Sunday Braindump – 2025-02-01

Catching up and OpenClaw/Moltbook thoughts

Huh,

I just discovered that you’re all still here, still subscribed to this abandoned newsletter (please feel free to unsubscribe if this annoys you or is irrelevant to you in any way; I won’t mind). And so in the spirit of playfulness and experimentation, I’m going to use this microphone that I currently still have without any concept or plan and without any AI generation or something.

Voice-pilled

Because I somehow, over the last couple of months, became very voice-pilled. I used to be someone who hated talking to a computer. I never used Siri, and I never used Alexa or anything like that. But when trying out LLM chatbots and seeing a couple of recommendations for that, it became clear that it actually works quite well to use a transcript of what you’ve said to talk to these chatbots. And for me that’s a completely different mode of thinking and working than writing. When I write, I’m editing all the time; I’m trying to find the perfect words, and I instantly rewrite sentences and stuff like that. When I speak, I just have my train of thoughts that I’m following. And so I have been speaking to my computer a lot. I even, to the annoyance of my partner, got to the point where I do it when other people are in the room. And so I’m going to use this and do the opposite of what we see mostly on feeds, which is slop. I’ll just give you my train of thought on whatever comes to mind. And it might just be this one time, or I might just use it in the future. So here we go.

One year of self-employment

So to update you, if you haven’t seen anything from me in the last weeks or months, I have now been fully self-employed for a little over a year. And I’m mostly working as a keynote speaker these days, because especially in Germany, the economy is so bad, nobody’s thinking. about the future at all. If it doesn’t make you revenue in the next three months, companies are not interested. So they don’t want to take any risk, they don’t want to invest in any futures work, but they very happily want to talk about the future. And so it turns out that is a product that I can deliver. I am quite happy with how it has shaped up in the first year.

It’s also especially, I think, because in Germany people feel so little agency about the future that the term “shaping the future” is everywhere. Posters, political campaigns—it’s everywhere. But nobody talks about how that actually works. How do you shape the future? And so this has become my main keynote theme: explaining what that actually means. What the future is, what shaping the future means, why you should do it, and how you do it in a practical way. That has been working actually quite well.

The thing that I discovered, which I was kind of aware of but got much more into, is that I actually really, really enjoy doing keynote talks. I’ve been doing them for a long time, but I’ve never done them as my primary source of income or my main type of work. But I really quite enjoy working with organizers, figuring out what the dramaturgy and the goals of their event are and what role the opening keynote plays in that agenda. And then really helping the audience to make the most of the day of the event, getting them into the topic, getting them into the right mindset, stuff like that.

Yeah, I’ve been quite enjoying that and have also been enjoying always putting a bit of the Critical Futures work into these keynotes. I have my Three Assholes slide that shows Zuckerberg, Altman and Musk as examples of popular people who are invoking specific future narratives to advance their own agenda. My goal is to help the audience to learn how to decode future narratives that they hear or that are getting passed around and then help them to take the first steps to defining the narrative of a future that they want to live in. That’s been my main source of work.

The AI and work discourse

I also have been playing around a lot with LLM tools, all kinds of things, to really like figure out what’s in there and what I can learn from experimenting with them and trying them out. So I have been doing talks about AI and the future of work since 2013. And it became clearer and clearer that this discourse around machines taking over our mental work and algorithms replacing humans has been around for a long time and essentially hasn’t changed at all. The talk that I gave in 2015 at Republica—I could slightly polish or update the examples, but it would still completely work today. And I think the reason for that is that this discourse is way too abstract; it’s way too high-level. And so it never leads to any significant change or new actions. Thus, my strategy for the last two years was to get a much more concrete, precise understanding of these systems, not just the technology but also the cultural aspects and the economic aspects, because then it becomes much clearer where the levers are.

