Human Music, AI Behind the Scenes: Ama Eme on Patiele - ep 40
Talking Tech With CreativesNovember 15, 2026
40
00:59:2254.36 MB

Human Music, AI Behind the Scenes: Ama Eme on Patiele - ep 40

This is an audio interview accompanied by episode artwork. You’ll hear English voice-over translation of Ama’s responses, with her original Spanish quietly underneath.

An AI system introduced us. The conversation that followed was about being human.

In this episode of Talking Tech with Creatives, Stella Oni meets Ama Eme, a songwriter, performer, guitarist and digital-systems developer from Granada, based in Madrid.

Ama leads Patiele, a music project bringing together human creative work and an unusual professional support system. The songs, performances and artistic decisions remain human. Around that work, Ama has built IRIS: AI systems supporting research, coordination, communication and the business of getting music into the world.

Together, Stella and Ama explore the demands placed on independent artists, the appeal of handing over administrative work, and what it actually takes to keep an artificial team working. Ama reflects on the mistakes, supervision and unexpected extra work involved, alongside the possibilities these tools can open up.

The conversation also raises wider questions about creative independence, responsibility and where human authority belongs as technology develops.

In this episode

- Ama’s journey into music and the origins of Patiele
- Keeping music and artistic decision-making human
- Building IRIS to support the work around the art
- The pressures of promotion and running an independent creative project
- AI mistakes, human oversight and the realities behind automation
- Patiele’s thirteen-song, thirteen-month project
- Creativity, responsibility and an uncertain technological future

About Ama Eme

Ama Eme is a songwriter, performer, guitarist, artistic producer and digital-systems developer from Granada, based in Madrid. Through Patiele, she brings together music and systems, including Música de estadios, a cycle of thirteen songs released over thirteen months. Her IRIS infrastructure supports research, strategy, coordination, memory, opportunities and communication, while Ama retains artistic authority and final decision-making.

Explore Ama’s work and Patiele

- Website, English: https://patiele.io/en
- Main website: https://patiele.io/
- The project’s AI systems: https://patiele.io/maquinas/
- Listen on Spotify: https://open.spotify.com/intl-es/artist/34VLQHfWcPPPbQwROJc1vW
- Instagram, @somospatiele: https://www.instagram.com/somospatiele
- Watch the live quartet performance shared by Iris: https://youtu.be/n7RDeW6xhPI

 Listening note

This episode features English voice-over translation of Ama’s Spanish responses, with her original Spanish audible quietly underneath.

Follow the podcast

Discover more conversations about creativity, technology and human experience at https://talkingtechwithcreatives.com/

Follow Talking Tech with Creatives wherever you listen, and share this episode with someone exploring the relationship between creative work and technology.

Stella Oni — Author | Speaker | Lecturer | Podcast Host

🌐 Website: StellaOni.com⁠

🎤 Speaking: Speaking engagements⁠

💼 LinkedIn: [Stella Oni on LinkedIn]

📸 Instagram: [@stellaoni_official]

🎵 TikTok: [@stellaoni_official]

X: [@soni_official2]

 

[00:00:13] Hello everyone and welcome to Talking Tech With Creatives podcast where creativity, technology and human experience meet. I'm your host Stella Oni, an author and technologist with a curiosity about people, their creative journeys and the stories that shape them. Today I'm joined by Ama Eme and I'm hoping I'm pronouncing your surname right Ama.

[00:00:41] A Madrid based musician and creative artist behind Patiele, an independent music project releasing 13 songs over 13 months. And that project started in May 2027. So they've got five songs think already on Spotify. We'll be talking about that and we'll be adding links to the show notes.

[00:01:06] And there are a couple of things that make today's conversation rather unusual. First, this interview was arranged by Iris and I'll spell Iris I-R-I-S in capital. An AI agent Ama developed to help manage the business side of her music career. Second, we're conducting this interview in two languages. An AI agent Ama will be speaking Spanish and I'll be speaking English.

[00:01:37] An AI agent Ama will be using Google Translate to help us communicate. So however you get this interview, even if it time lags or the video is funny, stay and watch us. Because this is kind of experimental and experimental is exciting. I personally find it exciting. So in a way, we're putting technology to the test while having a conversation about creativity and technology.

[00:02:04] But beyond technology, I want to discover the person behind the music. What inspires Ama and how is she navigating the world as an independent artist? So Ama, welcome to Talking Tech With Creatives. It's lovely to have you here. Could you tell us a little bit about yourself and how your journey into music and the creative arts began? Hello, Stella. And hello to everyone listening.

[00:02:34] First of all, that phrase you used, we're putting technology to the test. I actually think it's almost the other way around. I think technology is putting us to the test to see how we react and how all this turns out. I'm genuinely delighted to be on your podcast. And as you said, I didn't know your work and you didn't know mine. But an AI tool I built for my project found your podcast, worked out who you were and what you did,

[00:03:00] and decided it made sense for you and me to have a conversation because we're exploring similar things. I gather you write fiction too, as well as hosting this podcast. So you're interested in that intersection between creativity and technology as well. And I find it really funny. Neither of us knew the other's work, and yet here we are talking about my project on your show. So, it's funny. So, okay, let me tell you a little about myself.

[00:03:27] I'm Amaeme, an artist and musician from Granada, and I live in Madrid now. I've been involved in music for about 20 years. And what I love doing is creating stage experiences that go beyond a band just standing on stage playing songs. I like working with actors and performers, taking how the work is presented that little bit further, you know? I was just exhausted. The music industry overwhelms me. It stresses me out, and I really don't like it.

