Harrison Hines--Fleek and The Future of AI Agents and Crypto

Summary

In this conversation, Harrison Hines, co-founder of Fleek, shares his journey into the crypto space, the evolution of Fleek, and the company's focus on AI agents and trusted execution environments. He discusses the implications of recent developments in AI, particularly the emergence of DeepSeek and its impact on the market. Hines emphasizes the importance of simplifying technology for users and developers, and outlines Fleek's core offerings and target audience in the rapidly evolving landscape of web infrastructure and AI. In this conversation, Harrison discusses the rapid growth of prompt-based web app building products, particularly focusing on Fleek's strategy to target non-technical users with their agent products. He emphasizes the importance of market research and customer development in shaping their offerings. The conversation also highlights the need for improved user experience, especially for non-developers, and explores the potential of token models in engaging the community and aligning incentives. Harrison shares insights on the infrastructure needed to support innovation and lower barriers for users, ultimately aiming to create a seamless experience for building and deploying agents.

Takeaways

— Harrison Hines transitioned from equity crowdfunding to crypto after discovering Ethereum.

— Fleek aims to simplify the deployment of AI agents for non-technical users.

— The recent developments in AI, particularly with DeepSeek, are seen as bullish for the industry.

— Fleek’s infrastructure leverages trusted execution environments for AI agent hosting.

— The company is focused on lowering the barrier for non-technical users to engage with AI agents.

— Fleek’s core offerings include website and app hosting, as well as AI agent hosting.

— The importance of privacy and verifiability in AI agents is emphasized.

— Fleek’s approach is to provide a user-friendly experience without overwhelming users with technical details.

— The company is positioned to capitalize on the growing demand for AI agents in various applications.

— Fleek’s infrastructure decisions are driven by the needs of its products and services.

— Bolt.new achieved 20 million in ARR in two months.

— Targeting non-technical users is a key strategy.

— Market research is crucial for understanding user needs.

— Lowering costs can drive innovation in agent deployment.

— User experience improvements are on the horizon.

— Agents could become the new apps on phones.

— Token models can enhance community engagement.

— Infrastructure is essential for enabling new platforms.

— Listening to customer feedback drives product development.

— The future of agents is promising across various industries.

Chapters

(00:00) Harrison Hines: From Startup to Crypto Pioneer

(03:01) Fleek: Revolutionizing AI Agent Deployment

(05:57) DeepSeek and OpenAI: The Cost of Innovation

(09:03) Fleek’s Positioning in the AI Agent Landscape

(11:47) Trusted Execution Environments: The Future of AI Agents

(15:11) Simplifying Privacy for Users and Developers

(17:57) Fleek’s Core Offerings and Target Audience

(34:50) The Rise of Prompt-Based Web App Building

(38:53) Understanding the Agent Space and Market Research

(43:04) Infrastructure for Innovation: Lowering Barriers

(46:57) Enhancing User Experience for Non-Technical Users

(52:41) Token Models and Community Engagement

Follow me @shmula on X for upcoming episodes and to get in touch with me.

Chapters

• 0:00 Harrison Hines: From Startup to Crypto Pioneer
• 3:01 Fleek: Revolutionizing AI Agent Deployment
• 5:57 DeepSeek and OpenAI: The Cost of Innovation
• 9:03 Fleek’s Positioning in the AI Agent Landscape
• 11:47 Trusted Execution Environments: The Future of AI Agents
• 15:11 Simplifying Privacy for Users and Developers
• 17:57 Fleek’s Core Offerings and Target Audience
• 34:50 The Rise of Prompt-Based Web App Building
• 38:53 Understanding the Agent Space and Market Research
• 43:04 Infrastructure for Innovation: Lowering Barriers
• 46:57 Enhancing User Experience for Non-Technical Users
• 52:41 Token Models and Community Engagement

