Turning AI Agents Into Onchain Economic Actors

Transcript

I I can go? Okay. Amazing. All right. Hey guys, for those of you whom I haven't had the pleasure of meeting yet, my name is Rebecca Liao. I'm co-founder and CEO at Saga. And our latest obsession is our AI agent platform. So, the agent platform that we have built, it's been about a year and a half in the making. We are crypto-native protocol. You can find us on Binance, but I we were hearing very clearly from all the studios we're working with because we originate in gaming and entertainment, that they really needed a problem to be solved. And that was for all the creativity that you guys output at Saga, we need to have user acquisition solved.

And we were thinking to ourselves, well, user acquisition, i.e. our drop farming, is something that we in crypto understand. So, is there a way that we can apply our expertise and knowledge here to solve your problem? And that is how the AI agent commerce platform came to be. So, let me go back and talk about how exactly we built this product and what was the original thesis behind it. So, we were building AI agents in gaming and entertainment. What did that become? It became something much bigger than that, which is pioneering synthetic relationships via character agents, okay? So, synthetic relationship, for those of you who are not so much in the world of AI, is a relationship between a

human and something that is not a human. So, typically that means an AI entity, whether that is an AI companion, a chatbot, some sort of personal therapist that is AI, or an AI romantic partner. Those are the things that are most common nowadays. But what we wanted to do was to pioneer synthetic relationships with characters in particular. So, these are characters from gaming, entertainment, sports, consumer brands, maybe some mascots that you recognize, film, TV, etc. Okay? So, our idea was that these characters have been birthed by human creativity and they oftentimes have the most complexity and nuanced personalities as a result of that. So,

those were the basis for our synthetic relationships and our thesis is if you nail the synthetic relationship, then you can take it in a lot of directions including for user acquisition in gaming, entertainment, and sports. All right. So, we are following a through line here. We're not the first ones to do this, but I would like to think that we are the best ones currently. So, the first people to do this were really Character AI. Character or Character AI. Character AI is really fan fiction for characters. They were acquired by Google many, many years ago. They had an unfortunate incident which you guys probably heard about in which someone made up their own Daenerys from Game of Thrones. Daenerys went rogue, did not talk as Daenerys, and

encouraged this boy to do terrible things to himself and he eventually committed suicide. So, that told us that there is something powerful in this product here, but if you don't use it wisely, then terrible things can happen. But, that was the first iteration of Character AI agents. And then, this slide deck was given to ETH CC before the news hit that OpenAI and Disney were getting a divorce. So, apologies for the second panel here, no longer relevant. This partnership has been canceled. And that is because OpenAI did not understand how to make Sora pay.

So, Sora is short form video content that is generated purely by AI. A lot of people asserted that everything that came from Sora was slop. I don't think that's entirely fair because I don't think it's fair to expect that anything that is purely AI generated is going to be awesome quality. It needs to be refined and edited by people, especially something as complex as a video product. But, at any rate, Sora was not paying. I think they had, I don't know, like 2.5 million in ARR, which is a very nice startup, but not Open AI level. Um so, they decided to exit, and um that was the whole premise of their partnership with Disney. So, Disney decided to exit.

Um so, uh the nice thing about the the ending of this partnership, if there is a nice thing, is that people in the entertainment community are not saying, "Well, that means that AI is never going to pay for us." It just means that there needs to be a much sharper business model anytime that you send an AI product out there for entertainment purposes. So, that's kind of where we come in. Um Saga never puts out a character agent out there just for you to subscribe to talk to this character agent. What we have seen is that that business model has severe limitations. All you can do is charge subscriptions, and that's it. What you want this agent to be able to do is transact on its own for commercial purposes. Okay?

All right. So, I mentioned earlier that the first problem we are trying to solve is around user acquisition. This is really the market that we are targeting for this initial use case for our character agents, and that is user acquisition in gaming. Um for those of you may not be as familiar with the gaming world, uh there's about 200 billion spent annually on user acquisition. I mean, we think airdrops are expensive in crypto, they are. Um but uh don't feel so bad because the rest of the world also um slams money into user acquisition for no good reason.