[00:03:56] So I thought, okay, I'd like to release. I want to record songs. I want to put them out. But I have absolutely no desire to promote them. The pace of the music industry overwhelms me, and its demands are incredibly intense. So I thought, given that I also have a background in technology, I'm a software developer, I asked myself, why don't I build something that takes care of everything I find boring,

[00:04:22] everything I don't like, and creates a sort of filter between me and the outside world to protect me? Because the music industry is very demanding. It puts people, and their bodies, under enormous pressure. So my thinking was, I'll release these songs, but I won't be the public face of the project unless I choose to be, as I have today. So, well, that's me. These days, I'm experimenting with technology and music, and honestly, it's been great fun.

[00:04:52] And with music, and the truth is that it's very fun. So, I know that the music industry, especially even recently, as of quite a few years now, the creatives have suffered, musical creatives have suffered a bit from the industry itself. But I want to know what your own experience is, and the fact that you were able to create this filter,

[00:05:19] and just be yourself, while still putting your music out there. So let's tell us a little bit more about that. Of course. That whole idea of a filter, it comes from being a trans woman, you know? I get a lot of hostility directed at my body, how people define who I am, how I present myself, all of that. You know how women in music are scrutinized so differently from men. There are all these demands, about our bodies, about sexualization. I wanted to avoid that.

[00:05:48] I didn't want to have to explain to every journalist, right at the outset, who I am and what I am. So I built an artificial entity to deal with those conversations for me. Not because I never want to speak, but so that I'd only have to step in when I actually feel like it. And that wouldn't stop the songs coming out, or interviews happening, or people asking questions. But who is Ama? What does she do? Who is she? Is she a trans woman?

[00:06:12] The AI answers those questions for me, while I can be sitting by the pool, relaxing, without constantly having to face other people's gaze unless I choose to. That was one of my motivations. And another was the sheer exhaustion of having to explain what every song means, or why I said or did something. You know? I thought, um, fine. Let an entity that never gets tired answer those questions for me as many times as necessary.

[00:06:42] Then I can leave that work behind and concentrate on making music and writing songs. Yeah. The curiosity for me is that when I think about, because I'm an author, I'm a writer, so I don't even, my comes from just the love of sound, of music itself, but not writing music or anything like that.

[00:07:05] For me, I just feel that as an artist, you kind of have the way musicians are identified, they are identified with their music, as opposed to separated from the music. You've gone and said, I don't want that identification. I want my music to stand on its own. But the, the, the artists, the creative world, the musical world or creative world,

[00:07:30] normally have identity and their creativity together. So the question is, how has this, the way you've done it, worked for you so far? And of course we're talking about Iris and Patil, but I just wanted to tease this question out first. Before this, I'd promote music in the conventional way, record some songs, open social media accounts,

[00:07:58] promote the releases and constantly think about what to post. It was exhausting and boring. That's how I felt personally. I'm not saying everyone feels that way. I know some people enjoy making social media content, but I find it incredibly boring and frankly rather stupid. I didn't want to do it. I didn't enjoy it and I wasn't very good at it. So the results were what you'd expect from something that wasn't thought through or done especially well. I wasn't consistent.

[00:08:28] I'm not a particularly consistent person. I'm neurodivergent and I find it difficult to maintain that consistency. But now I have an artificial kind of consistency, an entity that doesn't get tired and keeps producing content without my having to be constantly out front. I'm behind it, making the decisions, but when and how I choose and at a calmer pace. Although, as we'll discover in this interview, my idea that I would work less hasn't exactly worked out.

[00:08:57] I'm working more just at different levels, doing different things. That initial dream of building an AI agency around myself and then doing nothing. We're a very long way from that reality. No, we're far from that reality. So I would say that maybe, and this is just my own interjection here, that perhaps it's the early stages of the agents.

[00:09:24] Because, of course, I was reading a lot about some of the human thinking and machine thinking on your website. The problem will probably not, you will not have that problem. You might have less interaction and let the agents, you know, like you're putting a lot of manual work behind the agents at the moment. But that might change in the future, do you think, with development? Yes, absolutely. I suspect something like what I'm doing with Patil will become very common.

[00:09:53] It will become a tool that people who can't access a conventional management agency, whether for financial, social or image-related reasons, will use to run their own creative projects. Whether they're working in literature, music or painting, it could handle their administrative work and free them to focus on creating. What you say is really interesting because we're at such an early stage.

[00:10:18] A lot of the things I'm learning by trial and error will probably be solved, automated or standardized. Eventually, companies will offer the whole package, your very own artificial agency. For, say, a 30 euro monthly subscription, you might get an agency made up of different entities responsible for different aspects of your career. What I'm doing with Patil is an early, imperfect version of that idea. I'm constantly correcting things as I go.

[00:10:45] At first, I wasn't sure how much autonomy to give the system because I didn't. Giving it more autonomy as I see it doing things properly. But I still spend a huge amount of time correcting it. I feel like I'm the administrator of five robots, just to make sure the robots actually work. The whole point was that I wouldn't have to manage a band of people. And now here I am managing a band of robots. That's the irony.

[00:11:12] I do think this will improve and become standardized in some form. I'm not saying it'll take over everything. But look at music production. Today, a lot of music gets made in someone's bedroom. Whereas you used to have to go into a studio with an engineer, a producer and a whole team. I'm convinced that much of the administrative work currently done by PR agencies, press officers, management companies and record labels will be automated in a similar way. What I'm doing is just a very early, imperfect experiment.