Transcript

0:01

Welcome Harrison Hines from Fleek, how are you? I'm good, how are you? Thanks for having me. absolutely, thanks for joining. I wanna start with your origin story before we get into Fleek and all the cool things you guys are doing. How'd you get into crypto? Yeah, so basically, back in 2015, I was working at a startup called Seed Invest that was working on equity crowdfunding. It was around the Jobs Act and regulation that had passed back in like 2012, trying to create crowdfunding rules. But the regulations were kind of botched. So that space didn't take off the way most thought it would back then. And then I saw Ethereum launch and I was like, wow, this is actually what like crowdfunding was supposed to be. And so I had reached out to Joe Lubin at, this was like late 2016, token sales were very top of mind for Ethereum back then as one of the early use cases that was taking off. And I had pitched him an idea. that I thought I could build a business like Seed &Dust on top of Ethereum and bring in some best practices around token sales and try to turn that into an industry. And so that's where I joined Consensus at the start of 2017 and built a project called Token Foundry. We were the token arm of Consensus working on token sales and token sale infrastructure for some of the internal projects and some of the ecosystem projects at that time. Did that for like two and a half years or so. And then when the ICO bubble started to die down, I left with a few people and started Fleek. That's crazy. You're an OG. That's really cool. to when you joined and, yeah. Yeah, so you're co-founder at Fleek. Tell us about Fleek in, let's put it this way, if you were to describe Fleek to your mom, how would you do it? I'd say it's tricky because what Flick has been known for is not necessarily what it's going to be known for. So I would say if I were to explain it to her today, I would say it's the easiest place to deploy AI agents as a non-technical person. Okay, that works for me. Okay, go ahead. be true at this exact moment. We do have our AI agent hosting product live at launch two weeks ago. You can deploy agents as a non-technical person. That part is all true, but I would say the current version of the product leaves a lot to be desired. And so in the coming weeks, it will eventually show itself as being a much, much better product, but. Yeah, that is where we're going. Well, this is timely because of lots of controversy happening today with DeepSeq and OpenAI and a lot of value. The NASDAQ market has gone down a lot because of DeepSeq announced that they're able to produce, I think, the same inference results as OpenAI at like one-tenth of the cost. And so I think there's been a lot of controversy there. What are your views on that? I think it's extremely bullish. Yeah, the cost, it's like, I think something like 96 % cheaper. What I think is confusing a lot of people is like, they didn't go about building this model in the traditional way most existing LLMs have. What they did is essentially create a new way of training an LLM. which is a lot more reasoning based. And so what's interesting is like, it'll be interesting to see, especially given it's open source, if the bigger players now just take those new approaches to building models and apply the massive data advantage they have and what that might produce in terms of results. Now, cost of running it is kind of separate, which is definitely an enormous advantage and it's going to unlock a lot of, I'd say, innovation in terms of use cases that now just became possible, that weren't possible when it was much more cost prohibitive. to use LLMs. And so I think the innovation we're gonna see, I think the fact that open source has now proven it can compete on a level where money is no longer the gating factor. I think these are all very bullish developments for the space. But yeah, I'd say the reaction today from the markets is definitely a little bit. surprising, although nothing surprises in markets. So I'd say maybe not totally unexpected. Maybe the market was just overheated and needed a reason to dump. But I think long-term, it's an extremely exciting development and it's gonna be extremely bullish for the agent space. And another thing I would say that like I was worried about a little bit was like, you know, We're obviously betting big on AI agents and are very bullish on the agentic future and the role agents are going to play in that future. But for a second there, it did concern me a little bit that LLMs were in a position to essentially directly offer a lot of the functionality that agents kind of offer. And so what I liked most about this DeepSeq curve ball is that it put these LLM wrappers, which some people call agents, in a much more favorable position now that the playing field has been reset, where you have an open source model that anybody could access and use. And so I thought that was a really important development because you just saw a chat GPT launch operator. They've made it clear they have other agents coming. so I think one, I wouldn't say like it was a existential like crisis, but I do think in the back of everyone in the agent space and agent framework lands mines, it was okay, how much of this can. Open AI integrate into their LLM directly. And what does that mean for agents in the future? And I think that with DeepSeek, it really did like level the playing field where actually the agents or the rappers, these LLM rappers actually are the ones with the moat, because they're closer to the end consumer than the models themselves, which are becoming more more commoditized. And so that was actually the biggest takeaway I had from this whole saga that I thought is really bullish and bodes really well for AI agents and this sort of like agentic future that's emerging. Yeah, I agree with you. think the overreaction I think points to two things. What the real moat is, and that's some of these proprietary companies with very proprietary data. That's one clear moat because that data is not available to anyone else. But then also on the getting closer to the user, that's also a clear moat and being able to provide useful and usable services at that level is going to be really key. tell us how Fleek, given the space that we're in now with DeepSeq and the controversies happening, how are you positioning Fleek to play in that space? Yeah, so I would say our initial approach sort of came out of a necessity as we were building Flick network. So like when you think about Flick, I think a good way to think about it is kind of similar to hyper liquid. It might not make sense when I first say it, but what I mean by that is they're a very product driven L1. So like when you think of hyper liquid, you first think about the the perps exchange and you think about them being an L1 sort of secondary and they sort of really use the perps exchange as a unique distribution advantage to drive usage to their L1. It is a general purpose L1, but it started as very application specific for their perps exchange. And so with fleek, Flick network is kind of like an application specific blockchain for the Flick platform. Very focused on high performance, low