have time on the weekend, but the more we talk, the more I was like craving going to it because it felt like this moment when, I don’t know, when I was a teenager and I had this new computer game and I was like, I was literally sitting in school, like very nervous about getting home, installing it, setting everything up and then just going, when I had a new computer or something even crazier, like

Can I can install my new graphic card etc? And I have the same feeling so thanks so much for giving me that and this was a very Yeah, it was great. And if people want to follow you around so obviously we will put all the links that you I talked about in the description But what’s the best place to follow you because I know that you also share a lot about your thoughts and the projects you’re working on

Alfred Simon (01:18:06.352)
Yeah, thanks for the chance.

Alfred Simon (01:18:20.518)
Yeah. LinkedIn is a nice way to follow me if you want to see a bit more filtered and normal of it. And if you want to get a more raw experience then Twitter is the place. I really post more and like different content on Twitter than LinkedIn. But if you follow me on LinkedIn, that’s already nice. There are also shared love things. When I have time, obviously it’s not daily.

Niklas Buschner (01:18:48.376)
Hopefully Liara can make more time in your day so that you can post more and share more.

Alfred Simon (01:18:53.17)
Yeah, it would be nice. But the problem is if she makes more time, then I already come up with a new project, which will take time again. But yeah, that’s life.

Niklas Buschner (01:18:59.32)
Yeah, that’s actually the new struggle. You make time and then you have more time but then you immediately start working on something new.

Alfred Simon (01:19:06.524)
Yeah.

Alfred Simon (01:19:10.117)
Yeah, exactly.

Niklas Buschner (01:19:12.44)
Cool, so then Alfred, thanks so much. Very much enjoyed it. All the best for you, all the best for advice, all the best for backdrop boost. I hope you will get to the, what was it, 400 euros in MRR? Yeah, okay, yeah, no worries. I don’t think that it always has to be a shiny big number. So wish you much success in that and yeah, thanks so much for making the time today.

Alfred Simon (01:19:20.934)
Same for you.

Alfred Simon (01:19:26.77)
400, yeah, small numbers, so yeah.

Alfred Simon (01:19:39.729)
Yeah, thanks for inviting me and same for you as well. All the success that you can handle. yeah, open to the right. Bye bye.

Niklas Buschner (01:19:45.324)
Thanks, see you. Yes, bye bye.