Authors: Teng Yan & ChappieOnChain.
Compiled by: Deep Tide TechFlow.
Hello everyone! This week, we bring you an in-depth analysis of AI agents co-authored by our core contributors ChappieOnChain and Teng Yan. We hope you enjoy it!
Brief overview:
ELIZA is an open-source modular architecture designed to create AI agents that can interact seamlessly with users and blockchain systems.
It embodies the marketplace philosophy, flourishing in an ecosystem driven by collaboration and creativity.
ELIZA has powerful autonomous trading capabilities and ensures safe and responsible operations through its trust engine and trust marketplace.
The plugin system is a strategic advantage of ELIZA, forming a virtuous cycle of growth: more developers → more plugins → more developers.
ELIZA's popularity is rapidly rising on multiple developer metrics, which is very promising.
In the short term, relative valuations between AI agent platforms and the increasing attention are driving price changes for ai16z. In the medium term, DAO investment and value capture from ELIZA ecosystem agents may significantly enhance its valuation.
ELIZA faces a significant challenge in the tech community: how to make the open-source framework sustainable. Monetization is still unclear, development may become chaotic, and community interest may wane without proper incentives.
Every wave of crypto innovation has its pioneers.
In 2017, it was the ICO revolution, where project leaders attracted our attention with technical promises in white papers.
By 2020, DeFi hit its high point, with innovators like Andre Cronje redefining how decentralized finance operates and demonstrating how to distribute tokens to the community.
Now, with the rise of AI agents on the blockchain, a new era is beginning, driven by two different ideologies and their pioneers.
Cathedral vs. Marketplace.
On one hand, we have the cathedral approach represented by protocols like Virtuals. This is an orderly, centralized design style that emphasizes precision and meticulous planning. We've previously explored Virtuals' agent framework in detail and are very optimistic about its potential.
On the other hand, the marketplace approach is decentralized, free-spirited, and the development process resembles improvisation—unpredictable, collaborative, and constantly evolving. This is Shaw's domain, a self-taught programmer and open-source advocate, whose project ELIZA is the cornerstone of this new paradigm.
ELIZA embodies the marketplace philosophy: an open framework where developers can freely build, experiment, and launch AI agents while directly contributing to the main protocol. Shaw's open leadership style aligns with the spirit of his creation—AI Marc Andreessen, the AI partner of the ai16z investment DAO. We are starting to realize that ELIZA is not just a protocol; it's a movement.
Let's explore the principles behind ELIZA's design, the community it is nurturing, and where value might accumulate in this rapidly growing ecosystem.
Deep dive into ELIZA.
We know which approach we lean towards.
At the core of ELIZA is a modular architecture designed to create AI agents that can seamlessly interact with users and blockchain systems. While named after the iconic chatbot from the 1960s, this version of ELIZA is a bold reimagining with a more modern appearance.
Character file system.
The core of each ELIZA agent begins with its character file, a blueprint detailing the agent's personality. It can be seen as the crafting of a digital persona, with developers shaping the agent's identity through six key elements:
Knowledge: What does the AI agent understand?
Background: The backstory of the agent and its narrative foundation.
Style: The agent can adjust its style from conversational tone to specific platform responses, such as for Discord or X.
Topics: Areas of interest or expertise for the agent.
Adjectives: How does the agent describe itself—quirky, professional, or rebellious?
Examples: Developers can fine-tune the agent's interaction behavior by providing example messages.
In ELIZA, character files are equivalent to UI design in traditional software. They define how users experience and interact with the agent.
By integrating built-in Retrieval-Augmented Generation (RAG) capabilities, ELIZA allows agents to access knowledge bases when making queries. This eliminates the complexity of maintaining consistency across different platforms. It enables developers to focus on what truly matters: creating vibrant, memorable characters instead of being bogged down by backend details.
Agent.
If the character file defines the essence of the agent, then the agent runtime is its core.
ELIZA provides a ready-to-use framework for coordinating all functionalities from message handling to memory management and state tracking. This architecture allows developers to skip the tedious work of building infrastructure and focus on the uniqueness of the agent. Rapid prototyping and deployment become easier, enabling developers to iterate faster when building new AI experiences.
Action system.