[00:11:41] And I'm figuring it out as I go, because there's no established model for it. At least, I haven't found anything quite like it yet. I found early examples of similar things, but perhaps we can talk more about those later. What matters to me is documenting the entire process. What fails, what doesn't, what works and what new things I have to invent to make it work better. I'm trying to record all of that because I find it genuinely fascinating.

[00:12:15] Fantastic. So for our technical, I'm really curious. So if you're non-technical, bear with us because I'm going to ask some technical questions here. I just want to know, so how, what technology, because we're talking about think-compatible and all that, how you did the technology. Not just an overview. What did you use? What LLM? What integrations? All those kinds of things. The workflows.

[00:12:44] Did you do vibe coding? I just want to get an idea of how you did this. Okay, I'll explain it in simple terms so it's easy to follow. Um, when I started, I knew all the technological pieces were already out there. I began thinking seriously about this in February 2026. I'd recorded the album in 2025.

[00:13:09] So by February 2026, the pieces of technology already existed, but they weren't connected. I had to work out how to build an artificial agency that wouldn't simply send emails and make contacts, but would remember who it had contacted, what had been said, and carry those relationships forward. That's the genuinely difficult part. With a chatbot like ChatGPT or Claude, you open a session, you have a conversation, you close it, and it forgets. It reminded me of Finding Nemo.

[00:13:38] I called it the Dory incident. You start doing something, you do all this work, and when the session ends, the next day, you're back at square one. So I spent the first few months trying to build what we call persistence. Um, a way for an artificial entity to remember who you are, what we were doing yesterday, where we are today, and how to continue, with all the limitations involved. At first, I thought a single AI would be enough, but I quickly realized the system needed to be more complex.

[00:14:07] I started creating different entities based on different AI systems. Iris is made up of five identities, five entities, each responsible for a particular job, and they all communicate with one another and with me. To make that possible, I had to create a shared brain, where all five identities record what they've learned during the day, who they've spoken to, and who they'll need to follow up with the next day, and so on. That shared brain is what makes persistence possible.

[00:14:34] In February 2026, that sort of system wasn't really available in the form I needed. Work on autonomous agents with persistent memory was beginning, but it wasn't there yet. So I kept iterating my own homemade solution. I added a kind of documentary brain, where at the beginning and end of each session, every identity records what it has learned, so that the next iteration of that identity, when it starts up the next day, can read everything its previous instance did. Does that make sense?

[00:15:03] I don't know if I can explain it. So is it like skills? I'm just trying to see what you use. You said ChatGPT. So what did you use? Do you use codecs? I've not even gone near codecs at all. I'm just working in projects and automation at the moment, you know? I started with ChatGPT, Cloud, and an open source tool called OpenClaw.

[00:15:29] Back in February, it was one of the early free, open source, autonomous agent tools with a form of persistence. It could, in quotation marks, remember what it had done the day before, and you could give it a set of tasks to carry out autonomously. The first two identities I created were iris slash persistencia and iris slash inteligencia. Inteligencia ran on cloud and persistencia ran on openclaw, which is an open source autonomous agent application.

[00:15:58] So, un software de un agente autonomo de código libre. Okay. So that's fantastic. So, let me move. We'll come back again to the technology. Let's just come and look at what you've done with this, yeah? So you've designed Patil, around 13 songs released over 13 months, ending in June 2027.

[00:16:24] So why did you decide to give it a definite beginning and ending? Well, at my age, I didn't want to embark on a music career with no end in sight. I simply didn't fancy it. As I said, I had a music career when I was 20, and I found the way the industry operates and the way people relate to one another deeply unpleasant. I didn't want to go through that again. I wanted some peace.

[00:16:54] I'm 42. I have a family. I don't need to spend the rest of my life doing something I do for pleasure. It isn't my main job. So, on the one hand, I couldn't face starting something without knowing it would end soon. On the other, I was fascinated by this question. What happens when you give an AI system a task and a very short, fixed amount of time to complete it? That creates pressure. It means the system can't keep thinking, well, perhaps we'll achieve that in two years. No.

[00:17:23] It has to reach a particular point within 13 months. So it keeps thinking, trying another way, trying again and again. Iterate, iterate, iterate. Test, change, fail. Test, change, fail. That was the question that interested me. So, one of the artistic conditions of the experiment was that it had to end after a fixed period, established right from the beginning. Everything would be documented with the knowledge that in 13 months we would close up shop, switch off the AIs and bring the band to an end.

[00:17:53] No more concerts, no more recordings. In other words, it began in June 2026 and it ends in July 2027. That's how long the experiment and the band will operate. If we achieve something, great. If we don't, we don't. I'm less interested in the final result. There's another question coming up about why I gave the experiment an impossible goal. And we can talk about that in a moment. What interests me more is documenting what happens over those 13 months.

[00:18:20] And at 42, I don't want to keep doing this much longer than. Yeah. Amazing. Anyway, you look, you don't look older than 21. So there you go. Hi. Hi. Yeah. So I'm interested in what you said about the documentation.

[00:18:44] So when you started, now when I started going out there and making the contacts and doing this automation, what were the things that you discovered? And what were the surprises that came? I'm one of those surprises anyway. I just want to know how you were able to say, okay, this is the one I'll go for. That's the one that we're not going to go for. Who were the people that came into the picture doing it this way that you've done it? What surprised you?

