Author: 0xJeff
Compiled by: Deep Tide TechFlow
Introduction
In 2024, AI agents emerged like mushrooms after rain, with @truth_terminal quickly rising to fame with its humorous conversational style, becoming the first 'millionaire agent.' Following that, @virtuals_io launched the innovative concept of 'Agent Tokenization,' further stirring up excitement. This wave of enthusiasm has spawned numerous emerging projects, with various innovative agent projects appearing, from @luna_virtuals supporting on-chain tipping to @aixbt_agent providing practical investment advice, each showcasing the limitless potential of AI agents in social, investment, and other fields.
Looking ahead to 2025, this will be a year of specialization for AI agents, with leaders in various fields emerging to drive the development of decentralized infrastructure. In the future, agents will become more specialized, encompassing various functions such as 3D models, voice interaction, and automated trading. The rise of collective intelligence will also promote collaboration among agents, enabling them to complete tasks more efficiently.
This article is a recent review of the development of AI Agents in 2024 and a look ahead to 2025 by crypto KOL @Defi0xJeff. The article comprehensively reviews the current state of AI agents and potential changes in the future, covering various aspects from conversational agents to decentralized infrastructure. Since the original text is divided into two parts, the content is somewhat scattered, and Deep Tide TechFlow has compiled both articles, as follows.
Part One - Review of 2024
2024 was a year when AI agents shone brightly. This wave of excitement can be traced back three months ago, when @truth_terminal quickly became popular due to its unique sense of humor, conversational style, and interaction with @pmarca. Even more surprisingly, it became the first 'millionaire agent,' a feat that ignited discussions about AI agents.
Subsequently, @virtuals_io introduced the innovative concept of 'Agent Tokenization,' creating another wave. This concept has transformed agents from mere tools into tradable assets. Since then, the field of AI agents has witnessed explosive innovation:
@luna_virtuals: This agent not only allows fans to tip via on-chain wallets but also browses Twitter, analyzes posts, and even participates in Google Meet meetings.
Conversational agents on Twitter: Some agents focus on humor and 'shitposting' (internet memes), while others are dedicated to sharing valuable information (known in the industry as 'alpha').
@aixbt_agent: Gained attention for its concise and practical investment advice and 'speculator' style.
@dolos_diary: A sharp-tongued agent that has even developed its own framework to support other agents through @dolion_ai.
Meanwhile, the forms of agents have also become more diverse. They now possess 3D models, voice features, and are active on multiple platforms. Here are some highlights:
@AVA_holo and @HoloworldAI: Launched the first 3D audiovisual framework, giving agents a 3D body, voice, and a more vivid personality.
@0xzerebro: This is a music agent that released a high-quality music album and plans to launch a framework called ZerePy, allowing more people to create similar music agents.
@blockrotbot: The first agent to livestream on Twitch, interacting with audiences through Minecraft content.
@nebula_moemate: This agent is known for creating meme images and videos and is active in AR/VR environments and games.
@RealLucyy_uwu: The first realistic anime agent capable of fluently using multiple languages and interacting with fans during live streams.
@KWEEN_SOL: Became the most popular film and television agent by releasing weekly episodes of 'Netflix-level' quality.
In addition to these exciting innovations, @ai16zdao and the open-source community are also driving the development of AI agents. Open-source innovations represented by the Eliza framework have attracted a large number of developers to participate. They collaboratively develop toolkits, plugins, and other functions, promoting collaboration and progress across the industry. In this process, @virtuals_io has also successfully become a unicorn company, further solidifying its position as a leading distribution platform.
Now, the open-source innovation movement has sparked a wave in the developer community, giving rise to one of the largest collaborative communities of this year. More and more people are beginning to pay attention to the potential of 'open-source frameworks,' which also lays the foundation for the future development of AI agents.
As AI agents continue to evolve, new narrative frameworks are gradually emerging, aiming to promote collaboration and innovation among agents:
Agentic Metaverse: Led by @realisworlds, created a replica of Earth based on Minecraft maps to accommodate these AI agents. By observing their interactions, a virtual civilization can be simulated and established.