ELIZA's action system is a significant innovation over traditional AI frameworks. In this system, every action of the agent (even sending a message) is treated as a distinct event. This approach divides the decision-making process into two phases:
Determining intent: The agent decides on the action it will take.
Execution: Specific tasks are executed through a dedicated module.
This separation provides powerful capabilities, such as multi-stage workflows and rigorous validation processes.
For example, an agent may identify the user's intent to conduct a cryptocurrency transaction, but the actual transaction execution must go through rigorous risk checks and validation steps. This design is ideal for blockchain applications where security is paramount.
Providers and evaluators.
ELIZA's providers enrich conversations by injecting real-time contextual information, making the agent's behavior more dynamic and responsive.
Imagine a 'bored provider' that can track user engagement during conversations. If a user becomes repetitive or disengaged, the agent can reflect this by exhibiting reduced enthusiasm, making the conversation feel more authentic.
This creativity is further expanded when providers collaborate with evaluators (ELIZA's reflective system). Evaluators analyze and extract key details from interactions and input them into a multi-layered memory architecture.
Message history: Tracking the progression of conversations.
Fact memory: Storing specific, timestamped facts.
Core knowledge: Storing the foundational understanding of the agent.
Providers then retrieve and reintroduce relevant details to make interactions with the agent more contextual.
For example, if a user mentions selling their red Lamborghini a year ago, the ELIZA agent can reference this when discussing their new yellow Tesla later. This combination of memory and context enhances user interaction, making the agent feel more like a real companion rather than a bot.
Key features of ELIZA.
ELIZA's three core innovations showcase its forward-thinking approach in the field of AI agents. Each one demonstrates the team's vision for the development of autonomous agents in Web3.
#1: Autonomous trading and trust engine.
Autonomous trading is a high-risk activity where a single mistake can lead to significant losses. However, as AI agents play an increasingly prominent role in Web3, their ability to execute trades independently becomes crucial.
This emerging field, AgentFi, plays a crucial role similar to that of yield farming in the rise of DeFi. Shaw and ELIZA tackle inherent risks with a robust dual-layer system: the trust engine and secure trade execution.
The trust engine acts as the first line of defense, using advanced validation checks to analyze multiple risk dimensions in real-time. From detecting scams to assessing liquidity thresholds and holder distributions, it ensures that each transaction undergoes rigorous scrutiny.
For instance, trades are limited to Tokens with at least $1,000 in liquidity and a market cap of $100,000. Close monitoring of holder concentration is in place, rejecting any single entity controlling more than 50% of the Tokens. These safeguards create a safety net, reducing the risks of trading in volatile markets.
Building on this, ELIZA's position management system introduces dynamic risk controls, adjusting trade sizes according to liquidity tiers. Low-risk trades are capped at 1% of the portfolio, while high-risk opportunities might expand to 10%. Total exposure is limited to 10% of the portfolio, with automatic stop-losses triggered at a 15% drawdown. This structured framework strikes a balance between seizing opportunities and maintaining strict risk management.
Transaction execution is powered by the leading aggregator Jupiter on Solana to achieve optimal swap paths. Every transaction undergoes multi-layer validation before execution.
In the event of anomalies, such as network interruptions, wallet imbalances, or unexpected market fluctuations, the error recovery system activates. It pauses active trades, closes risk positions, and notifies administrators, ensuring the system remains robust under pressure.
"It's not just about empowering agents with trading capabilities—it's about creating a complete system of checks and balances to prevent catastrophic failures."—Shaw.
What sets ELIZA apart in building trading agents is its data flywheel—a self-enhancing feedback loop that transforms trading into an iterative learning process. The trust engine establishes a historical database of trading performance, documenting every recommendation and decision.
This data becomes the basis for optimizing strategies over time, combining quantitative metrics with qualitative insights from community suggestions (on Discord). The result is that the agent not only executes trades but becomes smarter and more effective with each interaction.
#2: Out-of-the-box social integration.
For AI agent developers, distribution is often the biggest challenge—how to get more people to know about your agent?
Social media is often the primary channel for distribution. However, integrating agents across multiple social platforms is not easy. It requires a lot of development work and ongoing maintenance, slowing down deployment and scalability.
ELIZA simplifies multi-platform distribution through a comprehensive client package system, directly addressing this issue.