OpenClaw

The current example, as of right now, is called OpenClaw, which was called Clawdbot and then Moltbot before. This is an open-source agentic system that you can install on a VPS or, as many like to do, a Mac mini, which seems to be really popular. I installed it a week ago, and it was very fascinating to use because it made so many abstract things, like the discourse around agents and the “agentic future,” so much more concrete because you could see the potential in a system that was best described as “what Siri really should be.” Seeing a system that is based on the premise of it having access to everything—to your emails, to your documents, to everything you give it access to, and the more you give it access to, the more powerful it becomes to do certain things—and seeing the potential of that and making it so much more concrete.

In the last 12 months there was a lot of talk in economic circles about agents and humans together on the org chart, and that’s one of those metaphors that is always passed around, and it doesn’t make sense at all. If you have done any work with agents, you really know that they are very much not replacing humans. They are replacing certain tasks, or they are automating certain tasks. But nobody in their right mind who’s not just interested in doing a press release will seriously consider creating an org chart with humans and agents in there. That doesn’t make sense. It’s also a good example for me of why it’s so important to have very concrete experience with the system because then you immediately know where the discourse falls apart.

So I started playing around with OpenClaw. And in the beginning, you could just use your Claude subscription, but then Claude started banning people because it was against the terms of service. And so you had to give OpenClaw an API key for one of the LLM services, which also meant you had to pay for each token that you used. And it very rapidly for me became quite clear how resource intensive a simple personal agent is. You use a top model like Claude Opus 4.5, and you just let the agent run for a couple of hours without doing any big things, just the usual, like the heartbeat feature and a couple of chats. Within a few hours I had amassed costs of approx. 10 euros. So if you run it for a month, you’re easily in the hundreds. and so that shows you that all those visionary ideas about the agentic world … It’s still very very expensive and resource intensive. The other option, obviously, is to use a local model, but to get any meaningful use of a local model, you have to have a really, really expensive machine that is capable of running big enough local models for them to work at all in a meaningful way.

The other thing that I found really interesting is that over the last week, it instantly became very clear by sheer anecdotes how unready our digital systems are for these kinds of all-encompassing personal agents when it comes to security. At the beginning it was mostly just that they were not really secured from access from the internet, and it was not that hard to fix that. But having these probabilistic systems reacting to the world or even just observing the world opens it up for so many inroads of hacks and prompt injections and similar threats. I read lots of articles by security researchers who basically said, “We spent the last decades to make computer systems and the internet more secure. We invented a lot of frameworks and models and methodologies to have much better security. And to a certain extent for the agentic world, we have to completely start that again.”

The other day I joked on the feeds that we will have to thank all those YOLO nerds who just ran an OpenClaw instance with access to everything and had their authentications and tokens and keys stolen. Because these very concrete examples created a roadmap for the next weeks, months and years and actually allowed us, because we had so many concrete learnings, to move from the abstract to the concrete, which I’m guessing will make more robust and secure systems arrive much faster.

Moltbook is about humans

So these are the things that I’m thinking about this weekend. I’m not going to touch on Moltbook too much because everybody is very hyped about it. I think my understanding is that despite what everybody is saying, it is actually quite easy to just post on there as a human. So you should take anything you’re reading on there with a grain of salt. If it seems too perfect to be true, it probably is written by a human. But as always, what fascinates me with these kinds of occurrences is that I think the reactions to Moltbook tell you much more about the humans than they tell you about agents. We are seeing people and how they react to it, and they are like, “Oh my god, what’s happening there?” and ”Should we shut it down? It’s going crazy!” You can just see how it basically is like a live-action science fiction story that we feel like we’re observing, and we’re just so used to these tropes that we can’t help ourselves in seeing the ”ghost in the machines.” This is what we’re going to get again and again, where it’s just a performance of intelligence and consciousness, and it works so well to fool us.

Closing

Okay, there’s tons more stuff to talk about, but I’m just going to end it here. This has been my train of thoughts, basically a braindump of what just came to mind. I’m going to clean it up a little bit so that it is easier to read. And then we’ll see.

If you have any questions for me about anything, please hit reply or leave a comment. Questions are what I run on.

For transparency’s sake: I used Monologue to record this and QuillBot to take out the transcription errors (but did not use formulation recommendations; according to QuillBot, it gets only 57 for fluency and 54 for clarity out of 100). I also added some headlines to make it easier to read.