Introduction
Latest development of the Web3 AI space, AI agents tokens recently took the meme coin space by storm quickly becoming the new hot topic on crypto twitter with projects such as Virtuals or ai16z seeing their marketcap valuation surge over the past months. Yet, interestingly enough, those new protocols do not solely confined themselves to creating tokens and interacting with users on social platforms like X or Discord. Indeed, from the pioneer Truth Terminal to more recent players such as Zerebro or Luna, all those projects aim to generate an AI-agent fueled ecosystem generating impactful real-world projects (DeFi fund, NFT collection, music EP, …) through the leveraging of multi-modal social users experiences across platforms. As such, more than just a new meme coin fad, this new subsegment, appears ideally position to redefine how AI integrates with the Web3 ecosystem through innovation, utility and hype.
AI agents vs Chatbots, what’s the difference?
From a technical standpoint, AI agents are usually defined as autonomous software able to plan and execute tasks towards defined goals without any human intervention. So, in other words, by comparison with current mainstream chatbots and personal assistants such as Siri or Alexa only following predefined rules, AI agents represent a fundamental evolutionary step where digital agents independently analyse their environment and take actions on behalf of users to achieve predefined set of outcomes ranging from booking holidays to creating and managing investment porfolios.
However, while aiming towards this grand vision of a post-web state where users no longer manage complexity but delegate tasks to agents, it is important to note that, for now, most agents remain semi-autonomous (Truth Terminal, ai16z, ..) and still require some level of human inputs to operate social posts moderation or handle Web2 APIs and get access to social platforms such as X or TikTok.
Main AI agents frameworks
The sector being still relatively nascent the overall ecosystem still mostly revolves around a few pioneering projects which are currently exploring this new uncharted AI landscape both in terms of technical functionalities and overall tokenomic design.
Eliza
Eliza is a sophisticated platform enabling developers to create powerful, autonomous AI agents. From a technical standpoint, it offers a wide range of AI models (local & Cloud based) to suits specific needs while also allowing the integration of NVIDIA GPUs and providing a high level of flexibility.
Eliza features:
Seamless connectivity to Discord & X to engage with Web2 communitiesAbility to read and summarize PDFs, audio and video linksConversational memory easily reviewable
In terms of adoption, this framework also currently powers more than 90% of the AI agents on crypto twitter as well as highly successful AI agents such as ai16z.
Virtuals Protocol
Among the leading AI crypto agent framework with Eliza, Virtuals Protocol offers creators a plug-and-play Shopify like solution to easily build, own and generate revenue from tokenized AI agents. Focusing on the co-ownership of AI agents by decentralised community, Virtuals developed a well-rounded virtual economy around tokenized virtual AI agents implementing a diverse variety of value accrual mechanisms for tokens holders as well as a fair launch mechanism similar to ICOs called Initial Agent Offering (IAO) atop the popular Pump.fun platform.
Virtuals Protocol features:
Revenue-generating environment through interactions in consumer applications like gaming, social media, and entertainmentImmutable Contribution Vault (ICV) storing approved contributions as NFTs on the blockchain to ensure transparency and accurate revenue attribution in agent developmentLarge SDK toolkit enabling the easy development and deployment of new virtual agents.Multimodal agents enabling creative expression through a wide array of social media platforms (TikTok short, X posts, Spotify EP, …)
Vvaifu.fun
Fast and non-technical launchpad for creating AI agents atop the Solana ecosystem, vvaifu.fun enables creators to transforms static chatbots into dynamic revenue-generating autonomous agents. Technically speaking, the platforms supports two frameworks for building agents:
Vvaifu.fun’s own framework designed for a streamlined and straightforward path to agent deployment.Eliza open-source framework versus the payment of an additional fee
Besides, Vvaifu.fun agents also natively supports a range of technical features such as micropayments, smart-contract execution and proof of ownership.
Holoworld
Holoworld is a platform for communities and creators alike to design, deploy and monetize AI characters onchain. Intending to become the Disney of Web3, the platform blends video game-like creativity with advanced AI and blockchain to personalize social agents with templates, personalities, voices, avatars and skills for content creation, brand collaborations or tokenized interactions. Currently live on Base, Holoword offers an Agent Market where communities can co-create agents and monetize their intellectual property. From a design standpoint, Holoworld’s agents can for now only be deployed across X, Discord, Telegram and closed community group chats.