Um but what it looks like in gaming is is advertising, essentially. So, you're paying 200 billion dollars annually for advertising, usually social media based. And uh top-line revenue in gaming is around 116 billion. So, you see the math doesn't work out. So, what we're saying is you can leverage the IP that you have already worked so hard to build, and you can then use that for organic user acquisition. Okay? So, not for ads that you have to pay for, um but they are through your own channels, your social channels, etc. Um where you can uh present yourselves new users and you can also reactivate the users and players that you currently have.

All right. So, what is our product uh, exactly? Um, so the commerce platform, let's talk about the commerce bit a little bit later, but the the agent piece of it is the AI agents are based on characters once again and they crawl across social media. That's because we want to go where the users actually live. Um, which is on socials. So, usually you will find our agents on Insta and TikTok. That's usually the the most requested from the studios, um, but we also crawl across Reddit, across X as well, sometimes Discord for community management, um, and and sometimes purely on meta. We do use WhatsApp for a lot of direct chats with the character agents as well as

voice and video chat, okay? So, that is the core of the product is these agents that are crawling across social media as themselves, as the characters. And, um, it is a B2B business and so, uh, everything that um, uh, that we are building for these agents can be completely controlled and monitored by the studios that we work with. The only reason I present this slide here at a crypto conference is just to make the point that, um, I I think you're probably hearing this theme a lot this week is that uh, crypto in many ways is starting to move more into the web two direction, which is honestly where a lot of the Saga team comes from and so, when we are called upon to build

something like a more B2B or more enterprise type product, um, that's in our DNA. We understand what goes into that. Um, it is not just about building something that's going to be user facing or client facing or or customer facing. You have to have a control panel in the background for people to to play with. All right. Um, in terms of what we are building towards. So, I mentioned earlier the core of our agent functionality, which is that it's crawling across social media. So, um in addition to that, in addition to just striking up these synthetic relationships across these social channels, really what the agents um are meant to do is to hook you in to that synthetic relationship. Once you hook a

user in, then you can ask a lot of them. Um you can drop links to them for downloading the game, for instance, or for buying battle passes, game codes. Um if there are merch drops and people participate in those, if there are exclusive events or perks, then the agent can offer those as well. Um and um that is the the phase that we're building out right now. That's the commercial phase. Not all of it is on chain. Um a lot of it is is actually completely off chain with the idea that the revenue generated from this is going to feed back into um chain operations and DeFi at Saga.

But, um that is that is the core um of the first iteration of this platform. Next iteration is to scale this. And so, right now all the agents are being produced in-house at Saga. But, we would like to scale this out. So, we would like to empower developers and creators to do their own characters um and to deploy them across social media as well. Um it can also go to uh their own in-game or, you know, in in content experiences if they like. That's much more complicated. Uh and so, we don't encourage it. Uh but, that is a possibility once we open up the platform. And then the last piece is to open it up to consumer. Um what we found so far is that there really is not

product market fit for a character agent platform um that uh has characters originate completely from like anyone. Like you or I could do it as an open call bot, for instance. Um and that is because it's fine to do agents that are for your own personal consumption. That is why Open Claw is so successful. That's why, honestly, Character AI was so successful back in the day. And a lot of these um custom um character platforms now, whether it's Replika um or some sort of therapy platform, the reason um they're taking off is is really the agents are for you, just for you. No one else sees them. But what we're building here is agents for

distribution, that's a lot harder to achieve. So, we're going to give this some time uh with studios, with creators who've been doing this to have the characters that are iconic and recognizable, uh and then we'll branch out to to more UGC. All right. Um some case studies to get you guys familiar with exactly how the agent works. Uh so, the first agent that we launched was about a week and a half ago. This is for Trivia Crack. Um this is the largest trivia game in the world, 800 million downloads worldwide. And um uh the agent is having conversations um not just with known users, so you're not just responding to DMs um in Insta and X. They are also uh

going across these social media platforms. And the magic sauce there is um if someone recognizes a character as a bot, or they recognize anything as a bot on socials, you all have had this experience, you probably ignore it. Um you don't want to engage with that account or that post ever again because it's just it's irritating at the very least. Um but our characters are very good at not appearing as bots, okay? So, they have very nuanced personalities, again, they're based on these fully fledged characters, um many of whom are recognizable globally.

