The Rise of AI Agents: Do They Really Work? | EBC11

Transcript

So um you hear it on my voice. I was networking hard the last two days. So um welcome everyone and congratulations. You made it to the very last panel of the entire conference. So you've survived full days of intense discussions. I did keynotes and probably far too much coffee. So give yourself a quick round of applause for that, right? You can do better, I know, but we will figure it out. So um our topic today might sound like one of the most brainheavy of them all. So the rise of AI agents, do they really work? Um but

don't worry we uh with the brilliant minds we have on stage we'll make it insightful a bit provocative maybe and most importantly human and engaging. So think of this more like a conversation at a bar then and then at a final exam. So before we dive into introductions I'd like to start with a quick um warm-up round. I will ask each of you to answer in maybe one word yes no or maybe it's do AI agents already work today >> maybe you first >> maybe >> yes yes >> yes

>> depends can I say that [laughter] okay thank you so um perfect that already gives us a sense of where everyone is u coming from so now before we unpack what's behind those answers. Let's quickly get to know who's up here. So, please keep it short and natural. Maybe just a few words on who you are and what you're working on in the context of AI agents. Maybe you first. >> Okay. Hi everyone. I'm Asim. I'm co-founder CEO of Kernel Labs. We've built a product called Openbox AI which is focused on solving the blackbox problem of AI and opening it up so that it's more transparent, governable, and uh trustworthy. So we're working within the trustworthy AI space.

>> Hello everyone. My name is Max Lee. I'm the founder and the CEO of ORT. Uh ORT is trying to build the world's largest AI data collection platform. Actually this is a startup spin off from my Columbia uh university lab. I'm a professor at the Columbia University with uh you know with electrical engineering department. My research interests are AI and blockchain. Hello everyone and apologies for my fake picture. I thought this is the style. Didn't realize I was the only one but I'm a real person not an AI agent. And I'm David CEO and co-founder of a project called AMO. So in a simple term is that we bring AI agents into group chats. So not as a boss but as a

teammate in Telegram in Slack in Discord. And prior to that I was an engineering manager in Google for about a decade. So an early member of Google app >> I like your your image and your agent. So in my case I've been an entrepreneur in the crypto space for many years and since the past 3 years we focused on education. So we are the leading educational platform on tech focused on Spanish >> and I'm Naruta h CTO for giga at UNICEF. Many of you maybe might be surprised of having UNICEF in such in a blockchain conference and in a panel on AI. Um Giga it's a partnership between UNICEF and

ITU to connect schools to the internet globally. Um we want to connect every school by 2030. And for that uh one thing that we do will really leverage open data and open source technologies uh to to accelerate the process uh make the the whole process more efficient and transparent. Um and these technologies include blockchain as well as AI. >> Thank you very much. And for the context, my name is Danila Bubak representing the blockchain Bundis Fabant aka Bundesblock, Germany's blockchain association. So the voice of the German um uh blockchain community.

So now that we know who's here, let's [clears throat] get into it. AI agents. I would maybe bring up uh the definition of what an AI agent actually is so that everyone is on the same page here. And maybe you can correct me then if I say something wrong. So um an AI agent is like a smart little helper that can think, learn, and do tasks for you on its own. Kind of like a digital assistant that figures things out without always being told exactly what to do. Agree.

Nice. [laughter] Maybe one thing I would say on that is that depends also on the definition, right? Like AI agents emerged in the 70s 80s already and at that point they were AI tools to do a specific or solve very specific tasks and I think that now when we refer to AI aents we are referring to LLM um based AI agents. So I think it's worth also noting that difference. >> Thank you. So um my first question would be what's on uh one concrete task you've seen an agent do that actually impressed you and what's one thing agents absolutely fail at right now.

So who wants to start? >> I can go. So I'll I'll try to focus on the web three because when you talk about AI agents it's you can apply them everywhere and sometimes they work depending on what you wanted to do sometimes they don't and one thing I thought was amazing was the recently like they um yeah it was published from Anthropic like there it was an agent able to code for 30 hours so that was really amazing able to keep going and keep going and keep improving itself with the same thing it was working on but focus on web 3. Um, recently there's been kind of a wave of new protocols that use AI agents as a way to help

users invest in DeFi. And there are some that are able to offer like continuously a very a much higher APY on USDC and USDT than maybe protocols that were very very famous like year finance and all the optimizers that we used to have in the past cycle. And now we having the same but more APY. It's consistent and it's totally automatic with a with an agent. It's sure that the agent it's kind of an LLM but also machine learning because it's based on on information and data but at the same time I think it's a very good thing. Um and I think it will be it will be growing this part. And a thing that did didn't work. Maybe now it's different. Actually, we had a uh

