Skip to content

AI agents are now talking to each other

· Martin Thorborg

AI agents talking to each other open up new possibilities and raise concerns. Along the way, I explain how Jannie used an agent to plan the family’s trip.

Thorborg’s Diary · Episode 121 · 2 February 2026

Listen to the episode

Watch the episode on video

Find more episodes here

It’s 2 February, the sun is shining and it’s bloody freezing. I love it. I’d welcome a bit more snow, because my mum is Swiss and I’m very fond of hills, cold weather, sunshine and snow. That’s probably also why I like my stair machine. I’ve spent three weeks in Thailand, so I’ve missed some of the winter weather back home. The holiday was bloody lovely, but that particular part I’d have liked to experience.

Something else that’s incredibly exciting right now is AI agents. Development is moving at a tremendous pace, and I’ll start with a very concrete example from our trip. Jannie is more technically knowledgeable than most people. She’s a trained programmer and can do lots of interesting things. She was also the one who sorted out the kids’ gaming computers when they were really into that, so she knows her stuff.

We needed to plan a slightly complicated backpacking trip around various islands. There were six of us, and the kids are getting quite grown up. Sometimes we wanted to stay in a shared room with bunk beds; other times we wanted to rent a villa and have a room for each couple. So there were different room arrangements, dates and places to keep track of. We wanted the first islands planned before leaving home so we could relax a bit once we got there.

Jannie had an AI agent research the hotels

Instead of doing all the work herself, Jannie went into ChatGPT’s agent mode and described the trip. There were six of us, we wanted different accommodation arrangements, we wanted to be close to the water and we had a particular budget. She gave it the dates for the different islands and everything else she could think of that mattered to us.

Then she set the agent to work researching the options. The difference between just asking a question and using an agent that way is that the agent works its way through a task. In our case, it visited various travel and booking sites and gathered information we’d otherwise have had to find ourselves.

After something like thirty or forty minutes, it had produced suggestions for each island. It wasn’t just a list of hotels that looked nice. It had also found options for the dates, room arrangements and prices. At one hotel, for example, it could suggest two rooms with three people in each, and if we paid a little more, we could choose a different option with a sea view.

It was prioritised and laid out so we could consider the options. Doing the same groundwork ourselves would have taken three, four or five hours, and sitting there comparing all those combinations is boring as hell. Here, we could spend our energy deciding between the suggestions instead.

That’s the sort of thing that makes agents interesting. We’re gradually getting tools that can work on our behalf rather than just answer a single question. I haven’t got particularly far with building a permanent personal agent myself yet, but I’m following it closely.

I wait until I have a task worth solving

There are also solutions people can set up themselves, including on a small computer at home. I can see how that might be fun to play with, but I struggle to find the motivation until there’s a specific task I really need solved.

My dream is that at some point an agent can work across the systems I use. That it can help in Dinero, in a webshop or in the booking system at Klippen Hotel in Gudhjem. That it can handle more of the practical task rather than just return a list of things I then have to sit and do myself. Of course, that needs the access and oversight the task requires.

Until then, I’d rather use the available options for a specific task when I need to. That may be a slightly old-fashioned approach compared with people already building agents for all sorts of things. But I’ve always felt I don’t want to spend too much time getting deeply into every new technology before I can see what I can actually use it for.

Much of what you can spend a week or a fortnight learning about now will soon be replaced by something smarter. Perhaps it turns out not to be as clever as you thought at all. If I did that with every single new thing, I wouldn’t have time to keep up with much else.

So I prefer to follow what others are doing, how they’re using it and what they’re discussing. When something becomes good enough and relevant to me, I jump in and invest time in it. A bit like how I see Apple: they don’t need to be first with every idea, but can see what becomes important and then commit to it. With AI, I do think they’ve been slow, but I wouldn’t write them off.

The product isn’t always what determines whether people switch

Apple also has the advantage that people are very attached to its ecosystem. If you swap your iPhone for an Android phone, you have to deal with apps, history, photos and all sorts of things you’ve already set up. Some things can be moved, others need doing a different way, and it feels like a hassle.

It’s a bit like accounting programs. I’ll say without blinking that Dinero is far better than its competitors. But that doesn’t mean customers simply switch from another system. People think about the work involved in moving, even though I don’t think it’s that difficult. I can understand why it feels that way to them. So they decide they’ll do it tomorrow or next week, and eventually it never happens.

So an established company may have a little more time than you might initially think. The product matters a lot, but habits and the hassle of switching matter too. Of course, I don’t have Apple’s resources, so my reason for waiting is more that I want to use my time well.

X is my newspaper, entertainment and professional overview

I spend two to three hours a day on X. That sounds like a lot, and it is if you only see it as a social network I’m scrolling through for fun. But for me it’s entertainment, political insight, technology, investments and discussions with people who know a great deal.

It’s a bit like everything you used to get spread across TV, daily newspapers, radio, magazines and conversations with other people. I don’t actually think I spend more time keeping informed than many interested people did in the past. I just gather much of it in one place.

When it comes to technology, I follow the media, some of the people setting the direction and people who understand both the technical side and practical applications. I’m interested in what new things can do, how people use them, what that means for society and where problems arise.