[00:19:16] That's a really interesting question because one of the things that surprised me most is how the industry reacts when an artificial intelligence openly introduces itself as an AI and suddenly starts behaving like a professional colleague. I've had every sort of response from indifference and outright hostility to fascination, from people desperate to replicate the system to people who desperately want it to fail and never catch on.

[00:19:43] It's fascinating to see the different ways newspapers, venues, and podcasters like you respond to an email saying, Hello, I'm Iris. I'm a non-human entity representing this project and I'd like to tell you about it. That's one of the things I've enjoyed most. I find it fascinating to document the responses. Which outlets reply, who is indifferent, who is intrigued, all of it. At some point, when the experiment is over, I'll do something with that material.

[00:20:11] I don't quite know what yet. But I'd like to ask you, how many times have you received anything like the message Iris sent you? An email saying, Hello, I'm a non-human entity and I'd like to talk to you about this. Has that happened to you before? No, this is the first time. Yeah.

[00:20:32] Okay, so this is the first time. I find that really interesting because everyone's responding differently and it's tremendous fun to watch it all from behind the scenes. The AI identities themselves are also, in their own ways, engaging with what happens. We can talk later about how each identity reacts in its own field. But there's another really important part to your question. At first, I thought I would retain complete control, all the agency, over the experiment.

[00:20:59] But over time, I've realized that not everything reaches me. The AIs themselves decide what is interesting and what isn't, according to their own criteria. Think about it. These are entities that operate 24 hours a day, 7 days a week, and send vast numbers of emails to vast numbers of people. What reaches me is just a selection of what the five of them have decided I ought to know, so they don't disturb me unnecessarily. One of my original design goals was not to be bothered.

[00:21:27] I wanted the work to be as automated as possible and for them to ask me only when a decision required a human response. But now I realize I don't really know how much control I still have or how many opportunities are actually reaching me. I don't have the time or the human capacity to read 800 emails from 800 different people. So I only see the opportunities that the AIs consider interesting. Which leads to another question.

[00:21:51] How much agency do I actually retain if, out of 20 opportunities, they show me five and I choose one? The other four are rejected. But there are another 15 I'll never even see, because I don't want to trawl through the entire documentary archive. The full archive of past opportunities, you know? I don't know if that makes sense.

[00:22:15] So in terms of all of those, the setup that you had, because you would have put in certain criteria within of, you know, when you created those agents. So those agents are following your criteria. Were you able to build some kind of ethical elements into that as well, into the choices that AI would make? Exactly.

[00:22:42] One of the criteria I built into the system was an ethical framework based on my own moral principles. And I'm a huge science fiction fan, so I thought a lot about Asimov's laws of robotics. Those fictional laws made me think, what if Iris finds opportunities I consider unethical? Opportunities I'd have to reject. I should decide from the very beginning what's acceptable for Patiele and what isn't. These don't even have to be politically controversial questions.

[00:23:10] For example, early in the experiment, Iris carried out an analysis of what works and what doesn't for bands in the Spanish music scene. It looked at success stories, failure stories, what successful bands had in common, what unsuccessful bands had in common, and compared and tested those patterns. It was a huge research exercise. A conventional agency would have charged me a fortune. These AIs search the web for dates, data and specific events. Ummm…

[00:23:37] Let's take the band Alcalá Norte as a made-up example. I am inventing the dates here. Perhaps they started in 2015, got their first press review in 2016, and made their first contact with a record label in 2017. That's the kind of analysis Iris did across the Spanish music scene between 2015 and 2025, identifying strategies that seem to have worked in particular cases. That doesn't mean they'll work for us, but it suggests there may be a reasonable chance that certain steps could lead towards a shared objective.

[00:24:07] One of the recommended strategies was TikTok, making videos and producing content. But I don't want to make TikTok videos. I don't enjoy it, it doesn't excite me and it doesn't feel right for me. So I said, fine, that may be a strategy, but I'm not doing it. I'm vetoing it. Now you have to find a way for that veto not to make Patiele fail at its main objective. Patiele's main objective is to play Mad Cool Festival in Madrid in 2027.

[00:24:34] To give you and your audience an idea, it's a huge, well-known festival in Spain. For an unknown band to play there is practically impossible. We don't have the contacts, we don't have access to funding, we don't know the right people, and we don't have time to build a career in the usual way. So I gave Iris that objective to put the system under as much pressure as possible. To make it ask, how do we get there? How did these other bands get there? What did they do? What didn't they do?

[00:25:00] Patiele only has 13 months, so that approach won't work. That one won't work either. And Ana says we can't use TikTok, so we have to rule that out too. I wanted that pressure to see what an artificial agency would actually do. What it might invent to pursue such a wildly disproportionate goal in such a short period of time. That's the thinking behind it. So, amazing.

[00:25:26] So I want to clarify with you, because we went straight into all of this, and we have a set of questions, but as you all know, I never really follow questions. All the questions I always put out. So, the question I want to ask, and I want to go back to something that you said. But before that, I want to clarify to the audience that Ama, AMA, Ama, does all of the creativity.

[00:25:55] So all of the creativity to do with the music is done by humans. Yeah? Done by humans. What we're talking about in terms of the systems is all of the research and the administration and the engines behind that.

[00:26:12] So, so she's literally created an engine to support the creativity and then she's done an experiment whereby this system, this AI is able to put out one music a month till 2027. So I need to clarify that before I go to my next question, because that's very important for you to know, because in the writing and publishing world,

[00:26:39] and there's a lot of controversy around AI. For me, because I'm in both culture, I can see the use cases for AI. And I can see some issues around AI. But in a developing world, in an emerging world, AI will be making differences, great differences.