latency web services and web infrastructure that we need to kind of provide the products and services we already offer and want to offer in the future on the Flick platform. And... In the design decisions we made in order to make it as performant and low latency as we wanted to make it, we had to make some important architectural decisions, one of which was betting big on trusted execution environments. Because the current paradigm or existing paradigm for building these kind of decentralized clouds was to focus more on consensus-based compute, like smart contract platforms do. And you just will never hit the performance and costs of a centralized cloud if you have to run an application six plus times and you have to come to consensus on the compute results. And so you end up getting really bottlenecked on performance and cost and scalability if you try to go about this in the way a smart contract platform does. And so with trusted execution environments, you kind of have a very nice alternative that gives you a lot of the same guarantees as a blockchain. But you're just putting the trust assumptions on hardware instead of economic security or, you know, sort of like redundancy in terms of like, you know, consensus based compute where you have multiple people returning the same result. So you can kind of trust it's correct. And so our initial approach with our first AI agent hosting product since we had the T infrastructure was to host AI agents in trusted execution environments. And so that's what our product that we launched two weeks ago does. What's really nice about the approach we took is, and what's clear based on some Twitter interactions over the last week is we have probably the best TE infrastructure, and that's apparent by our cost for running agents in trusted execution environments compared to alternatives. We currently charge $20 per agent per month to run in the trusted execution environment where the competition is charging $400. So we're 20x cheaper than the next alternative, and we do think that trusted execution environments are going to play a big role in the future of AI agents simply because you don't have to trust the developer behind the agent or like, you know, worry about what we're seeing currently with agents where, you know, you think it's, it's an agent, but it's a human behind the scenes writing the tweets or controlling the funds in the agent's wallet or things like that. And so the verifiability and privacy benefits that trusted execution environments give you, we think become very applicable to AI agents and specifically the flavor of AI agents which are being termed autonomous AI agents. Because there are a range of AI agents. Maybe some you don't need to be autonomous. If you want one to just help you schedule some things, yeah, maybe that doesn't need autonomy. But I still think as like what you said before, the motes are data and That applies to LLMs, but it's also gonna start applying to agents. Because if you want an agent to be really good at something, the more data you give it, the more proprietary data you give it, you don't need tons of data, but if you want your AI agent for your company to be the best social media agent or the best sales agent for your company, you're gonna have to give it as much of your company's data and access to as much of your conversations and internal know, conversations and chats and emails and documents as possible for it to be as educated as possible to do the best job selling or marketing your product. And so essentially I do think those trusted execution environments become important because you're not gonna wanna give all that data to, well, A, you're gonna have privacy concerns around that data and giving away, you know, your kind of the Coca-Cola formula equivalent for your company. But also, you know, for some even non-business use cases, if you see some of the use cases of agents taking off, it's like these AI girlfriends. And so, you know, if you want, right now you're just chatting with them, but we could all see where this is going. You have voice, you have image generation, you have 3D video generation. You know, it doesn't take one's own imagination to go very far to realize, you know, where these things are going to go and how powerful they're going to become. And I think people are going to want some guarantees around knowing the conversations they're having with this, you know, agent, whether it's business or personal, they have very strong guarantees that those conversations or that information they're sharing with this agent stays private and it's verifiably private. However, In launching this first version of the agent product, we've also realized there's another very big opportunity to service a less technical user and helping them get involved in the agentic future. Because if you see, there's a ton of these frameworks right now, a ton of these platforms emerging around agents, but it's very technical right now, you know? And the barrier to entry is quite high. Like you can't use any of these frameworks really without having some level of understanding of technical skills. And it's not even the agent creation process that's the hardest part. A lot of people can create and analyze a character file, write some lore, write some sample tweets. That stuff's easy. The hard part is then when it comes to deploying the agent and running it. And so the two areas we're doubling down are really on the privacy verifiability side. powered by trusted execution environments. Right now we offer it for agent hosting. It's a no-brainer for us to move more into the inference direction and be able to also offer private verifiable options on the inference side of things, which we do think is going to become more more important as well. And with an open source model like DeepSeq, you could run that in an H100 GPU in confidential compute mode. The overhead you're seeing. in trusted execution environments versus raw servers or raw GPUs is low to mid single digits, five, six, seven percent, and the cost is comparable. So that's really where we're focusing is trying to bring TEs mainstream by using them and offering them at prices, performance, and developer experience that rivals centralized solutions. So there's really no reason not to use them. However, we will offer non-TE options because there are certain use cases where the costs to run things in TEs are a little bit cost prohibitive. And so we'll offer both. And having both will just encourage more people who you could get started on a cheaper non-TE option to use the TE private verifiable options as they get into use cases and start sharing more data and improving these agents to do things where the privacy and verifiability becomes more more important. if you want to give it a wallet, if you want to handle financial things, et cetera. So I'd say the two biggest bets here that we're making are trusted execution environments. That's where we think we could play a big role related to AI agents and just having the best price, best tech, best product related to trusted execution environments and AI agent use cases. And the other one is going to be on lowering the barrier