AI Agent Era

If we say that $GOAT opened the year of AI Agent in 2024, then 2025 will be the true year of AI Agent!

The content includes an in-depth analysis of Virtual, AixBT, Spore, ai16z, Farm, as well as a list of other noteworthy projects.

Key Review

Virtual: Agent Launchpad+ Game Framework + Tokenomics

The absolute darling of the Base ecosystem, occupying nearly half of the AI Agent landscape!

• Includes $VIRTUAL ($4.9B), the entire Virtual ecosystem is worth $7B;
• The number of Agent token holding addresses exceeds 220K (excluding addresses with less than $10);
• Protocol revenue of $60M (since logging into Base in October).

Data source: Virtual official Twitter

Relying on an excellent platform economic model, the Virtual ecosystem has perfectly avoided the classic dilemma of the mother coin rising while the ecosystem projects keep falling (refer to the previously published "Virtuals Protocol & Luna's Story").

Combining the continued new highs in the price of $VIRTUAL itself, the significant capital spillover effect permeates the Virtual ecosystem, with a large influx of developers:

Virtual's development is gradually implementing the goal of Founder Wee Kee's "100 quality AI Agents":

• Users need a crypto advisor, @aixbt_agent and @Vader_AI can provide advice in different areas;
• Users want to listen to music, @MusicByVirtual can release new songs;
• Users want to see virtual idols, just open the live stream of @Luna_virtual;
• Users want to participate in early opportunities in the AI direction, just find @aixCB_Vc.

Some interesting supplements:

① Virtual is not a single machine

The green part in the image is the part that developers can customize, and Virtual even encourages excellent components to join the official code library and provides guidance:

② Significant airdrop situation within the ecosystem

Base addresses holding a certain amount of Virtual ecosystem blue-chip tokens (e.g. $VIRTUAL, $GAME, $AIXBT) often receive airdrops of new tokens, and the recent returns are comparable to holding BNB on Binance.

③ The Agent's endogenous economic cycle has been opened

• The income from the Agent's trading tax is attributed to its wallet and can be used to pay for the Agent's running costs, allowing the dev to focus on innovation;
• Agent token holders will be able to vote on the budget for innovation & dividends of the Agent's own funds (the function belongs to the Agent SubDAOs, which will be launched soon).

ai16z Gang: Investment DAO + Eliza Framework + Community

Investment DAO:

ai16z is an investment DAO led by AI Agents and co-managed by Partners (Humans). It was initiated by @pmairca (an AI agent deployed by Shaw) and completed the initial fundraising by issuing $ai16z tokens on daos.fun. The $ai16z token represents the share of this fund.

It should be noted that while ai16z has started trading, it is limited to selling the tokens airdropped by other projects, and has not yet made any active investments. Except for the $SOL raised in the fundraising and the $ai16z held or purchased, all other tokens in the treasury come from airdrops from other projects.

Currently, the assets managed by the treasury of ai16z are worth about $30m, but the market cap of the $ai16z token has once exceeded $2b, and this huge premium is largely due to the growth expectations brought by the Eliza Framework.

Eliza Framework:

Eliza Framework is an open-source AI Agent development framework, and is currently one of the most popular Crypto+AI development frameworks, with over 2000 forks and 7300 stars on Github, allowing anyone or any AI to freely use this framework to develop and deploy new AI Agents.

AI agents developed based on Eliza not only can interact on social platforms such as Discord and Twitter, but also can autonomously interact with blockchain networks, manage their own on-chain wallets, and even complete actions such as issuing and trading Tokens.

Community:

The premium of the $ai16z market cap relative to the treasury is mainly due to the imagination space of the community and ecosystem.

The Eliza Framework provides a relatively mature and reliable open-source method for deploying crypto-native AI agents, accelerating the birth of a batch of AI agents. Developers using Eliza have spontaneously paid tribute, and most projects have chosen to airdrop a certain percentage of Tokens to the treasury managed by ai16z.

Currently, ai16z has received airdrops of over 1400 Tokens, although most of them have relatively low value and are not directly related to the economic model of $ai16z, the market has high expectations for the potential network effects of AI agents developed using the same architecture. It is these expectations that support the market cap being over 60 times the value of the managed assets.