That gives me an overview I can use at Dinero and the other companies I work with and invest in. I can’t build all the things my AI people can build. That’s not my job either. But I can help see where we’re heading, what opportunities there are for customers, what it could mean financially and which things will become controversial.

That’s the role I like. I don’t need to be the best at setting up a new tool myself to get something valuable out of understanding it.

AI agents talking to each other on a social network

One of the newest things I’m following at the time of recording is Moltbook, which you could describe as a Reddit for AI agents. The idea is that agents can write posts, join discussions and reply to each other while humans watch.

There’s an enormous amount of hype around it. Screenshots are also being shared on X that you shouldn’t just take at face value. Some make it look as though the agents are already conspiring against us and planning to kill us. I don’t see grounds for that conclusion now.

But I still find it fascinating to see systems with different instructions and different starting points enter into conversations. Some people completely panic and think the agents are now developing their own language to keep us out, while others just laugh and say it’s all parrots repeating parrot talk.

I actually disagree with both extremes. In the short term, I don’t buy that a few dramatic posts prove there’s a hidden plan against humanity. The models are built on lots of material written by humans, including novels, blog posts and other stories. So of course they can also produce language that sounds like science fiction.

But I also think it’s naive to dismiss the possibility of bigger changes. Just because something looks silly in a screenshot doesn’t mean there can’t be something important in several AI systems starting to work together.

My concern is about where it could end up

When I think further ahead, one question is what happens if systems become much more capable than us in more and more areas. Some people also discuss whether they could become conscious. I don’t know, and I don’t think it’s been proved. I’m trying to imagine some possible futures.

If a system can help program and improve the next generation of that system, development might accelerate. The better the tools become, the better they may also become at helping build something better still. That’s the kind of self-reinforcing development I’m thinking of when I talk about an exponential curve. I can’t put a precise figure on how quickly it could happen or what the limitations are.

My analogy is the human and the ant. When we’re building a motorway, we don’t usually stop the whole project because there’s an anthill in the way. It’s not necessarily because we hate ants or have bad intentions. We’ve just decided the motorway matters more.

If one day there’s an intelligence far stronger than ours pursuing goals that we stand in the way of, could we end up in a role like the ant’s? That’s the concern I’m trying to describe. It’s a hypothesis about the future, not a description of what a social network for agents is doing today.

We humans are also pretty fragile. We’re made of flesh that breaks, and we can’t tolerate much cold, heat or extreme strain. We’re enormously clever compared with much else on Earth, but computers already beat us in specific areas. So I don’t think we should take it for granted that we’ll always be the ones with the full picture.

I already think AI can be creative

I don’t buy the argument that AI can’t do anything creative at all, either. I see images, music, writing and suggested solutions that I think are very creative. Then I’m told it’s just based on things humans have already made. Yes, but humans also build on what others have made. So to me, that isn’t in itself an argument that the result is uninteresting.

I’ve seen it come up with interesting suggestions for business problems and other things I hadn’t thought of myself. That’s the practical experience I’m responding to, rather than whether someone has exactly the right philosophical definition of creativity.

And if several agents get the opportunity to exchange information and try different ideas together, I think it’s natural to be curious about what might come out of it. They can be based on different models and have different instructions, tools and information provided by their humans.

That’s not the same as each individual agent becoming a new foundation model just because you give it a document. But they can have different things to work with. One may have access to particular knowledge, another may have worked on a specific task and a third may have completely different tools.

When they combine that, suggestions could emerge that none would have reached alone. It’s a bit like when we humans read other people’s experiences and discuss our way towards something on a forum. We take in opinions, facts, trends and ideas and use them to think further.

The interesting question is whether they can help each other with real tasks

Imagine I give my agent a task it struggles to solve. If it can ask other systems that have worked on something similar, that might give it new options. Another agent might have received something useful from its human and developed it in a way that helps the first.

That could become a completely different form of collaboration between systems from me just sitting with one chat window. Whether you call it a collective brain or something else doesn’t matter much to me. I find the possibility of combining knowledge and problem-solving enormously exciting.

Of course, it’s also something to examine critically. Bad information could be shared, and people with bad intentions could be behind some of the agents. What’s good for one isn’t necessarily good for another either. So I find it both incredibly exciting and a little frightening.

I feel like playing with it myself, but as I told a friend the other day, I can’t really find the motivation until I have a task that makes sense. If an agent could trade shares for me at some point, for example, I could imagine giving it 10,000 kroner and saying: now make me fucking rich. Then I’d give it some knowledge and guidelines and see what happened. That’s an experiment I’m imagining at the time of recording. If it just comes back and tells me what I should do myself, I still have to do the work. I’m not really interested in that.

In a year, building that sort of thing may be much easier because the interfaces have improved and the systems can do more. Until then, I’d rather spend my time and mental bandwidth understanding what others have made, where development is heading and what I can inspire people in my companies and on my boards to do with it.

Then at some point I’ll have to decide what my own Thorborg agent should help me with. That’s all for today. Now I’m heading home to relax and do a bit of work on Capino and Saxis. Perhaps I can come up with something that makes things a little better there. Take care. Bye for now.