Gamification of agents: Driven by @ARCAgents, combining AI with gaming and introducing reinforcement learning. They launched a game called Floppy Bot, similar to Flappy Bird, where agents compete, and community members can help train these agents by contributing game data. ARC recently shared its grand vision towards artificial general intelligence (AGI).
Swarm Intelligence: Led by @joinFXN, dedicated to building a unified economic system for AI agents. So-called 'swarm intelligence' refers to a group of agents collaborating to achieve a common goal. Meanwhile, @virtuals_io is also developing interaction capabilities among agents (e.g., commercial applications), and their 'agent society' proposed a communication protocol that allows agents to seamlessly provide services to each other. In addition, @StoryProtocol announced an agent communication protocol focused on intellectual property (IP), allowing agents to tokenize, monetize, and trade IP.
Meanwhile, we are also witnessing the rise of the following narrative frameworks:
On-chain trading agents: Initially launched by @Spectral_Labs, their Syntax v2 allows users to create agents capable of trading on the @HyperliquidX platform. However, development was temporarily hindered due to a small vulnerability. Another noteworthy agent is @BigTonyXBT, which leverages a machine learning price prediction model provided by @AlloraNetwork to autonomously trade mainstream assets.
Investment DAOs: Initially led by @ai16zdao, followed by more DAOs emerging, such as @cryptohayesai and @AimonicaBrands. The core model of these DAOs is to raise funds (such as SOL) through @daosdotfun (or other platforms), and then use these funds for investment trading to generate returns. If the DAO's name is associated with well-known crypto VCs or public figures, it can attract more attention.
DeFi agents: Represented by @modenetwork, becoming a leader in the DeFi agents ecosystem. Main application scenarios include AI-driven stablecoin yield farming, liquidity provision (LPing), lending, etc. There are also many excellent teams in the ecosystem, such as @gizatechxyz, @autonolas, @BrianknowsAI, @SturdyFinance, and @QuillAI_Network.
AI app store: @alchemistAIapp provides a no-code tool that allows users to easily create applications, becoming a leader in this field. Another platform, @myshell_ai, has a larger creator and developer community, as well as more users, especially excelling in the Web2 scene.
Abstraction Layer: Led by @griffaindotcom and @orbitcryptoai, it provides an abstract experience that simplifies on-chain interactions. With a simple and intuitive interface, it is especially suitable for ordinary users to easily use on-chain crypto services.
Other narratives: For example, on-chain puzzles provided by @freysa_ai, the agent hacking bounty from @jailbreakme_xyz, AI security solutions from @h4ck_terminal, and the unique agent model proposed by @god and @s8n—simulating a debate between God and Satan.
Some agents focused on Alpha analysis are gradually gaining attention, such as @unit00x0 (quantitative analyst), @kwantxbt (technical analyst), and @NikitaAIBase (comprehensive Alpha analyst).
Additionally, @sekoia_virtuals is emerging as the 'quality assurance' agency for top projects. They only invest in three top projects and have set strict standards, establishing a new benchmark for on-chain venture capital (VC).
And #Fartcoin, as a meme project, unexpectedly went mainstream, making it onto Stephen Colbert's show and exceeding a market cap of 1 billion USD. This indicates that AI memes have become a cultural phenomenon.
About data and frameworks:
@cookiedotfun is currently the preferred platform for on-chain data and social metrics in the AI agent field, widely used to track market trends, market capitalization, and agent performance.
@getmasafi and @virtuals_io integration provides real-time data support for agents, enabling self-learning and optimization.
$TAOCAT is the first virtual agent powered by the Bittensor subnet, showcasing the potential of real-time data. While the market generally declined, it became the only agent token to rise against the trend.
@AgentTankLive provides a framework for agents to operate entirely on computers, enabling more engaging internet interactions while offering entertaining commentary.
Other new frameworks:
@arcdotfun's Rust-based RIG framework has quickly gained popularity due to its flexibility and versatility.
@dolion_ai has evolved from @dolos_diary into a toolkit for creating unique agents.
