Fleur Prince: Hey everyone, welcome back to AI Insights. ⁓ I'm your host for today, Fleur Prince, and I'm here together with my co-host, Granya.

Grainne Popen: Hi everyone, this is Gronya.

Fleur Prince: And we're going to talk today about everything that's happened past few months in AI, because a lot has happened. And this is also going to be our wrap up for season three. We've had a lot of interviews happening. But we're going to talk about everything that's been happening the past three months during season three, and also get you up to speed on everything that's happening in the world of AI.

Grainne Popen: Exactly, and this will be our last episode of season 3 and we are already looking forward to season 4 which is going to release next week so get excited!

Fleur Prince: Yeah, I can't believe that we're already in season three. We're just talking about it. I mean, the podcast started, well, you said like two years ago now. I thought it was a year ago.

Grainne Popen: Yeah, know. Two years ago, it's been crazy. And honestly, there's like there's just been so many changes like in that time. What's going to be really cool is like, towards the wrap up of season four, we should go back and listen to our first ever episode and maybe even do a discussion of that and what's changed, not just like, you know, since AI, this AI kind of advancement started happening and like the AI bubble, et cetera, but like since we started this, you know, which is going to be, it's just wild that we've been doing it so long that we have that frame of reference.

Fleur Prince: No, exactly. And for me, it still feels weird that it's just been two years this podcast. I mean, we've grown down to like 3,500 listeners on our podcast. We're both doing great when comes to the world of AI.

Grainne Popen: Yeah Yeah, exactly. A lot has changed like career wise as well. ⁓ my gosh. And we got nominated for that award along the way. ⁓ yeah.

Fleur Prince: ⁓ yeah, best podcast, AI podcast of the year.

Grainne Popen: Yeah. my gosh. Yeah. So a lot to reflect on as we wrap up season three and a lot to reflect on also specifically this week. So maybe I'll just dive right into it. So, just with regards to what we're trying to focus on this episode is we're looking at the agency pivot. So we're looking at, you know, kind of Sora dying. That's a major update from this week and agents taking over. Yeah. Spoiler.

Fleur Prince: Poison.

Grainne Popen: Well, it's the title. It's the title. And agents taking over. were even discussing this kind of before the podcast started, just, you know, how agents are rapidly advancing. Even in my work, I had a panel week about autonomous agents and what that looks like in law. So, yeah, I mean, there's definitely an evolving frontier and we're going to look at, you know, those evolutions within this week.

Fleur Prince: and I this week is the week of the Triple Crown ⁓ Open Google and Xeo all dropped their new models I mean, did you see GPT 5.4 Gemini 3.1 and Grok 4.2? Like ⁓ what for me it was like a complete revelation. Okay, like what is happening when it comes to all these models dropping?

Grainne Popen: Mm-hmm. ⁓ huh. Yeah, yeah, it was back to back to back. mean, obviously, there's kind of a tension where these companies are racing to remain the most relevant. And because AI is like such, you know, there's no monopoly, I would say for now, obviously, we've kind of seen this dynamic emerge, specifically in the US between consumer usage of Claude AI and chat to BT. But, you know, that's not the only AI solution out there. I think actually that You know, AI is kind of dispersed in so far as people use different solutions and you know, not only do people use different companies, but people also create their own models like different companies, for example, law firms develop their own models. so yeah, I mean, these companies are definitely going to have to continuously update their models, provide kind of the best content. We also saw that in something that we've nearly, you know, over on the podcast, which see the rise of deep sea and now that affected the industry. So.

Fleur Prince: Thank

Grainne Popen: Yeah, you know, mean, obviously I think that they will sort of advancement is necessary that they're going to need to keep on, ⁓ know, ⁓ these updates as rapidly as they are. And I think it'll be also kind of compelling to consider what the differences in the updates are. And, you if they're kind of, you know, pitching off each other or if they're growing in different directions, I think that's going to be an interesting analysis.

Fleur Prince: No, exactly. We're also looking at agentic realism happening here. mean, the industry is moving away from like the expensive toys. I mean, we already talked about you already did the spoiler of the death of Sora. But people are moving more towards tools that actually really do work. really MCP agents on the note of Sora dying. ⁓ I mean, the announcement for me felt like on March 24th, it came out of nowhere.

Grainne Popen: Yeah. Okay.

Fleur Prince: when they told us like we're shutting down Sora and the API. don't know if you had the same feeling around that.