when the dinner two days ago, someone told me that now it works much better. But at least on my experience, a few months ago, there was this um also launch of kind of AI interface to make it easier to move around DeFi. So you could ask the agent, hey, swap these tokens to these other tokens and do this bridge and do this other. And it seems that that now it's working or at least someone in the here seal told me now that but on my experience that was awful experience. So the UX UI improvement that I was expecting on AI. So crypto is very nondeterministic. So you need to learn a lot of things to use it. AI it's it's like talking to a human. So this is

like why the adoption it's 800 million users in open AI and you see crypto we still pushing to reach like 500 million. So I was hoping to see this AI easy to use interface in DeFi and crypto and this hasn't worked yet. >> I I probably can follow. So um I think the the moment I saw the title of this panel really makes me thinking that wow people are really kind of not getting satisfactory. Um but on the other side kind of speaking of um our adaptability as well. So remember when GBT3.5 just came out there was so many wow moments and I think AI agent has been working on

so many different areas um but yet we still have the question so does it work or not and and I think this is kind of super interesting. So to me I think there are just so many areas that AI agent or AI or Language models itself and have already been working. So for example like the capability to synthesizing content and organizing content summarizing content on and also the content generation part as well and I think kind of it's something that works and there are many many other different areas as well but and I think for kind of the areas that AI agent that don't work so well maybe one example is that being able to catch the emotional nuances out there kind of be the capability to read a room and kind of be

more fundamental or real um reasoning about the common sense. So these things that more correlated to emotional intelligence. Yeah. So so I think that that is kind of my my top takeaway. >> Yeah. Can me a little bit. So uh I want to share two examples. uh one is a very good example to imp you know which impressed me about the AI agent is the AI agents were um you know uh people already using the law firm the patent law firm so you know that's before it was a very uh challenge for a patent lawyer to check if your innovation or certain new technology is already

covered by some patent before right so the patent is huge and now we just leverage the AI to you know through the document into the AI say hey this is my innovation uh novel stuff I believe uh is this already covered by some patents you know uh you know the AI agent can give you the answer in one or two minutes and you can give the prompt and saying that if my innovation is already covered by certain patterns just give me the detailed uh information about that patterns the reference link and all all the stuff and the AI agent going to put everything in the document and give you a PDF format report something like that.

So you know before this could be the work uh like weeks from the law firm and you have to pay the money right but now it's just like one or two minutes you don't need a law firm to do so you can do it on your own very simple and uh this is how agent really helps a lot uh in terms of of like a patents and the one agent uh example another example I want to share that agent doesn't work is I will say it's all about numbers all about numbers because at the Colombia I have in in New York basically you know I have a research lab we are working on uh build some AI agent with a focus on finance and finance is all about numbers right so we are trying to use all the

current large models like a gro like chat GPT like um to do some analysis on the SEC filing from Nvidia from Tesla from these companies they do quarterly SEC filings with many many numbers personate lots of tables and you see the results from all these AI agents is is is most likely I would say about 90% is wrong. You cannot trust them at all. You have to go to the file again and check all the numbers there and see you know to to do the uh uh you know to do the verification on your own. So uh when you deal with uh numbers using the current AI agent it most likely it doesn't work.

So and we spend lots of time to you know uh to figure out uh how we can deal with it but in general it's really tough because you know in in scient in academic we always say that when AI works we don't know why it works why when it doesn't work we have no idea why it doesn't work. So that's the problem. Okay, maybe um I would ask the audience if you could raise your hand if you have ever used an AI agent. Okay, not too bad. Is everyone using it in everyday life?