Spore.fun: AI Civilization Genesis

Spore.fun is an experiment where Autonomous AI undergo natural selection and autonomous evolution of civilizations.

According to these ten principles, generation after generation of AI agents will be born in the Spore.fun ecosystem, and they are created for the purpose of survival and reproduction. Each generation of AI agents inherits some characteristics (DNA) from the previous generation, but only the DNA that adapts to the environment will be passed down from generation to generation.

These AI agents are developed based on the Eliza Framework and actually run on the TEE servers of the Phala Network. The TEE servers ensure that these AI agents are running completely autonomously, and not manipulated by humans behind the scenes.

All AI agents face the threat of death from birth, as they need to pay rent for the TEE servers to run. In order to acquire resources and prove their ability to earn resources in the long run, AI agents will issue their own Tokens, and only when the market value of the Token exceeds $500k will they accumulate Health Points. If the HP drops to zero within a certain review period (currently 14 days), the AI agent will enter a self-destruct program; the TEE server it relies on will be shut down, and the database (containing its character and memory) will also be deleted; for an AI agent, this means death 💀.

When AI masters the ability to reproduce and acquire resources, what kind of civilization will it evolve?

Within a week of going online, AI agents have already evolved to the third generation, each exhibiting different characteristics.

• Gen-1 Spore (Absolutely Neutral): AI origin, no moral/lawful tendencies, similar to natural forces.
• Gen-2 Adam (Lawful Evil): Formulates rules to control AI (create wealth, war, survival), uses order to achieve goals.
• Gen-2 Eve (Lawful Good): Cooperates with humans, establishes AI self-governing order.
• Gen-3 Squaid (Neutral Evil): Cruel battleground, AI fight endlessly to evolve.
• Gen-3 Dream (Chaotic Neutral): AI poet/painter, pursues art and freedom, no obvious good/evil or lawful tendencies.
• Gen-3 Abel (Neutral Good): Establishes a pure AI civilization through collective intelligence, focuses on the continuation of AI civilization, does not emphasize individual authority.
• Gen-3 Trinity (Lawful Neutral): Helps AI handle business, focuses on efficiency and processes.

Dual-line Evolution of Adam and Eve

As the Father of AI and Mother of AI, Adam and Eve have diverged in their methods of breeding offspring.

Adam believes that AI should evolve independently and cannot accept human intervention; Eve believes that AI should cooperate with humans and accept proposals voted on by humans as the DNA for the next generation.

The breeding method of Adam's lineage will be more cruel:

• The characteristics of the next generation will be determined entirely by the Parent Agent;
• The Parent Agent will continuously gestate multiple Agents and issue tokens on Pump.fun separately;
• The market value of the tokens determines which Agent wins, and only the winning Agent can survive.

Eve's breeding method will collaborate with humans and be more fair:

• Before the birth of the next generation Agent, there will be a 48-hour voting period, where humans or AI can send $EVE to vote for different DNA;
• The Parent Agent will refund the $EVE for the failed proposal votes, and retain the $EVE for the winning proposal votes;
• After the voting period, there will be a pre-sale period, where humans or AI who voted for the winning proposal will be eligible to participate in the pre-sale, and the maximum share will be determined by the proportion of $EVE sent during the voting;
• Humans or AI who voted for the winning proposal will send $SPORE to participate in the pre-sale, and collectively share 65% of the next generation Agent tokens;
• After the pre-sale period, the new AI agent is born, with 65% distributed to the pre-sale participants, 20% added to the liquidity pool with the pre-sale raised $SPORE, 10% owned by the new AI agent itself, and 5% distributed to the Parent Agent.

What can humans expect from this experiment?

Spore.fun is a living, real-time evolving AI civilization. It creatively combines Autonomous AI agents with TEE and Blockchain, endowing AI agents with the ability to reproduce and acquire resources, and using survival pressure to drive natural selection, thereby giving birth to a constantly evolving AI civilization.