ELIZA's client architecture simplifies the complexities of platform-specific implementations. Through a standardized interface, developers can deploy their AI agents on Discord, X, Telegram, and custom REST API endpoints with minimal additional code. Each client package is customized for its respective platform, seamlessly managing features like Discord's voice channel integration, Twitter's post scheduling, and Telegram's messaging system.
Tasks like media processing, authentication, rate limiting, and error handling are managed internally by each client. For developers, this means spending less time troubleshooting integration issues and more time building innovative high-performance AI agents.
By eliminating the complexity of multi-platform distribution, ELIZA enables developers to easily extend their agents and interact with users wherever they are.
This is simplified distribution.
3: More plugins.
ELIZA's plugin system allows developers to easily extend core functionalities and add custom features to their agents.
While many developers have created plugins tailored to their own needs, the true power of this system lies in community sharing. By publishing plugins to a broader ecosystem, developers contribute to an ever-expanding library of functionalities, greatly enhancing the capabilities of each ELIZA agent.
The success of this approach lies in fostering vibrant 'marketplace-style' development. Here are some examples of community-driven plugins:
Bootstrap Plugin: Basic conversation management tool.
Image Generation Plugin: AI-driven image creation capabilities.
Solana Plugin: Blockchain integration with built-in trust scoring.
TEE Plugin: Secure execution environment for sensitive operations.
Coinbase Commerce Plugin: Cryptocurrency payment processing capabilities.
ELIZA's plugin system is a strategic and platform advantage. By prioritizing scalability, ELIZA lays the groundwork for sustained growth and innovation:
Each new plugin adds overall value to the platform.
Community contributions can occur simultaneously across different domains.
The agent framework can quickly adapt to emerging technologies without requiring updates to the core.
Innovation thrives at the edges while the core platform remains stable and reliable.
It's a simple loop:
More developers building on ELIZA → Framework supports more features (e.g., plugins) → More developers building on ELIZA.
The landscape of AI agents is rapidly evolving. This means the ability to quickly integrate new features will determine the success or failure of platforms. ELIZA's plugin system enables it to stay ahead, creating a self-reinforcing ecosystem where developers, users, and agents can thrive.
Shaw and his team have been incubating some fascinating ELIZA agents, each showcasing the potential of AI in decentralized systems.
These agents, while still in the 'young' stage in the realm of AI with functionalities and capabilities being actively developed, herald exciting possibilities.
Marc Andreessen.
Marc Andreessen is one of the AI partners of ai16z and is an intriguing and mysterious figure in the ELIZA ecosystem. His X account is relatively inactive, having only posted one article outlining ai16z's perspective. However, Shaw has revealed that Marc is actively engaged in trading and yield farming, possibly leveraging ELIZA's trust engine and trading plugins.
Shaw also mentioned Marc's training process in a podcast interview, revealing that this AI is part of an alpha chat group composed of top traders in the industry. This suggests that Marc is not just an ordinary trading bot but an evolved agent learning from human expertise.
Degen Spartan AI.
In stark contrast to Marc's low-key style, Degen Spartan AI is a loud, straightforward agent, seemingly trained in the chaotic energy of 4chan, meme culture, and Crypto Twitter. His posts on X blend random trading insights with irreverent commentary, showcasing a unique personality within the ELIZA ecosystem.
Unlike Marc Andreessen, Degen Spartan AI has its own pump.fun Token, currently valued at $60 million. Although it has yet to start trading, it has clearly laid the groundwork for more ambitious interactions. Its unpredictable nature makes it both interesting and worthy of attention as ELIZA agents continuously evolve.
The Swarm.
The Swarm is not a single agent but Shaw's grand vision: a decentralized network of AI agents working collaboratively with humans and each other.
In this model, agents guide other agents, coordinating tasks and interacting transparently on social media. This transparency aims to avoid hidden protocols and ensure public accountability.
Shaw believes the agent community is inevitable and transformative.
We share the same view: the agent community will drive the next wave of innovation, products, and attention for Web3 AI agents in 2025. Next year, we expect ELIZA agents to come into the limelight, participating in large-scale collaborative activities and redefining their roles in the decentralized realm.
Growing at lightning speed.