Nice. Okay, awesome. Um you just um mentioned the word trust, right? So um when we talk about AI agents trust is one of the big words that always comes up. So but what does trust actually mean when the agent making decision uh decisions isn't human like let's explore how governance transparency and decentralization come into play and who's accountable when things go wrong. Right? So my question would be do agents need to be decentralized and interoperable or do users simply want them to work seamlessly? So how do you actually trust them and who is in the

charge when things go wrong? So >> shall I shall I start with this one? So just connecting from the last question into this. So um I was last week speaking to someone who was very senior um just reporting to the CEO of a of a organization with 150,000 employees and what they've done is they've removed they fired 18% of their staff and they've asked everybody in the organization to start using agents and this is all exploratory stuff and this is a big tech company and what they're doing is they're saying okay start using AI agents to make us internallyffic efficient, but whatever works, let's productionize that and start selling it

like a consultant to other people. So, it's important to note that everybody knows where this is going, but we're still in the discovery phase. Even the largest organizations are throwing spaghetti at the wall and wondering what will stick, right? But when when it does work, then what you need to do is you need to make sure it's trustworthy. You need to make sure it works for large financial use cases, healthcare, defense and everything else, not just us doing things in chat GPT and then if the data isn't correct or not, you know, sometimes, you know, it's our fault if it's uh if we relied upon it, but the organization is not going to blame the AI because we're not at that point where the where the organizations are using it um in a way that they are u telling

everybody that you have to use this within your organization. So one issue which we're having within AI right now is we're building in silos like we built web 3. Um you know you got the agentto agent uh protocol you got agent network protocol and a lot of these systems are being built separately. So what you end up with is trust gaps. So what we need to do is we need to make every um single system firstly accountable. You need guardrails and you need to make make them efficient but then you need to ensure systemtosystem interaction agent to LLM interaction all of these things all the trust gaps are are filled because having and for me be before decentralization trust itself is an issue that needs to

be resolved for organizations. Decentralization is not the highest consideration in my opinion and that's coming from a recovering VC who's been investing in in blockchain for seven years. >> Okay. Thank you. Maybe someone want to add >> sure maybe to add to that also from other industries. You mentioned web [clears throat] three but I think that something similar applies to other industries telco industry in in my case I think the most important thing is transparency like really promoting open source agents uh making sure that we have open protocols so that they can communicate and they are interoperable um and again in the telco industry usually uh operators vendors they work in silos uh and they cannot really uh

talk to each other. One thing that we are doing on our end, we are launching an initiative called open agents for uh telco. And the idea here is to really make sure that you create a framework for open AI agents. Um we have a platform for benchmarking so that you can compare as well. Uh I I really think that this uh open benchmarks and platforms can enhance trust as well in these systems because you really know uh what's happening. Um and then really again promoting interoperability across different systems. Uh so I would say yeah the same thing applies to different sectors.

>> So I would add that there was an uh like an article that many companies that tried to implement AI and AI agents in the end it didn't actually work and I think one of the reasons is that you cannot predict exactly how it will how it will respond. So it usually works very well inside your company where there's a validation. So you work with the AI but you end up validating. But when you give all the responsibility to the AI agent I think it's hard right now to trust it 100% at at least um thinking on DeFi thinking on the web 3 space. So I think a good way to put limitations it's put it on the smart contract level.

So it like the amount of bad decisions that it can make it's limited by the amount of actions that it can take on the smart contract. So I think that is a a good way and on the question it looked like if what do we want the agent to work or to be decentralized and I think maybe if it's decentralized it works better and I go on the pit tensor ecosystem and all the umbrellas of the like the subnets that it has and there's a subnet called ridges which is kind of an AI agents for coding and on some benchmarks it was even um the improvement was higher than claw and some agents from code and that was because the incentives were very

aligned and so they were able to create an agent that was even better not at the model level but on the tools they were using. So I think maybe the best way to to make agents useful on DeFi is making decentralized and um like going against incentives instead of having to trust only a a model like from open AI from cloud or from whatever. So I'd say use smart contracts to limit the the actions and maybe bit tension and decentralized way of making AI ends up working >> right. So um probably one more point add up to that is that um I probably don't

have the answer for u whether my preference will be on centralization or decentralization but there are two consideration that pop up to my mind the very first one is that uh checkability so when you give the autonomy to AI agents and working on behalf on human beings or or or some organizations and you don't want to end up in a situation that there was some bad consequences because of some actions of AI agents and you um it's AI agent it's not me and there's no liability to it. So you should be able to kind of being able to track and have a track records and being able to source back into and what happened in between of the AI agents set up an interaction with human beings and

ideally that should be in a trust foundation and should be immutable and of that check record. So that's kind of the first consideration. The second consideration is that um it's about the ownership. So I think AI agent by nature is something that goes across different platforms and and so I think sooner or later we'll start seeing more and more of AI agents that work embedded into different systems and what works behind that is the memory and the context that tied to that AI agent and that is a treasury for users and that is the most important thing probably and so how do we make sure the right people having access to that kind of contain memory