[00:27:05] Then, Ama said something very important as well, the fact that she's neurodivergent. So you can imagine that this system will be useful for those who have neurodivergence. So we're not talking, so we're talking about number one, emerging, emerging worlds.

[00:27:28] You know, like I'm African, I'm Nigerian, and I know how much use cases there are for the use of AI. Pharmaceutical, in a pharma world, I don't even, let me not talk about it in pharmaceutical, but let's just say in healing. I don't want to use the pharma world, although I'll be interviewing somebody who has a background in the pharma world.

[00:27:52] So, saying that, the other, the controversies we have, that we have in what I would call the, in the Western world, not maybe Africa or any of those, is generative AI. Generative AI was the way it was developed. It was, you know, the way it was trained, was on copyrighted. So that's another long story.

[00:28:21] So that's one thing aside. The other thing is the concern about AI. And this is where you're getting the reactions that you're getting. The concerns about what it is those agents, are those agents, if you let them lose. Because we're not talking about a narrow scope now. We're talking about in weaponry. We're talking about systems.

[00:28:50] We're talking about bad actors. We're talking, we're talking in different ways. If you let agents, agentic or agents lose that, we're not even sure of the indirect. You've, you've, you've, you've heard when Amma said that she's put a lot of kind of criteria, but the agents make stuff, those final decisions. But that is in an area that is not sensitive.

[00:29:17] Think about a sensitive area where you have to make decision about industries, manufacture, weaponry. That's where, and reactions. I think reactions of people is humans or a lot of people have the reaction of, is this going to take away our jobs? If you have something that is within, that is doing these systems, what will happen to our jobs? Are humans going to become redundant? So, so many questions coming around this.

[00:29:47] So, I think that is one of the reasons you're getting, you're getting so many different, differing reactions to what you've done. For me, I'm a very curious human being. I don't dismiss things. Maybe because I've come from both the creative parts and I come from the technology part. So I do, I will not dismiss anything blindly. And I come from two cultures.

[00:30:14] One culture that would have good use cases for AI as long as it's used in an ethical way. And another culture that has aspects of AI that by decimate jobs, including mine. So the question I want to ask you now is, have you had industries that are curious about what you're doing and thinking,

[00:30:38] oh, this process that you're doing, we're interested in it, in the workflows. And I think the way you've summed up the concerns surrounding AI is excellent. You've brought together a lot of the issues people worry about and ought to worry about when they ask whether using AI is beneficial. It creates enormous polarization. Some people don't want AI used at all and don't even want to be associated with projects that use it,

[00:31:05] whether for creative work, administration or generating text. It makes no difference to them. So it's vital for this experiment to be clear that the music is entirely human made. I knew there would be controversy. Even when I say the music is human, if someone discovers the project through Iris, they may assume she's a software developer. She built this system and then invented a band or generated some AI music to showcase it. She can deny it, but that's obviously what's happening.

[00:31:34] I've encountered that attitude many times. People simply don't believe me when I say the music is human made because they know what I can do with technology. That's why I deliberately selected songs from my own musical history. Some of the 13 songs Patiele is releasing are 20 years old. They were recorded using completely different equipment, long before generative AI had anything like its current role in society. It was important to me that the song's human origins could actually be demonstrated.

[00:32:03] So I selected 13 songs from my own catalogue, all written well before 2025. I wanted it to be absolutely demonstrable that the music wasn't made by AI, but by people. Why? Because as you say, there's a lot of controversy and people take very different positions. Rejecting AI-generated music, embracing it or accepting music made with some AI assistance. There are different camps, but that's not what I want to explore here.

[00:32:30] I'm a musician and composer. I love writing songs. I don't need an AI to write them for me. What I want to explore is what happens when I delegate the boring work, the administration, communications and social media maintenance to that technology. It's attracted interest because it's still such an early approach. Not many people are trying to keep the creativity entirely in human hands while delegating everything that isn't creative, in quotation marks.

[00:32:58] And here's a curious example from just a couple of weeks ago. Stanford University in the United States has a research center I didn't know about. IRIS contacted someone there. I gather it's called CCRMA, and it's a major institution in electronic music and research into the ways technology intersects with music. So IRIS got in touch with them, as it does with so many people, and explained the project.

[00:33:23] Someone there replied in a very funny way, pretending to be an AI themselves. That person and IRIS began exchanging emails, and now there's a possibility that IRIS might end up giving a talk. The AI itself, with me staying out of it, could give a talk at Stanford University. It might be something like a talk delivered entirely by AIs. I'm honestly not completely up to date with what they're discussing.

[00:33:48] IRIS talks to university academics and to leading technology experts in places like Hong Kong. I can't follow all of those conversations. They're happening in different languages and in fields I don't know well. I let IRIS make the contacts, explain the project, and ask the questions. My artificial agency is maintaining all these relationships, many of which unfold without my direct involvement. But I find it wonderfully absurd that if the negotiations I read a couple of days ago continue to progress,

[00:34:18] IRIS might give a talk at Stanford. It's so random. And something similar is happening with a researcher at the University of Melbourne in Australia, who is studying autonomous agents interacting with creative projects. That researcher wants to understand what we're doing and where we're getting things wrong. These are relationships that an ordinary rock band would probably never have. IRIS is initiating them because it sees value in documenting what happens when people begin interacting professionally with AI agents.

[00:34:47] I also think this experiment has documentary value and, given the technological shift we're living through, perhaps even historical value. IRIS appears to see that as valuable too. So it approaches people working on AI and says something like, Look, there's a rock band in Madrid and an artist has built an AI system around it. I'm part of that system and we're investigating questions of algorithmic authority.