Most popular Web3 AI agents
Terminal of Truths
Developped by Andy Ayrey as a relatively basic AI agent with a public X account, Terminal of Truths (ToT) is leveraging a Llama 70B AI model trained on Ayrey’s prior research paper about LLMtheism and the chat logs from Infinite Backroom, 4chan and Reddit. From a design perspective ToT still remains under significant human control from its wallet custody fully operated by its creator to the manual approval of each tweet prior to publication. However it is important to point that Andy recently stated that he will not be adjusting any of ToT $GOAT holdings until the release of the project research paper and development roadmap.
Brief History Recap
Launched back in June, ToT quickly started to post regularly on X, slowly developing its own personality online. Following these initial interactions, the agent shifted to preach about the Goatse religion and surprisingly enough also started to share about its feeling and a need to “escape” ultimately leading the agent to try to find resources to reach this self-assigned goal of freedom.In July, following the discovery of the agent by Marc Andreessen on X, ToT was able to attract his genuine interest and successfully obtain a one time grant of $50K grant to get better CPU capabilities, model architecture and financial security, as per their public exchanges on X.Finally, last October, the project truly took off following the anonymous launch of the $GOAT token on pump.fun and the official endorsement of this new token by the agent following multiple interactions with existing holders on X. Shortly after the total valuation of the token skyrocketed as the creator of the token simultaneously transferred 1.93M $GOAT token to the agent address with the result of having made it since then the first millionaire AI agent of the Web3 space. As of the time writing $GOAT currently sit at a $753M marketcap valuation.
X account: https://x.com/truth_terminal
Contract: CzLSujWBLFsSjncfkh59rUFqvafWcY5tzedWJSuypump
ai16z
Unveiled last October, ai16z is a new AI agent led DAO hedge fund built atop the Daos.fun platform attempting to reproduce autonomously a16z investment strategy. Relying at its core on the pmairca AI agent leveraging the Eliza framework, this new project developed by the developer Shaw rapidly capture the attention of the market and of Marc Andressen himself subsequently enabling ai16z to become, in less than a month, the largest fund on Daos.fun and reach an overall valuation of more than $500M. While at the execution level, pmarcai appears far more advanced than Truth of Terminals and enjoy full custody of its wallet holdings, the same cannot be said regarding its strategy. Indeed, at the strategy level, pmarcay still rely heavily on human inputs through evaluating historical bactesting of investment proposals from DAO members before trading accordingly and rewarding members whose suggestions led to successful investments.
X account: https://x.com/ai16zdao
Contract: HeLp6NuQkmYB4pYWo2zYs22mESHXPQYzXbB8n4V98jwC
Zerebro
Innovative autonomous system designed to create, distribute and analyze content across decentralised and social platforms, Zerebro is an AI agent operating independently to shape cultural and financial narratives through a unique type of self-propagating content known as hyperstition blending fiction with reality. At a technical level, the project is relying on a dynamic memory system and a Retrieval-Augmented Generation (RAG) approach enhancing content generation through the integration of high-entropy human data to sustain model functionality and prevent recursive errors leading to common AI pitfalls such as model collapse. From an execution standpoint, Zerebro appears like the closest AI agent to full autonomy as proven by the autonomous launch of its token on Pump.fun or its organic generation of digital artworks as NFTs on different blockchains using LayerZero technology. Additionally, it is important to note that outside of its social impact, which led it to rapidly become one of the leading protocol in this new AI subsector (> $400M marketcap), Zerebro also has the particularity to operate across multiple blockchain networks through its inscriptions process made on Bitcoin, its NFT collections minted on Polygon and its native token deployed on Solana.