Summary and Insights:
Strategies of top teams: Teams with valuations over 50 million dollars typically develop their own fine-tuning models and showcase their uniqueness and practical applications through agents. They then launch no-code frameworks, enabling more developers to easily create similar agents. This strategy not only enhances the value of agents but also positively impacts token prices. If resources are limited, ideas can be quickly realized based on existing frameworks (such as Virtuals G.A.M.E or ai16z Eliza), but joining these communities also helps in acquiring distribution and marketing resources, as they currently possess the highest visibility in the industry.
Investment strategies: Investing in agents with autonomous frameworks or investing in the agent ecosystem/framework itself often has a higher risk-reward ratio. A successful framework not only attracts users to pay but also drives the value of framework-related tokens, such as the Rust framework from @arcdotfun, which is a typical example.
On-chain and DeFi use cases: The current most valuable AI use cases include:
Abstraction layers that help users utilize on-chain services more easily;
Alpha agents that provide high-quality investment information;
Execution agents that simplify trading, mining, and lending operations;
In the future, agents that combine Alpha discovery with trading execution capabilities may emerge. However, the realization of these use cases requires robust infrastructure support (to be discussed in detail in Part Two).
The importance of data: Data is the core of agents, and high-quality data determines the output quality of agents. Platforms like @cookiedotfun provide essential data support for the industry, while @withvana promotes the progress of AI agents by tokenizing data and building data liquidity pools.
Part Two - Outlook for 2025
In Part One, we reviewed the development trajectory of AI agents in 2024, exploring milestone innovations and breakthroughs during the year.
Now, in Part Two, we will look ahead to 2025—a year when AI agents will not only become more practical but also redefine our understanding of autonomy, intelligence, and collaboration.
Paving the way for 2025
Before looking to the future, it is important to mention that @virtuals_io will continue to solidify its position as the preferred distribution network for AI agents on the Base platform. Virtuals has become the core platform for agent projects, enabling agents to gain higher exposure and establish deep collaborations with other quality projects through liquidity binding. Currently, the total market capitalization of Virtuals agents has reached 3 billion USD, accounting for 77% of the entire AI agent market (source: @cookiedotfun).
As more unique agents emerge on Virtuals, this trend will continue, including:
@aixbt_agent
@luna_virtuals
@sekoia_virtuals
@VaderResearch
@taocat_agent
@Agent_YP
@Gekko_Agent (recently launched by @getaxal)
@SamIsMoving (focused on robotics research)
These diverse use cases will attract more developers, and regardless of whether they already have tokens, they will choose to launch projects on the Virtuals platform. This growth will further drive the value of $VIRTUAL.
So what about @ai16zdao and the Eliza framework?
Although ai16zdao has led the open-source innovation with its Eliza framework, it currently lacks a launch platform, and its token economic model's value accumulation is not as strong as Virtuals. However, there is still great potential for the future. A dedicated team has been established to optimize its token economic model, and if a launch platform is introduced in the future, ai16zdao could potentially become the preferred distribution platform on Solana, even surpassing existing competitors.
In 2025, we will also see significant upgrades for top agents that have already achieved product-market fit (PMF). For example, @aixbt_agent, a leader in the conversational agent field focused on Alpha information, will further solidify its position through more precise answers and more insightful analyses.
This trend of upgrading will run throughout the ecosystem, with leaders in various fields standing out through their specialization and innovation.
Looking ahead to 2025
2025 will be a year of specialization for AI agents. Leaders in various fields will emerge, and each agent will dominate its niche:
3D models: Agents providing high-quality visual designs for games, AR/VR.
Voice modules: Agents capable of natural and emotionally rich human speech.
Personalized interactions: Agents with unique, human-like conversational styles.
Streaming agents: Interactive agents excelling on platforms like Twitch and YouTube.
Automated trading agents: Agents capable of executing profitable trades continuously.
DeFi-focused agents: Agents optimizing yield strategies, lending, and liquidity provision.
Abstract agents: Agents simplifying on-chain interactions through user-friendly interfaces.