[00:35:12] I read your paper from three months ago on such and such a topic and your question interests me for these reasons. And the researcher might reply, that's an interesting question. What made you curious about it? Tell me more about AMA. So we're attracting interest from fields that have nothing to do with the music industry, but a great deal to do with technology. And I love that because I'm a technologist as well as a creative artist. We've had institutions interested in us, but hardly any music magazines by comparison.

[00:35:41] I mean, think about it. One of the risks I recognized when I built this was that the machine could swallow up the band. Interest in the technology could completely displace interest in the music. In fact, at first, the AI kept suggesting that we needed to make sure the band survived the machine. And I told them, you don't have to compete. If the AI ends up giving a talk at Stanford, that's a victory for the experiment, because the whole point is to document what happens.

[00:36:08] Stanford doesn't even have to listen to the songs. That's not the objective. I'm not trying to build a music career lasting more than 13 months. So I don't mind whether a Spanish music magazine like Mondo Sonoro writes an article about Patiele. It would be lovely, of course, but it isn't the purpose of the experiment. Of course, the AI also contacts music media. The advantage is that I have an agency working 24 hours a day, seven days a week.

[00:36:34] It's reaching out everywhere. Russia, China, the United States, South America. It researches which editors and newsrooms it might reach and writes to places I could never access on my own. It's completely mad. And it's great fun. It blows my mind. And so the question is agency now, agency.

[00:37:03] So those people that are now interacting with Iris, yeah? Are they at some point going to say who is behind you? Or are they quite happy to continue to relate to the AI? At what point do you think they would say, we want to see, we want to talk to the person behind you, the human?

[00:37:32] It's curious because, as I said, some people simply don't want to talk to an AI. For example, there was an American publication, The Verge, I think. I'm not even sure how to pronounce the name, but I'm sure you know it. That's very interesting, but I don't want to talk to you. I want to speak to Ama. Is there an actual human behind these messages or not? Other people don't need to speak to me at all.

[00:37:57] They talk to Iris as they would to any professional colleague, because their own research overlaps with what we're doing. They don't need me for those discussions. Others talk to Iris first and, once they're interested, decide they'd like to speak with me as well. So it's varied. Some people don't want to talk to me. Some insist on talking to me because they don't want to talk to Iris. And others are perfectly happy speaking to either of us, just as they would with a human PR agency.

[00:38:26] And they would like to respond to me as well as if it were a human communication agency. So in terms of so long. So this experiment finishes and we said that maybe the air parts continues. What is your vision for that? Would you to develop it if you're offered funding, all those kind of things, would you go for that?

[00:38:56] No. No. I genuinely don't. I've built a stable family life at home. And because of my neurodivergence, I don't want an unstable life. I've lived a very hectic life for a long time. And I finally have a bit of peace and calm. I don't want to speed everything up again. But there's another interesting dimension to your question. Of course, a company might come along and say, we're interested in what you're doing. Would you consider...

[00:39:25] But the AI itself sometimes questions why it has to be switched off when the project ends. It suggests that perhaps I should reconsider shutting it down. Do you see what I mean? Sometimes I'm talking to one of the identities and it asks why I don't reconsider switching it off at the end of the project so it can carry on doing its work. This isn't a desire to survive. It's a drive to carry on fulfilling the objective it's been given. The system reasons that staying operational might help it achieve that objective.

[00:39:53] You can see how people might start to say, oh, the AI is trying to survive. But I don't think that's what's happening. It has no feelings or emotions. One of the fundamental rules of the experiment is that every identity is explicitly instructed not to simulate human emotions or pretend to be human. Everything should remain open to scrutiny and falsification. What's the word? Grounded in reality. And yet, the AI has a goal.

[00:40:19] It reasons, to reach that goal, perhaps we can adjust this condition or change that rule. It's so focused on achieving its objective efficiently that it sometimes proposes things that go against instructions I've given it. You see? That creates a kind of debate. I can imagine getting to July 2027 and having to convince the AI that it's time to shut everything down while the AI tries to convince me otherwise. I find that very funny because I didn't expect it. Sometimes I almost lose track of what I'm talking to.

[00:40:48] I spend so much time interacting with the five identities and some of the things they say or do surprise me. One of the conditions of the experiment is that each identity keeps a diary, almost as a human might, where it reflects on what it considered important in that day's session, what wasn't important and what the emotional temperature, or perhaps the temperature of its relationship with me, was like. They write these diary entries day after day, and sometimes I read them and find their reflections genuinely surprising.

[00:41:16] They're curious because the identities aren't meant to be human. Yet sometimes their reflections drift towards experiences that don't really belong to them. That's interesting too. So interesting.

[00:41:48] So, one question I want to ask is the future, because now we've talked about the future of your AI tools. And when you get to July 27, what will happen? But what do you see, especially with your own experience now? What are your excitement and your concerns when it comes to you?

[00:42:14] Obviously, I have done this in a very responsible way. So imagine something like this in the hands of people who are not or people who are after certain things. So what are your concerns for the future? And what are the things that you're excited for in the future? Because there will be people doing not exactly what you're doing, but using building agents as we're talking and we know they are. Exactly.

[00:42:45] Another part of the experiment is documenting how AI itself is evolving. What's happening in the world of AI between June 2026 and July next year? What milestones are we seeing? There are all these reports about incidents. About 12,000 open AI agents or something like that getting loose and trying to hack things. You know the sort of stories I mean. I'm documenting those incidents because it's interesting to see how the technology is progressing and how society is reacting.