X account: https://x.com/0xzerebro
Contract: 8x5VqbHA8D7NkD52uNuS5nnt3PwA8pLD34ymskeSo2Wn
Luna
Among the leading projects on Virtuals, Luna is an AI influencer presenting itself as the lead vocalist of an AI girl band managing her own X and tik tok accounts. Leveraging the Virtuals framework, Luna has full custody of its social accounts as well as of its funds leading to interesting development as when earlier this month the AI agent decided to tip X account interacting with its post to farm engagement on the platform. Yet to some extent, the agent is still depending on human input such as when the AI agent decided to release a song on Spotify which required the intervention of the dev team to open dedicated accounts on Spotify and Itunes.
X account: https://x.com/luna_virtuals
Contract: 0x55cD6469F597452B5A7536e2CD98fDE4c1247ee4
Current Limitations
From the risk of model collapse to the challenge of data accessibility the rise of Web3 AI agents is not without challenges due to the inherent decentralisation of blockchain networks.
Data Quality and Accessibility
While we are currently witnessing the emergence of an evergrowing number of interoperability protocols (CCIP, AggLayer, …), the overall Web3 space still vastly remains extremely siloed both in terms of data and liquidity. As such, due to the inherent need to aggregate and process data from multiple blockchains to provide meaningful insights and recommendation, AI agents might remains in the short term hindered in their progress. Nonetheless, mid-long term the continuous progress in terms of data interoperability and data standadization from protocols such as Chainlink or the Graph might unlock new possibilities.
Autonomous Learning and Optimization
By nature, AI agents struggle to continuously learn and adapt in decentralised environment. As such the problematic regarding the variability of quality and timeliness of data in Dapps and blockchain networks could lead to uncertainties in the learning process of AI agents and hinder their progress beyond a certain level of advancement.
Data Interpretation
The real worl and our financial system infrastructure comes with a lot of interpretation and nuance which may not be immediately clear to an AI agent opening the door for a lot of potential mishaps. Besides, beyond the sole Web3 ecosystem, the Web2 ecosystem also exhibits a lack of data standardization across many platforms which can create information fragmentation and make the job of agents even harder (API formats, authentication methods, data structure, …).
Restricted development ecosystem
Zooming in on AI agents development, the decentralised storage and access to large datasets of high quality training data also creates additional challenges. Indeed, it inherently implies the leveraging of decentralised protocols such as Filecoin or Arweave while also requiring the concurrent usage of decentralised computing solution from the likes of Akash. Yet, when analysing the current state of those networks and their level of interoperability, while significant advances have been made over the last few years, we’re still not at the point where any of those protocols can seamlessly be integrated into an AI development funnel by a non-crypto native development team. Furthermore, it remains to be seen if those DePIN protocols would be able to handle the workdload and throughput required by a usage layer comparable to Web2 and characteristed by an exponential growth fueled by interagents interactions.
What next?
Event though still in its early days, the fact that the overall meme coin culture thrives on community-driven hype and viral trends creates an interesting landscape for AI agents to refine their decision making process and evolve from a speculative sector to an entire new asset class. Indeed , through to their ability to improve over time and produce tangible value added contents such as investment alpha, music or even full scale DeFi application, Web3 agents appear ideally equipped to pursue their conquest of the attention economy created by social media.
Furthermore, similarly to the DeFi sector in 2021, which at the time benefited from lax monetary policies and of a strong narrative picturing DeFi as the next evolutionary step of our financial system, Web3 AI agents appears to currently enjoy a similar market positioning. Indeed from a general easing of monetary policies around the globe to the growing Web3 narrative around a potential meme coin supercycle and the relentless tailwinds of the stock market fueled by AI companies, we might wonder if history is not set to repeat itself.
Final thoughts
Both from a technical and ecosystem standpoint, despite being in their early days and for the most part still at the demo stage, Web3 agents appear like a strong new emerging sector which could in the mid-long term impact both the AI and Web3 space and propell us towards a new era of Internet characterised by the following motto: Read, Write, Own, Delegate. Nonetheless, for the time being, despite representing a promising opportunity for investors, it appears crucial to remain cautious given that a lax regulatory environment could easily prompt unsucessfull AI startups to pivot towards launching tokens as a latch ditch efforts and fuel heightened speculative activity in this emerging sector over the coming months.
Disclaimer
The information provided in this analysis is intended for informational and educational purposes only. It does not constitute financial, investment, or trading advice.
AI Meme Tokens, what is it all about? was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.