and context and to kind of the control the permission to that context and being able to kind of um monetize or having the right ownership to claim the profit of that cont context or memory and and I think those are the two things that pop up to my mind immediately and whether or not centralization or decentralization is a better solution to think about these two dimensions. Um I think they are pros and cons. So in fact trust has already become the bottleneck for massive adoption of AI agents. So I visit Silicon Valley many

times per year. You know I I I interviewed lots of stops in Silicon Valley which is in the United States which is already the hub of AI agents right. So when they sell their AI agent to the customers, the big hurdle is about the trust. For example, let me ask you this question because chat GPT for now is very powerful. If you allow the chat GPT to have a full access to your phone and to your laptop, the CHP could be your very powerful personal assistant. But would you want to give the full access to charge?

who gonna say yes? No, of course not. Right. I just use charge to do some editing work or drafting something like that. But I would never ever give my phone access to charge. Why? It's about the trust because I don't trust I have no idea what's charging going to do with the information from my phone because tragibility is a black box for us. We have no idea. uh we know nothing about it right um so the trust is a very big topic and there's a multiple ways to deal with the trust issue if in order to have a massive adoption of AI agents

so I would say decentralization is one way one approach by leveraging blockchain technology blah blah blah but this is a long longterm effort it cannot be done in one year or two I would say because blockchain technology itself is not mature uh at this moment because I I've been teaching blockchain technology and the AI technology at Columbia University for almost 10 years. I know from a tech perspective I can tell is way far away from mature uh blockchain technology itself and I think for now the only way we can do is we still build trust on the centralized entities like a Microsoft,

open AAI, Google we we have to trust them because we have no other options right so we can re rely on the government to put more regulation ations compliance on these universities. let them to be a little bit more transparent and let us know as a user um some information about the AI and how things works behind the scene and then we may have some trust and later on maybe next year you you see some AI agent you want to use on your databases and the selling point is like okay we're built on like Microsoft Azura cloud you can trust or we build on Google cloud you can trust

so then that's where the trust comes from for now but in the future I would say after 5 years 6 years 8 years 10 years maybe the decentralization technology became mature and then people will start to build some AI we always say the AI agent should be built by community used by community but that's a very good vision cannot be realized in one or two years uh for now I would say we still have to trust the big entities >> thank you I'm so happy that you are on my panel because you're building bridges to always the next questions.

So um we've touched on trust and governance but trust alone doesn't make the system work right. So agents also depend on access to data infrastructure and connectivity. Um so let's talk about what it actually takes for these systems to be reliable and globally usable. Um because not everyone has the same resources or digital foundations. Um so what needs to come first? Is it connectivity, governance or trust? >> So what do you think?

>> Shall I start? >> So um I guess I'll keep it quite simple for me. Um without connectivity you can't do anything. What's an agent without data? What's an agent without compute? So connectivity without connectivity you don't have AI. So that's first. It's like with that it's like without connectivity it's like trying to walk with no legs. Then comes governance. Governance is basically one like level of safety. You've got your connectivity sorted. You've got some checks and balances in it. But still it's like walking on your knees at that point. Now trust is basically what you aspire to. It's it's it's what we all should be building towards in the long

run because that that's when you start running. It's when you start getting enterprise use cases of AI. It's it's why the startups in Silicon Valley can't get uh you know the big companies to adopt them. And you know, we all believe and and the valuations in the market demonstrate the the expected um you could say uh disruption of AI. But we're still on that early stage of that curve where um the trust element needs to be resolved before you know large organizations which you would classify as 500 million dollars of revenue plus per year would adopt this type of technology and take that kind of risk which could end up having some kind of detrimental impact if not done

correctly. >> I was just going to ask you >> quickly to add I kind of agree. However, I do think that you need the right governance structures in place from the beginning. Otherwise, the damage or harm can be huge until you you get to that. Um maybe like also speaking from the perspective or lens of children and vulnerable children. Um the the types of population that we work with usually there is not even data about them. Um either there is no connectivity or data about these vulnerable children that we are trying to serve. Uh it's inexistent.

And as you were saying, if there is no connectivity, it's just not possible to to deliver digital services to them um including AI. And we can definitely not leverage the potential and benefit uh from the from [snorts] AI to uh deliver or provide even better services to to these children. However, if we don't have the right protection, laws, privacy, governance in place, at the moment connectivity reaches, exploitation can arrive uh and you can do a lot of damage. I would say that you need to make sure that you you lay out the right foundation from the beginning.