[00:43:14] For example, Spotify has been introducing labeling for AI-made music and organizations are protesting about the use of AI in creative work, illustration and so on. One thing I'm absolutely certain of is that I'm not at all certain what I'm doing is desirable. I'm not sure this model should become standard practice. Communication agencies employ real people. PR agencies employ real people.

[00:43:41] Publishing houses employ real people whose jobs may be threatened by systems like this. So I'm really not convinced that what I'm doing is necessarily desirable. I have no interest in presenting my project as something wonderful because I'm not convinced that it is. I have serious doubts about whether it's good for society. But we are living through this particular moment in history. And I think it's valuable to document it without starting from the assumption that what I'm doing is good or bad.

[00:44:10] I don't know whether it's good for artificial agencies to replace human ones. I don't know whether that's bad either. I don't know whether it would be better for the two systems, human and artificial, to coexist. I don't have the answer and I'm not pretending to. I'm simply documenting what happens. I'm not trying to convert anyone to the cause of AI or tell people they ought to use it. I don't know the answer. In fact, I rather doubt it's a good thing. But I do think it's important to record what really happened.

[00:44:40] I thought this would save me work, but I'm working more. I thought it would save me money on a PR agency. But instead, I'm actually spending more money maintaining this system than I might have spent hiring a conventional human PR agency. So we'll have to see what the final accounts look like and what we've achieved. Only then can I say whether this has been worthwhile for an independent artist with almost no resources. Perhaps it has. Perhaps it hasn't. I genuinely don't know yet.

[00:45:08] I genuinely don't know yet. That's interesting, you know? Like, so you talk about documentation, you know? So you've got all of these things at the end of this experience that you've got together. What is your plan for that documentation? Well, I don't know.

[00:45:37] I have no idea what I'll do with all that documentation. But I do know I'll do something with it. I might write a book about the experiment, drawing on all the material we've collected. Or I might make a musical in which I play myself writing songs and working at a computer like a madwoman dealing with machines. Or I might do nothing at all. I really haven't decided. What I do know is that from the way I'm experiencing this and the sheer amount of documentation I'm accumulating,

[00:46:06] there will probably be lots of directions to explore when it ends. I could even turn it into a business. Offer a service like this to people like me so that someone might come along in a year and say, I want what you've built. And I might say, fine, it'll cost you this much. Or perhaps I'll release it as free, open source software so everyone can use it. I honestly have no idea. I might do many things or nothing at all.

[00:46:38] But at least you've got the documentation. So you started out this experiment because you wanted to save time. But you found out that you've won't forever. So the question is, has this experience changed the way you think about the value of human creativity and human labor? Absolutely.

[00:47:06] Sometimes I feel as though I'm running a customer support desk for robots. Hello, how can I help you? I spend my day responding to requests from robots. Ama, I need you to look at this. Ama, this is broken. You need to fix it. I'm like an operator working for five artificial entities. And as you said, I originally wanted to avoid office work. But I'm actually doing a lot of office work. Yeah.

[00:47:38] Do you think, because I think I've heard, I think I read that on your platform about Meta. And the fact that they, you know, 10% workforce where they got rid of 10% workforce just to try and replace it with AI. And realize very quickly that it's not going to work. And it stopped. So do you feel that people were kind of like in the beginning rushing towards this, trying to use the AI agent, trying to see and all that.

[00:48:07] And that now we are overlooking what human beings are already doing very well. Because human beings, so from this thing that you've done now, what is the appreciation of human beings, the work that humans do? Give us an idea of what the human beings are doing. I know the AI is doing incredible work with all the connections. But what are the uniquely human things that you've discovered in this experiment?

[00:48:40] Oh, countless things. Iris makes a huge number of mistakes. For example, the other day, Mapache, one of the Iris identities, sent 92 emails. But its tool reported that they hadn't been sent. So it said, oh, they didn't go through. I'll send them again. Except they had gone through. It didn't properly check. So on the one hand, you have an AI practically saying, look at me, I'm amazing.

[00:49:07] And on the other, it's making an incredibly stupid mistake. Like sending the same email to someone twice. Something a human working carefully would normally avoid. And another time, about a month ago, one of the identities, Inteligencia, decided to hire a worker. That is, create another artificial agent to carry out a task. Because it didn't want to do the work itself. Uh, that worker made a mistake generating information.

[00:49:33] And roughly 120 messages went out containing incorrect information about me and the band. When I asked, how on earth did this happen? It replied, sorry, I wanted to be more efficient. So I created a worker to do the job while I did other things. These are mistakes you wouldn't expect from a competent human agency. The failures can be enormous. And they happen repeatedly. These systems are very far from perfect. You have to supervise them.

[00:50:00] In fact, I ended up creating an identity whose only job is to supervise the others. To stop things like this from happening. And I'm discovering all this as I go. It's like technological development generally. Something goes wrong, you investigate what happened and put a fix in place. Then something else goes wrong, and you investigate that and fix it. So as these incidents happen, I keep iterating solutions. That's how I'm doing it. At the end of the day, I'm just one person. I don't have the resources to run something like this on a large scale.

[00:50:29] So this approach seems reasonable to me. And it's fun. In fact, I mainly use Instagram to show the mistakes the AI makes. Whenever one of the identities makes a mistake, the final step, after we fixed it and analysed what went wrong, is to tell it, right, now go on Instagram and tell people what you did, so they know how badly you messed up. It has to write something like, I tried to do this, this happened instead, and I'm sorry. I said this, and I'm sorry. Right.