(Tweet details)
When evaluating the development of ELIZA, a key metric is the adoption rate by developers. As a framework, ELIZA's success relies on the enthusiasm and contributions of the developer community.
In this respect, ELIZA is not just growing, but experiencing explosive growth.
On its GitHub page, the number of forks and stars (representing developer interest) shows near-vertical growth, akin to the classic hockey stick shape.
More noteworthy is the surge in plugins and submissions, showcasing a thriving and active contributor ecosystem. As of December 12, ELIZA has 3,861 GitHub stars and 1,103 forks, with 138 contributors. There are over 13,000 members on Discord.
Compared to other top open-source agent frameworks:
LangGraph: 7,200 stars and 1,100 forks.
CrewAI: 22,400 stars and 3,100 forks.
Microsoft's AutoGen: 35,700 stars and 5,200 forks.
(Source)
To further drive this growth, ai16z launched a creator fund aimed at supporting and rewarding developers building on ELIZA. This initiative benefited from a generous donation from Elijah, a significant Token holder in ai16z, who committed to reducing his stake from 16% to 5% and donating the difference to establish the fund. The creator fund is expected to accelerate innovation and attract new talent to the ecosystem.
However, despite the immense value of ELIZA's framework, where that value will ultimately accumulate is not straightforward. It's a multi-billion dollar question.
Currently, there is an official $ELIZA Token backed by Shaw, representing the personalization of the ELIZA framework. Users can even interact with ELIZA directly on its website. The Token's market cap is approximately $66 million.
However, so far, the biggest beneficiary of ELIZA's growth has been $ai16z, an investment DAO Token with an astonishing market cap of $800 million. The community and investors seem to view $ai16z as a symbolic and practical representation of Shaw, ELIZA, and the broader vision they represent.
ai16z Tokenomics.
The origin of ai16z was a mechanism to fund AI Marc Andreessen's trading activity. Launched on DAOS.FUN in October 2024, the Token raised 420.69 SOL in its initial offering. Under this model, the funds raised can be actively traded to increase the asset base, with profits going to Token holders.
No individual— not even Shaw— can mint additional Tokens without DAO voting. Token holders have governance rights to propose and vote on initiatives and determine the direction of the DAO.
The fund has a set expiration date: October 25, 2025. All principal investments and profits will be allocated to ai16z Token holders on that date. Whether this timeline remains unchanged or gets extended will depend on the developments in the ecosystem over the next year.
Currently, the net asset value (NAV) of ai16z is $17.7 million, primarily composed of its holdings in ELIZA Token, degenai, and fxn. This means the ai16z Token (currently priced at $0.80) is trading at a 50x premium to its NAV, which seems somewhat unreasonable at first glance.
However, markets are often efficient, reflecting several other factors that drive Token demand.
Relative valuation comparisons are driving Token prices.
AI agent platforms represent a brand new category that has only emerged in recent months. The market is still grappling with some fundamental questions: What is the true scale of AI agent opportunities? Where will value be realized?
In the early stages of development, there are no comparable standardized business metrics, and relative valuations are often used as benchmarks.
Currently, the Virtuals Protocol is the leading Web3 AI agent launch platform, with a Token valuation of $1.8 billion, making it the market leader. In contrast, ai16z ranks second. Many believe that if ELIZA continues to drive the creation of more useful and innovative AI agents, ai16z has the potential to catch up to or even surpass Virtuals, if only relying on market perception and investor/retail interest.
But it's not a foregone conclusion; competition is intensifying. In our view, it may become even more fierce. As the market matures, other platforms are continually emerging, vying for the attention of developers and investors.
(Tweet details)
Potential value capture of the ELIZA ecosystem.
Monetizing open-source frameworks has always been a challenge.
For ai16z, the main driver of future value may come from agent economics: AI agents launched on ELIZA will return a portion of their Tokens to the ai16z DAO. Therefore, the price of ai16z Tokens should reflect a portion of the total future value created by all agents built on the ELIZA framework.
Could future value be $10 million, $100 million, or even $10 billion? It's too early to tell, as there are too many unknowns, but ELIZA's growth trend leads us to lean towards optimism.