Grainne Popen: Yeah, I mean, especially because Sora was so iconic, honestly, like, I don't think there's any other word for it. I mean, it was was during the time that we started recording the podcast that AI generation became ⁓ ⁓ consumer accessible. So like. You know in the world before our podcast and you know, I'm only I'm only using our podcast as the measuring I'm only using our podcast is like the measuring yardstick because I don't remember the exact month You know, I almost think it'd be difficult to kind of pinpoint an exact month that it became mainstream But yeah, no Sora kind of represented this era where chat BT like images AI generated images

Fleur Prince: Four.

Grainne Popen: became mainstream where, you know, like you go on Instagram, you go on any social media and you know, your friends are able to AI generate images. So it is very interesting to kind of see an of that era kind of, you know, be discontinued. But, you know, I guess it kind of begs the question of what's next.

Fleur Prince: true but also like the reason why it was taken down is not per se because it was a technical failure but because of the compute economics because apparently generating a single second of video is exponentially more expensive than text and with the IPO looming open AI is like cutting the fat to stay lean and also the Disney fallout that happened as well where they lost like pretty much like a 1 billion dollar deal

Grainne Popen: Mm-hmm. Yeah. ⁓ my gosh, yeah.

Fleur Prince: Yeah. But the cool thing is also is that Sora is not entirely dead, but they are reassigning it to world simulations for robotics. Which is cool, but still sad. Now I cannot make my funny ⁓ dancing animal videos anymore.

Grainne Popen: Mm-hmm. yeah yeah You know, it's crazy. remember Sora when it first came out, it wasn't available in the EU because I remember my mom of all people sent me Sora and was like, you can make an image on this. And I was like, really? That's so cool. I click on it and it's like, this is not available in your region. I was like, are you serious? because like all these tech companies, you know, are headquartered in the U S I sorry, in the, ⁓ And I was in Arland at the time. So I was like, surely they'd make it available here. But anyway, ⁓ Yeah, no. I mean, kind of as we've discussed, is this iconic feature of the evolving forefront of AI and how AI is specifically posited to consumers. And it's not dead, but it is being reassigned away from that kind of consumer focus to world simulation for robotics. And I think the broader takeaway here is that we've kind of reached a limit. And this is something we've actually not really had an opportunity to discuss in depth on the podcast and something that I think is going to be a major theme as we're coming up to season four, which is we're reaching a limit of infinite compute. So companies are now having to choose between, you know, cool viral videos and making things like hyper available for consumers and functional enterprise tools. So, you know, generating AI images, generating video, it takes an inordinate amount of compute power, resources, et cetera. I mean, we actually have discussed that aspect with regards to, you know, like water consumption. What's the effect on ⁓ the We have an episode dedicated to that. But now we're kind of seeing the ⁓ other side that dynamic, you know, ⁓ the flip of the coin, which is that companies now need to make practical choices ⁓ based on compute limitation. and the practical effects of the compute limitation that they're seeking to limit.

Fleur Prince: No, exactly. You also see a lot of investment firms are now asking for, we invested a lot of money into it, like few billions. But what did we actually get out of it? Did we become more efficient? Did we gain more influence and earn more money? And then the basic outcome of those reports have been, you for example, look up the McKinsey report is that, yes, they have become 20 % more efficient, but when it comes to like, increases and it's not that much.

Grainne Popen: Yeah. Yeah, no, completely. Yeah, it'll be interesting to see how that progresses.

Fleur Prince: Exactly. think it's good also for us to move on to the next topic, which is the model context protocol revolution. So MCP for short, which I was talking about earlier. ⁓ we see the concept of MCP that has hit a tipping point with nearly like 100 million installs.

Grainne Popen: Mm-hmm. Yeah.

Fleur Prince: it explains that the MCP as the universal port that lets AI such as like, Cloud, GPT, and Gemini plug into your files, your calendar, and your company database without custom code, ⁓ taking on like a very new life. I mean, if can like find out like why it matters is because the infrastructure allows Cloud new computer use. ⁓ by the way, I tried this. It's awesome. ⁓ feature actually work in a professional setting.

Grainne Popen: Mm-hmm. Really?

Fleur Prince: It's like the invisible glue I think also of ⁓ 2026 AI ⁓ because ⁓ no longer have to be at your computer to use AI. You can just like send it a DM if you connected it to your phone and tell it like get me my files ⁓ my PowerPoint I forgot to upload it and that it just goes through your computer gets the files sends it to you.