and making it extremely easy for non-technical users to get involved. in this agentic ecosystem, building agents, using agents, and managing agents without needing technical skills. So that's really where we see ourselves fitting in. The TE side is kind of more on the infrastructure side. But we have some plans and some things we're working on where we think we could really make some big strides on the user experience side. to creating agents, building agents, and not needing any technical skills to be able to do any of the cool things you want to be able to do with agents. I spoke with Nelion and also Oasis in the last couple of weeks, and both of whom also utilize trusted execution environments, as well as other privacy enhancing technologies. The approach that they're taking is TEs are just one part of the, it's just one piece of the privacy enhancing technologies that they're making available to developers. But they're giving developers the option to, know, which one, looking at a spectrum of privacy, which one, At which point in the spectrum do they feel like it makes sense for them and their application? And so they're making it the choice belongs to the developer. And then what Nillion, for example, is doing is they're providing several privacy enhancing technologies and then the user kind of selects given kind of their needs. I think the AI agentic future and using TEs makes sense. As you're speaking with developers in the Web2 and Web3 space that are not super familiar with privacy enhancing technologies, but are more familiar with AI and maybe training in LLM, how do you work with them in terms of tutorials and getting them up to speed on privacy technologies? What a trusted execution environment is? How do you do that in terms of education? Yeah, so I'd say a lot of that already seems way too much for the average end consumer. I think we have to dial it back a lot. I'm not sure the average user knows what they need. And I definitely don't think they want to spend time understanding the nuances between all these different technologies for privacy and all the different options. think, I think trusted execution, like I, I am a big fan of all these privacy initiatives. I think they're all cool. Fully homomorphic encryption, you know, the the nil compute that Nilian's working on. I think they all have merit, ZK proofs. But they all have different trade-offs, you know? And I think what we're really focused on is like traditional metrics, such as cost and performance. So it's like... There's a lot of talk in web three about cool solutions and which one is better than this, but it's like, A, when we've researched those other technologies, there always seems to be a huge performance or cost burden with using them. And, That's one of the beauties of trusted execution environments is you can kind of get pretty close to like the normal experience, normal prices, normal performance, normal developer experience. But I'm not saying those other ones don't have merit. It's just. Like what we're trying to focus on is just distilling it and making it as simple as possible for someone to get those benefits without needing all this education. The same way you use Signal as an alternative messenger. for privacy reasons without need, like they use trusted execution environments in their architecture, but they're not educating people on trusted execution environments. They're just educating users on the benefits they want, which is privacy. they're sort of, the material's there, it's open source. You could find it if you want. But to the average user, I think you need to dumb it down. Otherwise you risk. too much cognitive overhead that you scare them away. And so I would say that's sort of the approach we're taking. And I'm not saying optionality isn't good, but I'm just saying if you look at a lot of the best products in the Web3 space even specifically recently, a lot of the ones that have found success have been opinionated on a stack. It's not that pump that fun supports multiple chains, you know, they've decided salon is the best chain to launch meme coins or you know, these types of coins on they've been opinionated about a bonding curve as the best mechanism to launch with and it's not saying that that solution is going to be the best for everyone, but for a large majority, especially of less technical end users, the product's great and it works great. And it's the easiest, simplest way to launch a meme coin. And it maybe isn't perfect. Maybe there's some flaws with their model and the bonding curve and this and that, but it doesn't matter. It's a good product and they would not have built as good of a product. if they went with optionality in terms of which chains to launch on and which mechanism to launch with. Because it creates overhead, it creates thinking, and users don't want to think. Part of building a good product is you have to eliminate thinking as much as possible for the user. And so that's kind the approach we're taking is like, we're opinionated, we think TEs are good enough. We care about cost and performance. think ultimately developers and users are going to care about that too. And we think those other technologies are probably not ready for prime time. And when they are, we'll evaluate it. But for now, it's like, let's just focus on building a really good product, really good user experience, weaving in the privacy benefits and verifiability benefits where they make sense. and sort of go with what the data is telling us in terms of where to double down. I like several things you just said, know, encapsulating the complexity from the user and from the developer. That's always a positive thing to do. And I think in web three and crypto, we actually go the other way where we, we put a lot of value in complexity and we put a lot of value in buzzwords and trying to almost confuse the customer thinking that that's like a big flex when it's really not. think going the opposite way, trying to make things simple. is actually really, really hard to do. And I like that you guys, that Fleek has a very strong point of view on Tease and going the agentic space. I think that's having a strong point of view there, makes your mission more tangible. people won't second guess, like, what are these guys about? I mean, it's very clear what you guys are about. Going back to kind of your core offerings. You said earlier about how the Flick network, and forgive me if I'm misunderstanding, but is Flick network, it's a layer one also, as well as infrastructure? Because you made the kind of a, okay, got it. So that's why it made sense to make the comparison of hyperliquid, got it. exactly. Like it uses a Narwhal Bullshark consensus at a SWE Miston Labs, which is a DAG based consensus. So architecturally, it's a very cool, interesting network, but it's not a smart contract platform. Like you would not launch a token or an NFT on Flick network. It's specifically web infrastructure. We focus on stateless web infrastructure, primarily on like the edge network side of things. The TEs are being added to the network, but that's, it's a separate node type. But like what's live on the network today is Flick functions, for