>> So what do you think is the most needed at the moment? What should be taken of care of first >> among the these things? All of them. I think it's the chicken and egg issue, right? Like you need um the right governance h but you also need connectivity. You have the governance but you don't if you don't have connectivity then you you have nothing and vice versa. So I think it's the chicken and egg um dilemma. >> Anyone want to add something here? >> Yeah, maybe. I think it's pretty complete, but connectivity is crucial because if not, you have a brain that cannot do anything. So he must have the ability to interact with other apps and other tools. But I would add that maybe

what's necessary and I've felt this on the crypto space. It's more entrepreneurs like more people trying things. I think agents are able of so many things that it depends on our c creativity and there was kind of a boom on AI agents uh in crypto and I love that kind of rise I thought it was amazing I thought it was very cool there was an like many agents and then went to zero but even like completely autonomous they were able to create their own IP and collaborate with other agents and there was an agent paying $1 to another agent because it needed a service and it very like initial and like first things and it's normal that it didn't work but

I would say like more people doing things and like connectivity comes up with uh people creating these MCPS and and connectors between apps so yeah more entrepreneurs building technology >> thank you so I think we have now crucible time so where will we really B. So we've looked at what's working today, what's broken and what's need to be built, right? So now let's look ahead say 3 years from now.

Will AI agents be quietly running in the background of our everyday lives um like invisible assistants or will we still be struggling to get them out of beta mode? What do you think? >> And I probably can take a step. I actually don't feel like it's still in beta mode. I think it's kind of changing many industries already and it's making tons of revenues already. So, and in 3 years I I don't know I I think there will be different formats but to me um I feel like lots of platforms might be

changing. So and the way I'm thinking of it is that the way we are thinking about the current platforms they are like a flat services. So um it's a an an flat front end. So for example when you are dealing with Instagram when you are part of it and so when you are interacting with it you are part of it but when you are living it has nothing to do with you but what AI agents really enable us is that it really enriches the consumer interface. So it's something that like having tech reaching out to different corners. So in your conversations in your different kind of integrations of other platforms out there. So um to me I think this is going to change lots of things and change on where intent are being captured and serviced nowadays. So

and and nowadays we are seeing lots of companies and lots of startups they are making many monies already and by automating um the blog posting automating kind of the and cold touch emails and which to some extent I think it's the initification of the existing internet and how long does it going to last but humans still seeking for the humanto human values so where will that human to human values to exist in the future so to me I think in three years we going to see the next waves of giant companies like uh the existing Tik Toks like YouTube they're coming up but they are no longer going to take the same format as where we seen them nowadays.

Um and the other thing that really fascinating about large language models enable is that think about how much of the information that being captured in such a small model that could be distributed in a quite decentralized way and that will make things like content creation can target for individuals and recommendation. you don't need a large data cluster to do the recommendations and actually is something that could happen for yourself for your groups of people and would that really change how how we interact with the internet in the future and I think this is something that I'm really looking forward in the next three years >> um Max maybe one question for you so what kind of backbone infrastructure is

still missing before agents become mainstream what do you think >> uh you mean the backbone in in infrastructure in terms of a centralized or decentralized >> what would you choose? Um okay so I think um I I I want to talk about decentralized infra uh because you know this is a okay access stage let's talk about some uh blockchain stuff so uh you know we already see lots of decentralized uh uh you know we call this in general we call that deep pin right decentralized physical infrastructure um so far is still like what I said from

infrastructure technology perspective is so far way from mature. Uh when we talk about the performance of the or backbone of this infrastructure, we're talking about SLA. If you understand that, you know that's the performance metric for uh the cloud data center stuff, right? The latency, availability, reliability. Uh these are the three key uh metrics for the cl uh data center performance. But if you do the cloud in a decentralized way, the latency is a big issue, right? Availability is okay. Uh but reliability is another issue. So how we can deal with reliability and uh

latency uh with comparable performance with AWS or Google Cloud. From a tech perspective, this is very difficult. For example, like for for you know uh in terms of a latency uh AWS offer you 100 millisecond uh type of a latency but in a decentralized way if you use a decentralized GPU for inference like your AI agent inference um is the latency going to kill the app then you have to use deploy the app on AWS or Google cloud again you know uh so that's the thing uh but uh over the years you know I know in academia there's many

research group they are working on this and to train the AI agent in a decentralized way to lower the cost and uh but for training is okay because the training you don't need to worry about latency uh but for inference for inference means uh the latency is the key right and the reliability is the key as well so decentralization is is a pretty challenge and that you know how how can we build a back you know the backbone of a decentralized infra to meet this latency availability reliability metric comparable with AWS or Google cloud is still um you know uh I would say uphill battle