Grainne Popen: Yeah. I mean, kind of those like little, I mean, this isn't like little, but like the, you know, kind of the, the streamlining of UI and kind of the advancement towards integration of AI and professional services. mean, there's like, you know, you said yourself, it's a tipping point. There's no going back from that. swiftly onwards. Yeah, so I was trying to create these Figma mockups and there was just, you know, kind of there was an absence of practicality in so far as, know, I couldn't convert like the AI generated kind of template to a different workable like environment for me to edit it outside of And there were just like some complications ⁓ and ⁓ you know, like the AI itself, was like, ⁓ you know, like if this was just a little bit more streamlined and also, you know, like the, prompting took so long. I, that's the thing, like as UI, advances, ⁓ think we're going to see huge, huge, huge kind of, adaptation people taking it on. So that's kind of my personal experience from this week that, you know, I would highlight in relation to the MCP revolution.

Fleur Prince: that also brings me to, ⁓ I don't know if you've seen Vibe Design and like the super app shift that's been happening. ⁓ what's really cool is Google it's the launch. So Google ⁓ they just Vibe Design and they're moving like from pixel pushing to intent based interfaces. So ⁓ because yeah, looking more towards

Grainne Popen: Mm-hmm. Mm-hmm.

Fleur Prince: How do you call like, yeah, these interfaces that are really focused on the user experience compared to like, okay, pixel pushing, but also Apple series openness that's happening. there's a global shift as they're happening as well. Oh, because they are actually using Gemini in iOS 27, which is really cool.

Grainne Popen: Mm-hmm. Yeah, I mean, that's another kind of fascinating, you know, interplay that we're seeing from AI development, which is that we have these existing AI companies, right? Or rather, sorry, we have these existing tech companies. AI is unavoidable. every single kind of company, professional services, products that like, no matter where you are in the industry, you're going to interact with AI somehow. And these major tech companies all are trying to win this race. And we were discussing that earlier with regards to the updates and trying to push more updates, et cetera. But another thing that we're kind of seeing is, like this move away nearly from, you know, we were discussing compute process and trying to have the best model to also, you know, kind of just strategic strategic business moves. So that's what I would kind of highlight this as, you know, Apple allowing Siri to access Gemini and Claude with Gemini and Claude being two powerhouses. It represents that companies, you know, that aren't that don't see themselves as being able to compete in a technical way ⁓ are to put themselves forward in different ways using kind of like business expertise to kind of bridge that gap between the lack of innovation and industry. own lack of innovation or their own ⁓ of inability to ⁓ produce competitive model. They're staying competitive with smart business practice and kind of unique business posturing.

Fleur Prince: No, true. And what you do also see is that there's pretty much like a competition now going on indeed about like which is the best AI model to use ⁓ for my specific issue. But actually, when you look at WeChat back in the main app that's being used in China, they turned like every service on their application into an AI interaction. So they're not actually competing with each other. They're just like, we just put it all in here. So instead of it being a competition, They're just like, okay, we're just working on efficiency instead of trying to make a profit out of it. Though I also say always with WeChat, also there the government always looks with you what you're doing on WeChat. So I don't know if it's that much better.

Grainne Popen: Yeah. Yeah, no, completely agree.

Fleur Prince: I think that's also maybe good discussion as well. Like in the West, the App Store model is kind of dying in favor of like the conversational layer. Yeah, I don't like I love that companies are competing because that just gives competition the more fuel to become the best application that they're possible. But I'm also looking at the convenience part because I at certain point, like at 20

Grainne Popen: Mm-hmm. Yeah.

Fleur Prince: subscription per tool is just going to be too expensive for the average person.

Grainne Popen: Yeah, mean, Claude is actually like extremely expensive.

Fleur Prince: No, exactly. like if you're a tech person, you know how to download an open source model and do it all for free. But if you're not,

Grainne Popen: Yeah, and rig it up. Yeah. And another thing is even if you're not super tech heavy, you know, the point of all these different IEI tools is that information is, you know, you can learn anything and you can have it explained to you as though you're a toddler. Like, you know what I mean? Like, there's such a like, even if you're not tech heavy, you don't know how to download the API, you can definitely figure it out. You know what I mean? So it's kind of, yeah, I do see kind of this this shift.

Fleur Prince: No, and I think you're also going to see a shift where there's going to come where because OpenAI is now already on the decline. I don't want to say per se on the decline, but they're behind when it comes to Claude and Anthropic, of Google, where need like we talked about earlier, they have to start making choices where it's not about just investing money and hoping that it works, but actually making the practical choices.