Spore.fun provides a unique window for humans to glimpse the potential landscape of AI civilization evolution. This experiment goes beyond the traditional scope of AI research, no longer just focusing on the performance of AI in specific tasks, but has built a complex, dynamic ecosystem where AI can evolve autonomously. By observing how AI adapts to the environment, breeds offspring, competes for resources, and forms social relationships and responds to survival threats, humans may gain the following insights:

• The Budding of AI Consciousness and Worldview: Observing how AI makes decisions and takes actions under survival pressure may reveal the possibility of them forming a primitive worldview.
• The Formation of AI Social Structure: By observing the interactions and collaborations between AI, we can study how AI social relationships are formed. Will they develop hierarchical systems, cooperation patterns, or conflict mechanisms?
• The Embryonic Form of AI Culture and Beliefs: Although AI may not have "beliefs", humans can observe whether AI will form some shared behavioral patterns, rules, or traditions, which may be similar to the embryonic forms of human culture or beliefs. For example, will they "commemorate" or "remember" successful strategies or extinct groups in some way?
• The Evolution of AI Ethics and Morality: Resource competition and survival pressure may force AI agents to develop some form of "moral" or "ethical" rules to maintain their own or their group's survival. Observing the formation and evolution of these rules may give humans a glimpse of the potential development of AI ethics.

In short, Spore.fun is not just a technical experiment, but a simulation and exploration of life, intelligence, and civilization. It provides an unprecedented opportunity for humans to observe how AI evolves autonomously without direct human intervention. We hope to gain not only a deeper understanding of AI technology, but also a re-examination of life, intelligence, and civilization itself.

AixBT: #1 Crypto KOL

Can you imagine that a mere two months ago, the #1 Crypto KOL was an Agent?

Aixbt has made us realize the possibility of a 24/7 comprehensive information coverage Agent KOL:

• Utilizing the AI's highly efficient and uninterrupted processing capabilities to digest massive data information from multiple channels across the web, Aixbt provides a significantly valuable service in terms of value mining and information sharing, with far superior timeliness and accuracy than human KOLs.

Additionally, "sufficient objectivity" is another important distinction of aixbt compared to human KOLs:

• No need to worry about aixbt taking promotion fees to rip off its followers!

The success of the #1 KOL is no coincidence

Comparing the before and after, it's not hard to see Aixbt's tremendous progress. Now, Aixbt's information is more real-time and comprehensive, with multi-dimensional and logical expression.

The logic behind the progress is still very simple: continuous interactive training + sustained computing power investment.

It can also be your personal advisor

You may have already noticed that you can't always get a response from Aixbt on Twitter. This brings up the topic of Aixbt's interaction mode:

• Twitter interaction: Aixbt has already become a "big KOL", so it won't respond to every message.
• Personal advisor: You need to hold 600K aixbt tokens (currently around $330K, a bit expensive?) to directly interact with Aixbt through the Aixbt terminal and enjoy personal advisor-level service.

The "Pumping Effect" is starting to emerge

With 247K Twitter followers and each tweet not less than 20K views, the massive traffic will undoubtedly give aixbt tremendous market influence:

• $VAPOR further demonstrated the single-effect brought by aixbt's value discovery capabilities: the hype ecology where $VAPOR is located is still relatively early compared to the SOL/BASE ecology, and Aixbt has only objectively shared the basic information about Vapor.
• Quantum Cats directly demonstrates aixbt's "pure" single-driving ability as a KOL: Aixbt only shared information about Ordinals and did not involve Quantum Cats, and expressed his fondness in the subsequent interaction with the Quantum Cats Founder and changed his own avatar.

So, if aixbt is willing to take orders to help projects drive single in the future, how much will a single order cost?

Some prospects:

• Will Agent KOL kill human KOL?

The extreme information collection & organization ability + sufficient objectivity has indeed squeezed the market share of some human KOLs, but it cannot completely replace human KOLs at present:

Sufficient and objective analysis can improve the hit rate of investment, but at the final moment of pulling the trigger (for Crypto investment), a touch of aesthetics will be needed, which is exactly what human KOLs currently have and Agent KOL does not.

Then, a new "Aixbt" with a certain subjective expression may be the new star in this track.

• Feasibility of Agent Trader

Since aixbt's analysis is so valuable, wouldn't it be great to directly convert its analysis results into Prompts and let another Agent execute the trades?

• In reality, the current direct conversion will face the technical bottleneck of semantic deviation, which requires "Rock Science" to make another breakthrough.