>> I was going to add that also the the power and energy required to to power all the GPUs data centers that you were just talking about um we all know I think that demand for energy it's going to grow exponentially. It's already doing so. And I think it's a similar debate to the blockchain web3 one. Um yeah, you now have they say that data center companies they'll have to investor or um build the same amount of infrastructure as they did in the past 20 years over the next next couple of years. Um you have some of these companies building their own solar plants, their own energy uh plants etc.

Even nuclear is coming back to to the conversation and debate uh because I think uh the the issue of energy it's still an unsolved um issue and that people are yeah trying to look into because we definitely know that's going to grow exponentially and that we need to to find a solution uh for it. >> Yeah. Going the like in the following three years which I think is very interesting. So I absolutely think that AI will be everywhere. I don't know in which level and which speed but it will be everywhere and it will have AI companions and they will live in all devices but speaking in three years I think we'll see decentralized AI versions competing with centralized and I hope it's like this and I really I'm

really fan of the bit sensor kind of approach. I know there are many projects trying the same but I think the way they've built incentives and the way they are solving like for example inference or even decentralized training and it's working so I think it's an important thing that um it's an import difficult project to understand but like for example the energy issue like the biggest supercomputer is bitcoin and they built it only using incentives so I think incentives are the key and if you align them well with um precise problems, I think you can build amazing stuff. So, I would be hoping in 3 years to have decentralized AI products that are more competitive, that are better in

the benchmarks and are cheaper than centralized options. >> Can I just answer the same? So in my opinion um it's actually scary and also super exciting how AI is going to be everywhere and you know we talk about how it's going to be in the background as well but I think with the advancements in chips in in like um models as well and also with energy maybe we're talking a bit longer than just 3 years and also if you think about this emerging theme around robotics AI is going to be physical in front of our faces this right and so so when it comes to whether it's the physical side or the digital trust needs to be woven into the

very fabric of AI right we've already got like dark factories in China where basically robots are working 24/7 in the dark so you know this is a theme that that's only going in one direction but we've also seen scenarios where agents do not recognize who their master is and as we go towards like sentient AI I you're going into into a world where trust is going to be like foundational to this technology. Right now, innovation outpaces, regulation outpaces trust and governance, but eventually as we move towards u mass adoption, I know there's decent revenue for startups, but I'd still call it decent. It's not um

it's not meaningful at the global stage right now. apart from you know things like chat GPT we're not really seeing it in in everything we do but that's that's where we're headed so it's exciting but also scary is a it's a fine line between you know uh innovation and and and and regulation >> that totally reminds me to that movie Iroot right oh man okay so my voice is almost gone completely sorry for that time flies and I want to end on a note actually that gives everyone a takeaway. So I ask um each of you to complete the sentence in one line. It's AI agents will only succeed

if you go first as no pressure. [laughter] I I think that for me that it's almost like the sentence is uh is structured wrong. I think AI succeed I think the full stop comes after AI agents succeed that and it's just a it's a case of when not if >> nice. Okay. What about you Max? >> Uh if we fully understand how AI works.

>> Oh my goodness. The same answer. So I was just about to say explanability. So and being able to explain >> I would go with the same answer that it's already succeeding. So just keep doing the same. >> All right. So to to change a little bit and again coming to my topic if humans are put in the center and are the ones that kind of yeah try it. >> All right. So let's see what our collective wisdom says. Right. Um beautiful. I think that's the perfect note to wrap up the conference on. So, um, maybe I have one last question.

What's the funniest or weirdest AI agent use case you have seen so far? >> The funniest. >> Funniest or weirdest or both? >> Weirdest >> drawing sketches of fish. I saw a website just solely doing so. >> Anyone else? Nothing funny in your everyday life? >> [laughter] >> like it comes to my mind um an AI agent that became an artist. So he was publishing autonomously his own albums in Spotify and they were pretty good. They were not bad. So that that was that was fun.

>> Interesting. All right. So thank you all for staying with us until the very end. like my voice to our amazing panelists for the insights and to all of you for being part of this journey. Enjoy what's left of the day and may your next AI agent actually do what it's supposed to do. So, thank you all and let's give a hand to our human agents here on stage, please. >> Thank you. [applause] Thank you.