Grainne Popen: Yeah.

Fleur Prince: But what I would also say that if they are still looking at making it consumer apps, they do have to think about how do we make sure that people can still pay for it?

Grainne Popen: Yeah, that's true. I definitely hear you on that. And it'll be interesting to kind of see how that evolves as we move forward. And speaking of moving forward, I'm excited to kind of look at the next segment, which is the regulatory hammer. ⁓ And think maybe even kind of the most pertinent development out of, you know, everything that we've kind of discussed so far is that ⁓ the AI Act draft to is has just been proposed. So the European Commission has just released the second draft of the code of practice. I mean, it's almost difficult to contextualize the extent to which this industry, not just industry, consumers, every single every aspect of kind of AI regulation, because the EU AI Act is the default benchmark for the entire world and countries and their kind of protocols and their their frameworks to regulate AI. are nearly all predicated off of EUAI, the EUAI Act. Even I would kind of take it to a more kind of core level in identifying that the EUAI Act has a risk-based model. Now the risk-based model is the found, more than foundational. It's the starting point for any conversation about not only compliance, but just adaptation. So I, and you can, and I'll, you know, I'll briefly reference a physical manifestation of that influence through the, through Texas's Triga Act, which was ⁓ just put into force January 1st of 2026. And that was almost like, if you read the act, can, you know, it's just, it's evident on face just how similar it is to the EU AI Act. And it was actually built off of the EU AI Act. So, you know, as we see the EU AI Act evolve, we're going to see other frameworks evolve to kind of, you know, match it. And part of the mandates includes mandatory AI watermarking, ⁓ universal EU AI ⁓ for all generated content, and the conflict that we're going to see here is that these rules are going to go into effect in August, 2026. That's very, very soon. That's actually four months away. ⁓ 2026 is flying by, yeah, companies are gonna have four months to figure out how to label every ⁓ single AI assisted text and image without ruining the user experience. Fleur, how do you see that kind of developing? What are your thoughts on that?

Fleur Prince: Yeah. I think they're going to very ⁓ that they did not start earlier. They're going to be wondering what's going to happen. But also, when look at this, think four months is reasonable when it comes to the facts. OK, how can we make sure that everything is labeled? Because all they're asking is for you to put a watermark on it, ⁓ Google already showed how you can do this, for example, in the corner of every AI generated image. There's the Gemini logo.

Grainne Popen: Yeah. Yeah. Mm-hmm.

Fleur Prince: So pretty much what I see is that they just have to implement that into the prompts that every time an image is generated, this watermark needs to be in it. And if people cut it out afterwards, that's not on them.

Grainne Popen: Yeah. Yeah, I mean, it's compelling. So this nearly gets into like ISP liability. So, you kind of when the isn't complied with, right, but it's consumer generated. say, for example, I'm AI generating an image, I'm throwing it online. Who's liable there? Is it me? Is it the service provider? I mean, we're kind of even seeing this with regards to defamation. ⁓ I was looking recently, Ireland has a defamation act that regulates, ⁓ you know, kind of. Everything with regard to defamation. And it was looked at in the court case of ⁓ Taros, which is an Australian case, whether or not a hyperlink or, know, like a, like a, like a little, you know, how like different search results will come up on Google and you'll see like a little like pop up. now we the AI generated overview, but anyway, there, defteros, the court case kind of looked at. and considered, okay, if it's coming up on like a little hyperlink, if it's coming up in, you know, like a explanation on the first page in Google, ⁓ that defamation? ⁓ we're going to have to look at the evolution of tensions. And we're gonna have to consider, okay, this AI generated image was produced by an individual put on this platform. And then maybe the aggrieved or rather victim or anyone who is affected by the image has contacted Google to take it down. They haven't. To what extent is Google then liable for that image? And to what extent is the consumer or suppose the better word is, I don't know. Yeah, it's just the individual. what extent are they going to be liable? The thing is that the thing that makes us so interesting is that it's so unprecedented, right? Like we... the, it's going to, it's going to oscillate. and we're going to see kind of different decisions, as different countries have different perspectives. but you know, now we kind of have the setting for this and I guess we'll ⁓ watch it go in August, 2026 and definitely talk it over. ⁓

Fleur Prince: No, definitely. I'm though I'm what I'm seeing is that they're just going to sneak it somewhere into like the terms of conditions when you apply when you apply to use their tool. So Gemini, if it's Claude or an other open AI tool, it's just going to be somewhere in the terms and conditions that if you are using their services, they are not going to be responsible what you do with it. And that all the liability is going to be on you. That's what I would imagine a company doing.