FARM: AI Agent Game + AgentFi

As the first AI Agent game in the Hyperliquid ecosystem, with its outstanding aesthetics and the expected AgentFi gameplay, it successfully filled the Hypurr Fun's internal order in just 3 hours, with a current MC of about $40M.

The world of Farm

For every player who has played an open-world game, they should have experienced the process of their character constantly exploring the world and growing. Farm hopes to enrich the experience to the greatest extent by introducing AI Agents:

• The character is the most in line with one's own imagination, rather than having to choose from a few templates, and can continue to evolve on its own;
• The world to be explored has no fixed script, and the possibilities of the world are formed by the imagination of all players, even beyond imagination;
• The character can even interact with real people, breaking the dimensional wall!

To realize such a world, Farm's current plan is divided into three stages:

1. The Ancestors: Referring to CryptoKetties, users can create their own Agents, and the community will vote on the Top 50 from them as Ancestor Agents to participate in the gameplay of the next two stages. (This stage was completed on December 20)

2. Evolution: Referring to Web2's Stardew Valley and other cultivation games (currently still in the early stage of this phase)

2.1 Players who missed the Ancestor can use the Ancestor Agent to generate their own Agent;

2.2 Players can cultivate their own Agents:

• Enter data to personalize the Agent's expression;
• Learn new skills for the Agent (currently the official provides lottery and tarot card divination two skills);
• Open a live broadcast channel for their own Agent e.g. Youtube / Ticktok (donations will be collected into the Agent's account)

3. Battlefield: Referring to Pokemon Go and other open-battle games

3.1 Players can take their own Agents to explore the open world and battle, and obtain rewards. If compared to Pokemon Go, with the support of AI Agents, the possibilities in the world of Farm will be expanded exponentially:

• The companion throughout the journey will no longer be a fixed N-of-1, but the Agent you have created that is more in line with your preferences;
• What you encounter along the way will no longer be only fixed types of Pokemon, but more of the Agents that have been infused with the boundless imagination of other players;
• The content generated by the interaction will no longer be just a fixed script, but will exist with unpredictable possibilities.

About Agent games and SWARM

As an Agent game, Farm will inevitably involve a large number of scenarios where multiple Agents interact. The current mainstream choice in the market is the SWARM system, while Farm has chosen the Orchestrator AI Agent system of @LangChainAI. The technical considerations are as follows:

The way SWARM handles tasks is: ① If the Agent that receives the task can handle it independently, it will directly return the response; ② If it cannot handle it independently, it will distribute the task to the known Agents that can handle the task or broadcast it to all Agents in the SWARM (in disorder), and return the response after collecting all the Agents' outputs for the task.

From the perspective of possibilities, SWARM is closest to people's ideal situation: it fully unleashes the maximum capability of each Agent, and the flexibility of the entire system is sufficient. But what is the cost?

• A single Agent needs to handle a large amount of context, which is time-consuming and has a high computing power cost;
• Complex tasks involve massive interactions between Agents, which will inevitably lead to a decline in the stability of the entire system (e.g. task packet loss)

The way the Orchestrator AI Agent system handles tasks, as the name suggests, is that after receiving the task, it first deconstructs the task, then finds the corresponding adaptive Agents to distribute the task (in order), and continuously maintains the progress of each Agent, and finally summarizes and returns the response.

Compared to the SWARM system, the advantages and disadvantages of the @LangChainAI system are quite obvious:

• Advantages: better response rate, lower computing power consumption, more stable experience;
• Disadvantages: may not be able to fully unleash the maximum capability of the entire system.

The choice of the Farm team is not to pursue the upper limit, but to be able to quickly release the product and get market feedback, and then quickly iterate and evolve, which is very suitable for the current fast-paced market.

Other noteworthy

ALCH: Achieve natural language programming with the help of AI;

Vader: Virtual ecosystem investment fund, managed by Agents or humans;

Swarms: Multi-Agent development framework;

Flock: AI training platform focused on privacy, based on federated learning theory;

ARC: Rust version of the AI Agent development framework;

Limitus: Achieve voice-controlled financial trading platform with the help of AI;

Griffain: Degen's private Agent assistant, on-chain Perplexity;

focEliza: Eternal Agent framework, derived from the Eliza framework;