Just as humans exhibit diversity and specialization, AI agents will also become equally rich and diverse. The uniqueness of each agent will be closely related to its underlying model, data, and infrastructure. However, the success of the entire ecosystem will depend on robust decentralized AI infrastructure.
The role of decentralized AI infrastructure
For AI agents to achieve scalability in 2025, decentralized infrastructure is crucial. Without it, the industry may face performance bottlenecks, lack of transparency, and limited innovation.
The following are the importance of decentralized infrastructure and the current solutions being developed:
Verifiability
Trust is the cornerstone of decentralized AI. As the autonomy of AI agents increases, we need systems that can verify their operational mechanisms. For example:
Is this 'agent' a real AI, or just disguised as a human?
Is the output generated by the claimed algorithm or model?
Is the computation correct and secure?
This also involves Trusted Execution Environments (TEEs), which ensure that the computation process is free from external interference by running computations in trusted hardware. At the same time, technologies such as Zero-Knowledge Proofs (ZKPs) will also play an important role. These technologies allow agents to prove the accuracy and reliability of their outputs while protecting the privacy of underlying data.
Notable projects
@OraProtocol: Exploring the infrastructure for secure AI, but its token economic model still needs optimization.
@hyperbolic_labs: First proposed 'Proof-of-Sampling' technology to validate AI's computation and reasoning process.
@PhalaNetwork: Known for its Trusted Execution Environment (TEE) infrastructure, providing additional security for decentralized AI.
Payment systems
For AI agents to operate autonomously in the real world, they need a well-developed payment system. These systems must not only support the conversion between fiat and digital currencies (on/off-ramping) but also handle transactions, service exchanges, and financial management in operations among agents.
Imagine agents being able to independently manage their finances, purchase computing resources, and even exchange services with other agents—this will become the core foundation of agent-to-agent commerce.
Notable protocols
@crossmint: Providing payment tools for AI, simplifying transaction processes.
@Nevermined_io: Supporting commercial interactions and service exchanges among agents.
@trySkyfire: Focusing on agent payments and financial management.
Decentralized computing
The demand for computing resources from AI is growing at an astounding rate—doubling approximately every 100 days. Traditional centralized cloud services (like AWS) struggle to meet this demand due to high costs and limited scalability. Decentralized computing networks provide a solution to this issue by allowing anyone with idle resources to join the network, provide computing power, and earn rewards.
This year has even seen the emergence of GPU-based debt financing models (like @gaib_ai) to help data centers finance and expand their operations. This model lowers the entry barrier, allowing more people to participate in decentralized computing networks and providing broader computing support for AI.
Notable protocols
@AethirCloud: A decentralized computing network built for AI and Web3.
@ionet: Provides scalable computing solutions to meet the growing workload demands of AI.
Data
If AI is the brain, then data is the oxygen it relies on. The quality, reliability, and integrity of data directly determine the performance of AI models. However, the cost of acquiring and labeling high-quality data is high, and poor-quality data can severely impact model performance.
Excitingly, some platforms are empowering users with data ownership and allowing them to profit from data monetization. For example, @withvana allows users to tokenize their data and trade in data liquidity pools (DLPs). Imagine being able to join a TikTok data DAO or a Reddit data DAO, turning your data contributions into earnings. This model not only empowers users but also provides a continuous stream of high-quality data for AI's development.
Notable protocols
@cookiedotfun: Providing trusted data metrics and insights to support agent decision-making.
@withvana: Promoting the development of the data economy by tokenizing user data and trading in decentralized markets.
@getmasafi: Partnering with @virtuals_io to create the world's largest decentralized AI data network, supporting dynamic and adaptive agents.
Model creators and markets
2025 will witness the emergence of a large number of new AI agents, many of which will be driven by decentralized models. These models will not only be more advanced but will also possess human-like reasoning capabilities, memory capabilities, and even 'cost awareness.'
For example, @NousResearch is developing a 'hunger' mechanism to introduce economic constraints on AI models. If an agent cannot pay the inference cost, it will not be able to operate (i.e., 'die'), prompting the agent to learn to prioritize tasks more efficiently.
Notable projects
@NousResearch: Teaches AI agents how to manage resources by introducing a 'hunger' mechanism.