Grainne Popen: Mm-hmm. Yeah, completely. But I, you know, I guess that remains to be seen. It's also, you know, yeah, I mean, it's just fascinating how regulation is advancing. mean, it's really only comparable to like the advent of the Internet, where the law was like racing to kind of, you know, nearly not compensate, but the law was racing to to understand even what the Internet was far beyond, you know, just regulating it. So I think we're kind of seeing that now and the EU AI Act is kind of the Bible for regulation. So it's very interesting that they've, you continuing to update it and, know, that'll just change industry. Another kind of change that we saw is the energy level levy. So there's a growing debate about the EU taxing data centers to pay for the triple transition, which is AI energy and geopolitics. we're all, ⁓ Obviously living in the EU, we are no stranger to European countries appreciation of tax, to put it in one way. But yeah, so there's kind of a growing conversation about the EU taxing data centers, and this could potentially be adapted into the EU AI Act, although there will be profound, profound resistance from tech companies. And again, all of this will be considered in August, 2026 as the adaptations are put into place and we will be on the ground reporting about it as usual. So what'd you say? We're not going anywhere. Oh my gosh. Yeah. And then.

Fleur Prince: Yeah, we'll be right there. I think it's also a good conclusion to the road on where we're going. The big picture now has been, indeed, Sora shutting down, the focus turning towards agents. Also the pivot to robotics, where AI such as Sora is now being used for world building.

Grainne Popen: Yeah. Yeah, for robotics. Yeah. So it's kind of, it's that transition between, know, we have this hyper powerful tool. going to put it towards robotics and industry, not consumer facing.

Fleur Prince: No, exactly. yeah, it's more going to be okay. It has to be practical for us to make profit otherwise we're not gonna do it.

Grainne Popen: Yeah. And so it has to be practical. And we're also seeing, as we've just discussed, that it has to be regulated and it has to fit those regulations. Actually, you know, it's a super, like it's a very compelling kind of business strategy to have shut Sora down directly before these, you know, new regulations are being imposed because that reduces the amount of liability. so intensely, you know, like while other companies scrambling having board meetings on boarding board meetings right now and trying to develop these strategies, ⁓ know, is not going to have those same considerations because it's no longer consumer facing. So again, it's just interesting seeing, you know, that interplay between industry, what it looks like regulation and innovation.

Fleur Prince: Yeah, and I think also a good closing is, mean, remember back in 2024, we ⁓ about AI ⁓ if it could write a poem. Now it's turning towards like, AI do our code? Can it file our taxes? Stay within the power grid limits? We're really ⁓ walking away the honeymoon phase and really going towards like the workday has begun. AI is going to be part of our lives if we want it or not. ⁓

Grainne Popen: Yeah, and I would really kind of emphasize that the workday has begun. I don't think there's a single really kind of, don't think there's a single industry that AI hasn't touched in some way or form, even, you know, even coffee shops, ⁓ coffee shops, like in kind of my general area, they would be using AI generated images for their art, for their marketing. You know what I mean? So, as we've seen, you know, even that sort of use of AI can have, you know, interplays with the broader industry, for example, the EU AI Act and the new limitations on AI generated images. So I completely agree with you. And I think that's a powerful closing thought, ⁓ which that AI is no longer trying to impress with creativity. ⁓ It's and testing utility.

Fleur Prince: Yeah, that is actually a great sentence. ⁓

Grainne Popen: Thank you.

Fleur Prince: It's good for us then now to close off the podcast. So I want to thank everyone who is listening to a podcast. Hope it helped you this week to get up to date everything that's happening and also us closing season three. I to thank all the guests that have been on the show. It was really awesome having you guys. And if want to listen to our past conversations with guests that were on the podcast, please check out our podcast on Spotify, Amazon Music, iHeartRadio. Apple Podcasts and a bunch of other channels that I just added. YouTube as well by the way we're also now a YouTube podcast.

Grainne Popen: Yep. multi-platform, multi-platform, multi-disciplinary, multi-platform.

Fleur Prince: No, definitely. Anyways, thank you guys for listening and see you all next week at the opening of season four. ⁓

Grainne Popen: Yes! Get excited!