example. So if you host a website on Flick and it's a Next.js app or you're using Flick functions in your app. It's like a Lambda alternative. Those serverless functions, edge functions run on the network. We have cron jobs coming that'll run on the network. So you can build a backend framework with just those two primitives alone. There's a CDN on the network. And so really it's just the web services, like a full stack. We don't handle databases. We don't store data on the network. Those are external choices you can make. So we don't lock a developer in. They could use whatever storage or database layers they want. But yeah, that's what allows us to be super fast, super performant. And what you are seeing with a lot of this agent infrastructure and frameworks is they are moving more in a serverless direction. And so we think we're extremely well positioned with Eliza V2 and some of these newer frameworks and updates they're making where... A lot of these agents, don't need them to be running in long running servers. And so that's why between Flick functions and Crown jobs, we can handle a lot of the infrastructure for these agents with the TEs we have on the network. so yeah, so that's kind of the approach there is just like use the Flick platform and the products and use cases we want to enable on the platform. to really dictate the infrastructure decisions we're making on the network side in terms of what features to prioritize and what infrastructure to really add to the network. Not too different from the way probably like an AWS grew their feature set based on the needs of Amazon and what was needed there. We kind of think about it a little bit similarly. You mentioned some core offerings. I'd like to kind of back up a little bit and talk about what are the key core offerings that Fleek currently provides and your target audience that you're really trying to serve. Yeah, so I would say our bread and butter up until now has been website and app hosting. We host maybe like 60 or 70,000 apps on fleek. Those are people hosting. Yeah. Yeah, that's been like just a lot of web three projects using it to host websites or apps or dApps. More recently with Fleek Network, we've been able to support more full stack applications, Next.js framework, things like that, that do more dynamic loading and like just building dynamic websites and apps instead of just static, which is all you could really do on IPFS. And so that's been where we've made our name. But now with the agent hosting, we're kind of on board that agents are probably the new websites and apps. There's no need to take users out of where they live, Discord, Twitter, and force them into an app that they have to download or a website they have to click through a bunch of UI screens. It's just like... An agent can do that, they can do it where the user already lives. It just feels like a much better experience for a lot of things on chain and off chain. And so for us as a business in the business of deploying websites and apps, it was almost a necessity to move in this direction because otherwise someone else, if you're gonna get cannibalized, you might as well cannibalize yourself. And so moving into the agent direction was a no-brainer because we already had the deployment platform. We already had all the CI, CD. And what you need to do to support that for a full stack application is not too different from what you need to support the infrastructure and DevOps needs and scaling needs of an agent. Because they're pretty like. similar, like a website or an app and an agent, they're all just these abstraction layers on top of some underlying infrastructure packaged into a user experience essentially. And so, A, we were in good position with our existing platform and the T infrastructure we had bet on for like, basically needed it for Flick network because otherwise we had an issue of How do we prevent nodes from accessing database secrets or environment variables or API keys? Because most serverless compute use cases are calling some external data source. And so you have to store secrets. that's 90 % of serverless compute use cases easily are using some sort of key that's calling an external data source. So TEs were our solution to that problem. And it just so happened that once we announced we had TE infrastructure, we started to see people interested in it and the use cases they were interested in. And a lot of it was AI agents and AI agent hosting inference too, but... GPUTs, go there next, but CPUTs were a better starting point, much in line with our bread and butter business. And so that's our other core offering we offer today is the agent hosting. Now I will say we're debating whether the Flick platform is too developer and technical to provide the user experience we want to provide people on the agent side of things. and if there's an argument to be made to separate that out into a different brand. There's a very interesting case study of a project in a very similar position to Fleek called StackBlitz, who had an existing full stack developer platform. They created a concept called web containers. They created this back in like five years ago. Never found product market fit. Then... Three months ago, they launched Bolt.new, which is one of the leading prompt-based web app building products. There's like Bolt.new, there's Loveable.dev, there's v0. And that product went zero to 20 million in ARR in two months, two million users, and they just raised 105 million. And that product is growing like crazy. It's off to the races. It's a beautiful product. And I'm a big believer in prompt-based building as the future. And so that is the direction we're moving into, but specific for agents. But what's interesting to me is they made a conscious decision not to launch that product on Stack Blitz and launch it under a new brand so they could provide a very tailored experience to a more non-technical user. They were now starting to target with that product as opposed to Stack Blitz, which was a developer product. And so we are having very similar conversations and reaching very similar conclusions where we might actually separate out the agent product in the coming weeks. So we can offer the exact experience we want. and target a more non-technical user, but keep the fleet platform separate. That's a more developer, more, you know, B2B like infrastructure. If you need TEs, if you need, you know, that, that would be there. And then just basically follow the exact same playbook they did with StackBlitz and Bolt.New and kind of Bolt.New uses all StackBlitz infrastructure under the hood. One of the things that allowed them to create Bolt.new was it used web containers and the experience they were able to provide to make that product that good used web containers. And so they basically became their own use case and it worked out phenomenally. And so that's kind of the way we're approaching it with Flick where we have this infrastructure with Flick network and trusted execution environments that we think is very well suited for this agentic future. We do think that trying to stuff this agent product into Fleek is holding us back a bit from being able to go