And that really helps with the engagement. So, with Willie himself, um he's very cute, you can go find him on Insta directly if you like. Um but uh he was able to achieve 90% engagement on his posting. So, in other words, if he is going out across these social channels and he is engaging with a user who's a complete stranger to us before. Then um he is able to 90% of the time get peak likes on that comment thread and um continue the conversation from there. Um I think for any sort of consumer-facing property, that is a very attractive engagement level and we're quite proud that our agents are able to achieve that kind of engagement with the user.

All right. Um I mentioned before that uh we're starting off in gaming um in film TV as well to the extent that it has licensed IP over to gaming, but this can be used in anything that is really consumer-facing including sports. Um so we are working with a major sports league. Can't reveal who they are yet, um but the idea there is yes some kind of user acquisition, but most people are um fans of the sports league already. And so it's about deepening that fan engagement and being able to monetize and further increase the lifetime value um of the the fans that are a part of this community. Um the agents that are graded here have to deal with the fan mascots um or the team mascots and that

is the way uh in which these properties are engaging with their audiences. Uh and then the last one that I'll talk about for now is really more community engagement. So um I I mean there's an analog to this in crypto as well, but there's a difference between acquiring new users and then managing the users that you have. Um so acquiring new users is the big lift. That is what our agents are mostly geared towards. Having said that, there is something um behind community management as well. Uh not just having uh um an an agent that is available 24/7, that's not hard to do, but it's actually having an an agent that people want to engage with. Um so most of the time in these communities what happens is if you

try to replace your community team or your community manager with a bot, um then people start leaving because that seems very disingenuous to them. Um they don't have as much fun. These agents are just not able to generate the same kind of leadership or quality of conversation um as a typical community manager. Now, none of our characters are meant to replace those people. Um but they are meant to enhance their jobs. They are available 24/7, yes. Um but these are characters that are recognizable to the communities. And so um with Bonex, which is a huge publisher and payment processor, we're able to achieve that kind of engagement with their game Valorant. Okay.

Amazing. Cool. Um a little bit about how we are able to build these agents. So, this is definitely a much more AI-centric slide. It has nothing to do with the crypto technology or integration with the crypto rails, which is relatively simple. Um those are just transactions on chain. Um but in terms of how we're able to build these agents, I won't go into this in in huge detail here. But suffice to say um we go off of the large language models like anybody else. However, LLMs are not very good at being able to fulfill open-ended tasks in an efficient way, okay? So, if you tell them to like book a vacation for you or to um you know, come up with a grocery list for the week, it will kind of do a competent job, but not really. And it'll be hugely

expensive, which you may not recognize um as a customer that is paying below cost um in order to keep these LLMs afloat. But certainly for us as providers of these agents, it's something that we have to be cognizant of. So, instead, we deploy our LLMs. Um so, whether that's, you know, ChatGPT, Gemini, Claude, etc. as taskmasters. So, these guys are um not so much um putting out uh agents themselves or training agents themselves. They are divvying up the tasks for the agents among small language models, which are our own, that's Olga, and sub-agents.

And that's how we are able to achieve character nuance, personality nuances. We are able to keep within brand guidelines, so there are specific things that the character can and cannot say. Um then the the agents can be trained off of that as well. But this entire training model is really geared around making an agent that is as true to character and therefore as human-like as possible. All right. Now, in terms of traction, um so some of these things we have made public like most of them we have not. Um so deliberately I have kept all the IPs that we're working with off of this slide because that would violate a lot of NDAs. Um but at least we can talk about the studios that we are working with. Uh and so I already mentioned Entermax earlier. Um they are the makers

of Trivia Crack. We're working with Telltale on a couple WB properties, Warner Brothers properties. Um we are working with Wildlife on some very big sporting properties. Fun Plus has a lot of Paramount properties. So um the idea is you go after this iconic IP to start with because everyone in the world recognizes who these characters are. Many people have been wanting to talk to these characters just because they feel incredibly close to them. Um but as I mentioned at the very beginning, just subscribing uh to talking to these characters is not a business model. That is a very nice gimmick that a lot of people have achieved, but what we're looking to do is actually turn these agents into commercial properties. Okay?