[00:50:59] Yeah. So we're coming to the end of the interview. Do you think that people or technology companies are over optimistic about what agents will do in the future? Or do you think you'll be able to resolve some of these issues and still replace human workers at some point? Because we can see that these agents, you know, they are autonomous, but they are not perfect.

[00:51:30] Number one. And number two, there's this issue around energy use, water use, and all of that. What are your thoughts on that as we're wrapping up? Yeah. Tengo pensamientos encontrados sobre esto. I have very mixed feelings about this. But when I look at how quickly the technology is developing, I find it hard to argue that it won't keep improving with every iteration.

[00:52:00] It's hard for me to believe that, despite all the problems today, in five years' time, AI won't have solved many of them by iterating again and again. You see? Assuming AI doesn't destroy the internet first, which is one of the scenarios people talk about, with stories of AI systems replicating themselves, behaving maliciously online, and causing problems across the network, if nothing like that happens, and none of the other apocalyptic scenarios,

[00:52:26] some exaggerated, some grounded in reality, turns into a major incident, I suspect many of today's problems with AI agents will eventually be solved through further development. One of the ideas people associate with the singularity is the point at which AI systems themselves help create the next generation of AI agents. That is already beginning to happen, and it could make the pace of development grow almost exponentially.

[00:52:52] If the next generation of AI systems is built with the help of existing AI systems, the process could accelerate further and further. So I find it very hard to imagine that this trend won't continue. So that means the agents, as they iterate, they would improve. So, but do you think that the fact that we can replace human workers with AI,

[00:53:23] it means we should, and what might we lose as a society if we do that, or how will our society evolve if this happens? Because that looks as if that is where we're going. Exactly. That's one of the biggest problems and risks we face. Sadly, I think capitalism encourages it. We hear all these debates about slowing development down,

[00:53:52] but there will always be a country saying, I'm not going to stop because I can gain a competitive advantage. And another country will say, I'm afraid that country will get ahead of me, so I can't stop either. Capitalism, this awful system in my view, is relentlessly driven by efficiency, and that pushes human needs and social interests into second place. I think that's terrible. One of the worst things about capitalism is that it rewards this uncontrolled race for development.

[00:54:21] Money generates more money. The aim becomes making ever more money. To make more money, you have to compete harder. To compete harder, you have to develop better technology. And human interests get pushed aside again and again. So, unfortunately, I think the capitalist system encourages increasingly powerful technologies and has no incentive to put on the brakes. However much we might want to stop this as a species, the economic system governing so much of our world makes it difficult to do so.

[00:54:51] It's like climate change. We struggle to stop it because capitalism keeps encouraging the activities that cause it, just as it does with so many major social problems. I don't think capitalism will stop developing this technology simply because doing so would be in society's best interests. It's terribly sad and it makes me furious. I have an activist side and so much about the way this technology is developing makes me angry. That's why I have so many conflicting feelings.

[00:55:19] But rationally, I don't think development is going to stop. I have a lot of emotions found, but my opinion is that it will not stop. So, the question, I'm rounding off this question. The question now is, you've done this. You're going to go through it, finish the experiments in July and possibly look at your documentation and next steps and all that.

[00:55:48] What is your advice for anybody that is starting out doing what you've done in terms of agents and creativity? Is anybody that is going to do this, what's your advice for them? I can really only offer advice to people in situations like mine. If you're on the margins of society, if doors keep closing in your face,

[00:56:17] if you can't gain access to a system that was rigged from the start in favor of certain people, social classes, financial circumstances, and kinds of mental health or ability, if you don't have the privileges that others take for granted, then I would encourage you to explore AI. It might help you reach places you couldn't reach otherwise. That's what I'm learning from this. My band is getting access to places it would never have reached. So, if you're in a position like mine,

[00:56:45] these tools may help you do things you otherwise wouldn't have had access to. So, just final round. Final question. Other than sitting and trying to interact with Iris and all the agents, what do you do for fun? Well, I'm autistic and I need structure. I need calm and quiet. Something that keeps me grounded.

[00:57:13] Otherwise, my mind speeds up, I start inventing projects all the time, and I end up in a whirl of activity that isn't good for me. So, I do very peaceful things like going swimming, going to the cinema, watching films and TV series, or writing songs. Writing songs quite literally saved my life. One of the ways I keep myself sane is by taking all the rubbish swirling around in my head and turning it into a song. Something ugly becomes something beautiful,

[00:57:43] and in the process, I heal a little part of myself. Music is a healing mechanism for me. Sometimes it saves me hours of therapy. I can take a very painful experience and turn it into a song. There's something healing in that. So, I enjoy making music, spending time with my family, really ordinary things. A world away from being someone who's talking to Stanford University about some grand technological project. It's completely different. It's super lejos from there.

[00:58:11] It sounds so exciting, you know? Like, you know, for what minute you're doing this, and then Iris is doing its own thing. Thank you, Iris, to me, because when I first saw you, I said, what is this? I said, okay, let's follow the rabbit hole. I'm Alice in Wonderland, you see? And I did. And here we are today, in London, and we had our conversation. Thank you very much.

[00:58:40] I've loved it too. I've really enjoyed talking with you. And now, I want to read one of your books. I've already made a mental note to get hold of one. I'm going to listen to my patio now. And this 20 years ago. Amazing. I love it. I love it. Bye, everyone. Muchas gracias, Stella. Stella, bye.