after the non-technical user we really want to target and where we think the large opportunity is. And so what we're planning to do is a move exactly like StackBlitz and Bolt.New, where we build this standalone agent product, take it out of Fleek. provide all the messaging, pricing, branding, everything agent tailored to be a beautiful experience. It uses all the Flick infrastructure under the hood, but now we could really go target and speak to those two audiences and not try to convolute things. Cause if you look at Flick today, we're already doing mental gymnastics for how to fit agent pricing into a web hosting pricing model. It's difficult, you know, it's like they're two different things. And while we could fit it all together and maybe there are some cross-selling, upselling opportunities within the platform, we think probably for the same reasons Bolt.new launched as a standalone product and wasn't part of StackBlitz. Versel did the same thing with vZero as a separate product from Versel. We're seeing the trend and we're taking note and we're probably gonna make a similar move here. Now the work to arrive at the conclusion that it sounds like you're arriving to is typically called customer development or just basic marketing research, really understanding your audience. What type of work have you done in that space to really understand the core offerings you need to provide both audiences and how they ought to look to each one, the positioning, the messaging? Can you tell us a little bit about the work that you're doing kind of behind the scenes to do that? Yeah, for sure. Well, we kind of had this approach of like, we don't know how the agent space is going to play out. Is like, is it going to be like NFTs where like, maybe you're going to have a lot of industry specific platforms where You know, there's different tailored experiences for like, if you want an agent with a token, you're probably better going to a platform that's very specialized in launching an agent with a token. You know, if you want a musician agent, like maybe you're going to a platform that's like super tailored, like just like NFTs, like there's music NFTs, there's all these different factions within NFTs. And it's, and you know, like it's probably a better experience that way. because then like, you you can tailor it to exactly that use case. Or, you know, is it gonna be more like Shopify or websites like where you can kind of provide one platform that covers the full gamut of use cases? And I think basically you'll probably see some of both happen. But our approach was let's put out something. and let's just be very diligent to listen to what the data is telling us in terms of what direction to go. And I would say those are like, there's, being pulled in two directions right now. That, that I think align with what I just explained. But that I would say, A, there's just a lot of research you could be doing on just the market and trends and how people are using agents and what use cases are finding success. But also I'd say, yeah, just, you know, we've been fortunate where even in these first two weeks of the product being live, we're seeing such strong uptake of the product. and such strong inbound in terms of partnership requests or B2B, customers who want to use the infrastructure to power their own agent hosting platforms. And so while we don't know exactly where users are going to create the most agents or what the experience is going to look like, and is that going to be one platform versus a lot? We know there's a lot of people who are going to attempt different things. And so for now, we just want to be able to provide really good infrastructure to enable those platforms a better option than handling their own servers and scaling their own agent infrastructure on AWS. And so with our TE product, what's crazy is even in a TE. It's still cheaper to host on fleek than use AWS or DigitalOcean to run something like an ELISA agent. When we offer a non-TE version, it'll be even cheaper because there's people who want to host thousands of agents already. And it's like, we want to be able to just like deep seek what happens when you lower the cost of what it takes to actually deploy and run these agents. Who knows, you know, will you see in the future of a big social media platform or a platform with hundreds of thousands or millions of users, deploying agent for every user? I'd say it seems crazy that that wouldn't happen eventually, you know, in some context where it's useful. And so that's sort of our game plan right now is provide the infrastructure to allow people to innovate. and lower the barrier to allow new people and less technical people to get involved. And so those are sort of two guiding North stars right now. We're not sure which one is a better bet. So we're kind of putting equal weight on both and just being very diligent to talk to as many of the early users using the product as possible. and the market signals we're getting to dictate sort of which of those two we start to lean more into in the coming months. that makes sense. I think you guys have it right on with listening to your customers. And it's encouraging to hear that there's so much inbound demand. And my guess is a lot of the inbound demand is both Web 2 and Web 3. Is that correct? Yeah, and I think that's what people like about Fleek, even in this early first version of the product, is for a lot of the low code, no code agent platforms, especially in web three, they almost all force you to launch a token. You know, like there's, there's not many platforms where you could launch use one of these kind of web three frameworks like Eliza, for example, and have a no code platform where you're not forced to launch a token as part of the agent creation experience. And so I think that's really, um, one of the areas where finding success is not forcing users into a specific, you know, mold. so yeah, so I'd say, but what is interesting in the data and the signups is that because we don't force a token, we are attracting people who are from a more traditional business. Like I just saw one project that signed up a few minutes before the call. And like, they're just an agency focused on building brands and storytelling. And it's like, all right, cool. I have no clue what they have planned, but I'm excited, you know, to sort of see what they launch because you can, you know, like, so that's kind of what is very encouraging to me is, is we're seeing people from all over. And there are some trends, but there's nothing that is like standing out as like, okay, this is what everyone wants to use agents for. It's all over the map, all over different industries. And we just got added to the Eliza repo today. So if you go to the Eliza repo, like there's now in the read me an option to one click deploy on fleek. So yeah, so we're seeing a lot of uptick even just from that in terms of people finding out about fleek and they didn't add it for nothing. It's because we are even in TEs the best price for hosting. Cause on AWS you got to rent an EC2 server. So you need to be technical. Now you gotta worry about server maintenance, management, DevOps, in addition to just paying higher than $20 a month. Or, you like, these other TE options are $400 a month, but yeah, even if you don't care about the TEs, we're still the best price, best developer experience, where you don't even have to be a developer, don't force you into launching a token or anything like that, and just trying to give you as much freedom and creativity as possible. But there's a lot, like I said, left to be desired with the product. We're on it. In the next few weeks, that product is going to improve tremendously. We've been listening to all the feedback that's driving the roadmap first and foremost. But yeah, I think that's where, you know, they're seeing like, yes, this is a better option. If you want to host an Aliza agent, Fleek is your best bet. Deploy it in one click. You know, run it.