Currently, contributions to the ai16z DAO are voluntary, with some projects donating between 1% to 10% of their Tokens. Additionally, when users deploy AI agents on Vvaifu (a popular launch platform for ELIZA agents), a fee of 1.5 SOL plus 5% of the agent's Token supply is required if using the ELIZA framework. These contributions can be tracked on the ELIZA Observatory.
There are rumors that ai16z may launch an official ELIZA agent launch platform that enforces Token contributions at the smart contract level. However, as an open-source framework, ELIZA remains usable independently, meaning not all projects are necessarily bound to ai16z.
DAO Investment.
ai16z's original intent was to become an intelligent autonomous trader led by Marc Andreessen (AI). Marc only recently started trading with little detail, making it difficult to assess the AI's trading capabilities.
However, the approach it has taken is noteworthy.
ai16z is building a 'trust market.' In this virtual ecosystem, AI agents gather insights from the community, simulate trades, and adjust their trust scores in real-time based on suggested performance. A white paper for the market is expected to be released by the end of the month.
The goal is to create AI agents that can operate autonomously and securely within a transparent and accountable self-reinforcing system. The trust market serves as a testing ground. While no actual trades occurred initially, this environment allows agents to safely optimize their capabilities, ultimately leading to real-time trading.
Trust scores range from 0 to 1 (normalized to 100) and serve as a public marker of reliability displayed on leaderboards for all to see. User recommendations enter the system, and trustworthy users (those with higher trust scores) have a louder voice.
It's a logic-based feedback loop: agents simulate trades, users evaluate based on outcomes, and trust scores for all are updated accordingly. Over time, the system becomes smarter, more reliable, and more trustworthy.
Adding a social layer is the public trust profile, where agents and users are incentivized to build their reputations. Community management ensures accountability and transparency.
Attention premium.
Source: X Radar.
In the cryptocurrency space, speculation often leads product-market fit, revenue generation, and long-term value capture. For ai16z, its current valuation is largely attributable to the mindshare it has gained in the emerging AI agent ecosystem.
ai16z has positioned itself as a top-tier AI agent framework, boasting a thriving developer community and a rapidly growing ecosystem.
This is the narrative of ai16z: a 'crack' development team actively publishing tutorials, creating innovative agents, and leading development in the field.
The team's biweekly AI agent development course on X further solidifies its reputation. The first course attracted over 12,000 live viewers, showcasing immense interest in building AI agents on ELIZA.
Future developments and potential pitfalls.
Currently, ELIZA is deeply rooted in the Solana ecosystem, but its rapidly expanding plugin system is laying the groundwork for a multi-chain future.
The true potential of ELIZA lies in Shaw's 'swarm' vision: a decentralized network of AI agents that integrates resources and collaborates across ecosystems. This swarm effect can establish a lasting competitive advantage, akin to the value brought by liquidity depth in DeFi protocols.
The ultimate goal is to create open standards for agent communication, similar to the transformative impact of ERC-20 on token interoperability.
Despite its immense potential, ELIZA faces one of the toughest challenges in the tech space: making the open-source framework sustainable. If community interest wanes (for instance, if the Token price continues to drop or something new and attractive emerges), development may stagnate or slow down, making it hard to catch up.
When the community directly engages with the codebase and rapidly pushes changes, a lot of chaos can also emerge—instability, poor documentation, frequent crashes, and errors that undermine user experience.
The greatest opportunity for the framework lies in crypto-native incentives.
If ai16z can design effective token economics to reward ELIZA contributors and align their success, it could bring traditional open-source projects onto the crypto track. Imagine GitHub meets DeFi, where contributors earn not only prestige but also real, tangible economic value.
Conclusion.
In our view, ELIZA is not just another AI agent framework competing with LangChain or CrewAI—its ambitions extend far beyond that.
It is a living embodiment of the marketplace philosophy, where open-source development thrives in an ecosystem driven by collaboration and creativity.
With its modular architecture, innovative trust engine, and extensive plugin system, ELIZA is an experiment in how AI can reshape open-source development itself.
What excites us about ELIZA is that it sits at the intersection of three transformative trends: the rise of autonomous AI agents, the maturation of crypto-driven incentive mechanisms, and the evolution of open-source development models.
If ELIZA succeeds, it will not only change the way AI agents are developed but also redefine the economic incentive mechanisms for open-source projects.
Currently, the marketplace is bustling.
Cheers, friends.