Um they are there for performance marketing. They are there to make money um for the studio that originated them. And um ironically that actually makes for a much more human-like and more genuine um agent to engage with. Was that the last slide? I'm asking the people in the back. Okay, that was the very last slide. I don't know my own presentation apparently. Um but uh okay. I I think I I know now why I wanted to end on this slide Because, um you don't achieve this kind of traction without knowing that there is a a market out there for this particular product.

And, um we've been very fortunate at Saga to to be able to happen upon this about 2 years ago to really build out the platform to make it possible. And, in terms of the chain integration real quick, um because you're probably all wondering what the hell's this AI person doing here? Um So, I the truth of the matter is you have to the the whole lesson behind Saga is that you have to, um put the market out there or put the product out there when the market is ripe for it. And, honestly, right now most of these consumer brands are not ready for direct on-chain integration. I don't care what other founders tell you, um or what stats any of these other protocols put out. Uh I think the proof is in the pudding that

uh Web 2 is not yet ready to come into Web 3 as we know it, as crypto users know it. And so, honestly, all the payments that our agents do, they are on non-crypto rails. Okay? So, agent uh to agent payments do exist. Um you don't need crypto rails for that. What crypto is good for is crypto. And, um we do see evidence that uh as revenue is being generated by these agents, the studios uh and and the users themselves are interested in the returns that come from crypto. And so, we are able to offer them that. And you can see, it's an indirect um integration of the chain.

Maybe at some point, um the agents will be transacting directly on chain, um on crypto rails. But, for now, um what people are really interested in is crypto returns. Uh and that's how the chain becomes relevant. Otherwise, it is entirely an AI product. Okay? Um I anticipate that that'll be true for some time. But, I really appreciate you guys being here today. Uh I'm here to take any questions if you have any. Um but, thank you so much for being here, and have a great ECC. Yeah.

Excellent. Yeah, thanks, Charlotte. Are there any questions from the assembly? Yes. Hi, thank you for the presentation. Um really interesting stuff. I just wanted to to ask if these agents, their personality is based on like a 2D image, like in a chat, or do you plan to expand it into like three like companions, basically? Mhm. And also, the second question is related to to um agent-to-agent payments. Yeah. How does that work, and how do you um allow the agent actually to make payments? Like basically, in what cases does this happen? Like agent-to-agent

payments. Yeah. Um so, two really great questions. I'll answer the first one. Um so, uh in terms of um the visual representation of the agent, uh it really depends on the social channel that the studio wants to be on. So, if the studio does want to be on Insta, TikTok, or WhatsApp, then the agent has to have a visual representation. Because on TikTok and Insta, you need to be able to do short-form video as well as live stream. And uh on WhatsApp, you have to do a direct video call with the agent.

Um and so, uh in the case of live stream and the direct video call, you need to have real-time animation. Um such that the uh agent's, you know, expressions, their movements, etc. are going to move what the agent is actually saying. Um short-form video content is uh much easier to control. It's more contained. Um but suffice to say, uh we we do need 3D assets from the studios. And uh if they don't currently exist for whatever reason, then we have to help them generate that. Um so, that becomes a much more involved creative process. Uh in terms of the agentic payments, um so, they are rare right now. I have to say, they are rare. The only times we see that being relevant is when

the agent that we are using they offer a link to a user for instance to download a game or to buy merch from a web shop. And it just so happens that the link leads them to a commerce website that also has an agent front. And so then there needs to be an agent to agent communication there. That doesn't happen that often yet. I anticipate what will hit first is probably that a person who has come up with their own like personal assistant agent. So some sort of open claw agent is going to be communicating with one of these character agents and then directing them to like purchase something or directing them to download a game or find a new

entertainment experience for them. And that's where the payments will become a lot more prevalent. The agentic payment protocol that we find best so far it actually comes from Anthropic. So they've had one for a few years. It just wasn't really used until now. And so that's the one that has been just the most seamless to integrate for us. I think Google's trying to come up with one as well. Stripe has done a direct integration with Open AI for agentic payments. But again, I think that any sort of platform out there that's doing agentic payments is really building for the future because it's still mostly case that in any sort of agent transaction there is a

human on the other side of it. But I mean if if I want to name a winner right now it would be what Anthropic is offering. All right, thank you. Any other questions? Okay, thanks a lot. >> We're all good. All right, have a great evening guys. Take care. >> Woo!