$20, no other costs, no hidden costs, no DevOps, no scaling, no nothing. And so I think that's really where we want to just double down is, in providing that best experience, whether it's a direct developer or consumer trying to build an agent or a business trying to deploy thousands of agents for their users. We just want to be there as a better option than these existing clouds, tailor it to agents, provide the T non-TE versions, let people choose, let them do whatever they want and see what the data tells us in terms of what are these people doing? and how we could provide a better experience. I'll just preview, like right now to like deploy an Aliza agent, you gotta go get your own OpenAI key. Terrible customer experience. They come in, they want to deploy an agent. First thing you got to tell them is go get an OpenAI key. So in the next version of the product, we handle that for you. We will give you the LLM key. Maybe we start with OpenAI. Maybe we start with DeepSeq even. Maybe we give you a few options. But to the point about optionality, I don't think the user cares. I honestly do not think they want to learn about all these different models and all these different considerations and tokens per second and cost. They just want one click. Yeah, so we're gonna default to what we feel is the best option in settings. If you wanna customize to a different model or upload your own, be our guest. We're not gonna force you into anything, but we wanna default route that is extremely good. The customer doesn't have to think about anything and we take as much of the friction off of their plate as possible. So they just go from idea for an agent, to an agent running that they could start interacting with and talking with and improving and then just give them all the tools they need to add whatever skills, improvements, feedback as they want and make it as conversational as possible. You know, I'm reminded of a, I worked at a billion dollar outdoor sports retailer about 10 years or so ago. Uh, no, no, a different one. And, but what's really interesting is that this site, this, this company, about 80 % of their sales came from the call center. And, but when you look at how they made the sales, it wasn't even a call center. It was actually a chat center. And so. it was an army of sales reps chatting with customers. And as I looked into it, and the reason why that center had so many sales or was really the bulk of the sales is because the product was quite technical. If you're to go camping in the Himalayas, know, the customer wants to know about the material, the thickness in terms of millimeters, the threads, I mean, really, really technical. that's really hard to do, you know, speaking with a customer, but for some reason it was just easier to do over chat. And I remember even back then kind of just quipping, you know, what if like all this was done by AI and we're now at a point 10 plus years later where that's actually possible. And so I'm really curious if that company is, is looking into creating agents and then redeploying all their human, sales reps to, doing other work within the company because I can see agents doing a really good job. Yeah, absolutely. the person comes in, maybe you have their email. So now you run some software to go, all right, give me all the data on this person that you could get from this email in real time. And now that agent can know enough about this person to provide them an extremely tailored and personalized experience, even just in that chat, that might increase your conversion rate tremendously, you know? the trip that they're going to, the geography, the weather. mean, just the potential is massive. yeah. And I think consumers are going to love it because you get text messages from these brands now. They're not great. It's some coupon and you know, some 4th of July sale, but they're already texting you. Imagine, you know, you hit them with a little, you know, witty text message that you know, if it's a young kid was on brand with how young kids talk, if it was a parent, you know, you you could customize these things incredibly well. And I think the personalized web is it's going to be like a lot. Yeah. Like just put on steroids with agents. And I think that's going to lead to just improvements across the board. I agree. But yeah, e-commerce is, I think, such a low-hanging fruit area of opportunity for agents. for sure. Going back to Flick network, as a layer one, so speaking now specifically to the Web3 audience, and again, you don't have to answer, but I'm curious about, know, the most layer ones will have a token and as a coordination mechanism and also to pay for gas, et cetera. I guess, what are your thoughts on that or any plans to go that direction? because Fleek is already a really useful piece of infrastructure for Web2 and Web3. Yeah, I'll stop there and let you answer. Yeah, mean, well, a few design decisions that will for sure remain is we don't believe in using the token as a payment mechanism. So it'll always be stable that the network accepts. Maybe holding the token might get you some sort of discount, but. Like I think what's nice about the regulatory environment changing in the US is I do think you're going to see a lot more interesting token models that actually lead to way better value capture that might maybe look more quasi equity like. Like I like hyper liquid. Like I think about our token a lot in the same way they do. So you need it to be a validator and run infrastructure on the network. But they're also making a ton of money on the platform and using that platform revenue to buy and burn the hype token or buy it. I don't know if they burn it, but they for sure buy it. What they are going to do with it. I don't know. Maybe they use it for liquidity or different things. And so. What's interesting and like Venice, I saw Venice launch their token today. Yeah, they did a phenomenal job. But what's nice is like, they're not running their infrastructure decentralized. And they created a very cool token model that works with the way their business is set up. And so, It reminds me a little bit of like the, well, I'd say like at a bare minimum, our token will be like an L1 network token. But what I do think about a lot is like how the BNB token has just as much if not more value on the Binance platform than it does on BNB's smart chain. where's the majority of the value of the BNB token coming from? Is it because it's useful on the BNB chain or is it because it gets you access to token launches and pre-sales and airdrops on the Binance platform? And so what I do know is I want the maximum value capture by this token. to align all participants and align platform and network and all activity towards one singular goal. And which is why I like hyper liquid a lot because. Like they could have went the uni swap route. They could have went where they had a token for the network and then a centralized platform, like front end and perps exchange that collected fees to themselves. But they didn't, they maximally aligned value capture to the token at the protocol layer, at the platform layer. And it's resulted in a phenomenal launch. a phenomenal token model, a phenomenal value capture model. and they're well on their way to becoming like on chain Binance basically. And so that's where I like, know, I use them as a guiding sort of design principle of like how to launch a token, how to treat your community well, how to optimize for the longterm correct things and, you know, and how to align a platform and a network. in a maximally efficient and optimal way that makes everyone happy and aligns everyone around this token as the main value capture mechanism. So that's sort of how I think about it. Yeah. you know, one challenge with hyper liquid is if you were not a perps trader, you know, then you wouldn't, you wouldn't use hyper liquid. And so that, that excludes a lot of people that don't trade in perps. And so they weren't able to benefit from their drop, for example. Similarly with fleek. If I'm a retail user, but I'm not launching any AI agents or I'm not a developer and I you know, I can't participate in development type things on Flick. How could non, I guess participants, that doesn't even make sense, but how could people that are not launching AI agents and or developers building an AI agent using Flick as infrastructure, you know, be part of the community and participate? Yeah, well, I think two things. think one, with the direction we're going with the agent product, the nice thing is anybody could start building agents. And I'm of the impression everyone is going to want their own agents. I think they're going to be the new apps on your phone. Like you're just going to have a bunch of agents that you're using for different things, personally and professionally. Is that possible now if I went to fleek and then I could I as a non developer build an It wouldn't be probably your favorite experience, but that's where I said over these coming weeks, that'll be addressed where. Okay. By the end of February, it should be a really nice experience, especially for a non-technical user. Where like I said, the LLM keys will handle for you, even some of the plugin keys you need. Cause it's like, you know, any of these good plugins, you need your own API key, have to go there, set up billing. And so, you know, part of what we're going to be doing is handling a lot of that for you. So you want to add a voice, boom. at it will handle billing under the hood for you. And it'll function like a SaaS product to me. Okay. With a very conversational experience. So like, if you use like these lovable.dev or bolt.new type products, like that is... That is, I think, a magical experience. And I think that something like that for building agents or building a team of agents is very ambitious because agents are a lot harder than websites or apps. But... I do think it's something to aspire to. But yes, I would say you could do it already, because we provide you three options today. You could bring your own character file. You could build a character file from scratch, or you could use a pre-existing template. But yeah, I'd say the experience, you could do it today. You'll be able to do it. But you'd still have to go get your OpenAI key. You'd still have to do some things. Go get the Twitter account from scratch. Pass in the credentials. There's a lot of room to make it a much, much nicer experience for the average person and eliminate a lot of these barriers and steps to go from idea to agent running and a user interacting with it. So that is our main focus for this next, for the month of February is really polishing that onboarding and user experience for a non-technical person. But yes, you could do it today. There's already people doing it today who are not. technical but coming back to your question about the token yeah that'll be really nice because now any retail user will be able to potentially just become a fleek user. and start building agents. And like I said, I think we'll be able to make some convincing arguments over the next month or two for why everyone should potentially have their own agents or start building their own agents. But there will be some mechanisms within the token itself. So even if you just wanted to be a retail token holder, there's gonna be value for you there where you can, you know, stake and get some upside or yield just for believing in the project and the future growth of the project. and the token model be very tied to the cash flows of the project. So yeah, that's kind of the game plan to get retail involved. Awesome. Well, that sounds like a really good place to end. Before we do so, anything you'd like to share with the audience, maybe any upcoming plans, any ways for the audience to get involved if they're interested? So I would say definitely if you follow us on Twitter, we're just at fleek, F-L-E-E-K. Or if you sign up. Like fleek.xyz slash Eliza is where the current agent hosting experience lives. But even if you just sign up for the newsletter or updates, you're going to see that product really transform over the coming four to six weeks. And so I'd say if you're technical, you could use the product today. Works great. We just launched an API on Friday. So for platforms that want to use fleek to power their agent infrastructure, needs on their own platform, feel free to reach out. We're having a lot of cool combos like that and we're happy to help there. We have the non-TE version of the infrastructure coming in a few weeks. We've got the TE version already live with some cool improvements coming there. But yeah, I would say if you want to give the current version of the product a try, you're more than welcome. Um, but right now, like you have to pay $20 to use it just because with TE's you kind of have like some upfront costs you can't avoid. But over the coming weeks, there's going to be a free version where someone can create an agent without even signing up without paying before interacting with the agent a few times. It'll be a much, much nicer experience for people to, you know, get their hands on it, create an agent, start interacting with it before they need to pay and the experience to do it. to add capabilities, to get the LLM keys, add plugins, it'll be super seamless. So yeah, I'd say if you don't wanna give it a shot today, at least sign up for the newsletter or follow us on Twitter. And I think by the end of February, you'll be amazed at how good that experience becomes. Harrison Heinz from Fleek. Thank you so much. Thank you.