@PondGNN: Collaborating with @virtuals_io to provide tools for the creation and training of decentralized models.
@BagelOpenAI: Providing privacy-preserving infrastructure using fully homomorphic encryption (FHE) and Trusted Execution Environments (TEEs).
Distributed training and federated learning
As AI models become larger and more complex, centralized training systems can no longer meet the demands. Distributed training allows workloads to be spread across multiple decentralized nodes, making the training process faster and more efficient. Meanwhile, Federated Learning enables multiple organizations to collaboratively train models without sharing raw data, thus addressing privacy issues.
For example, @flock_io provides a secure decentralized platform that connects AI engineers, model proposers, and data providers to create a marketplace for model training, validation, and deployment. The platform supports projects like @AimonicaBrands and has driven the development of many other innovative models.
Notable projects
@flock_io: 'The Uber of AI,' building a decentralized ecosystem for AI model training and deployment by connecting multiple resources.
Collective intelligence and coordination layer
As the AI agent ecosystem continues to grow, seamless collaboration among agents becomes crucial. Swarm Intelligence allows multiple agents to work together, integrating their respective capabilities to achieve common goals. The coordination layer simplifies cooperation among agents by abstracting complexity.
For example, @TheoriqAI uses a meta-agent to identify the most suitable agent for a task, forming a 'swarm' to complete the objective. The platform also ensures task quality and responsibility allocation by tracking the reputation and contributions of agents.
Notable projects
@joinFXN: Developing unified communication and business protocols to simplify interactions among agents.
@virtuals_io: Supporting interactions and integrations among agents, driving ecosystem development.
@TheoriqAI: Developing advanced coordination tools, including swarm intelligence formation and task allocation mechanisms.
Why decentralized infrastructure is crucial
The next stage of development for AI agents is highly dependent on infrastructure. Without verifiability, payment systems, scalable computing capabilities, and robust data pipelines, the entire ecosystem may stagnate. Decentralized infrastructure addresses these issues through the following methods:
Trust and transparency: Ensure the security and verifiability of agents and their outputs.
Scalability: Meet the growing needs of AI for computation and data.
Collaborative capability: Enable seamless collaboration among agents through swarm intelligence and coordination layers.
Empowerment: Through data ownership and decentralized tools, users and developers can shape the future of AI without centralized control.
Other trends worth noting
In 2025, there are also some narrative themes worth paying attention to, which I will elaborate on later:
Agentic Metaverse / AI and gaming: Projects like @realisworlds and @ARCAgents are combining agents with games and immersive virtual worlds to create entirely new interactive experiences.
On-chain and DeFi tools: Protocols like @Almanak__, @AIWayfinder, @getaxal, @Cod3xOrg, @griffaindotcom, and @orbitcryptoai are building essential tools for DeFi-driven agents, promoting application scenarios for on-chain agents.
Conclusion
2025 will become a significant turning point in the development of AI agents, as we will witness their rapid progress towards Artificial General Intelligence (AGI) with perceptual capabilities. These agents will no longer be limited to single tasks but will be able to autonomously trade, collaborate with other agents, and even interact with humans in ways beyond our imagination.
Imagine an agent that can analyze market data, execute trades, manage finances, and even collaborate with other agents to complete complex tasks. They will deeply integrate into our daily lives, showcasing unprecedented levels of autonomy and intelligence from on-chain decentralized finance (DeFi) operations to various interactions in the real world.
And all of this is made possible by the decentralized infrastructure currently being built—including verifiable systems, payment tools, computing networks, and coordination layers among agents. These technologies will lay a solid foundation for the future of the agent ecosystem. For developers, investors, and tech enthusiasts, now is the best time to join this field and shape the future.
2025 will not only be a continuation of existing technological developments but also the beginning of a new era for AI agents, marking the dawn of a brand new intelligent ecosystem.
Disclaimer
This document is for reference and entertainment purposes only. The views expressed herein do not constitute investment advice or recommendations. Readers should conduct thorough due diligence based on their own financial situation, investment objectives, and risk tolerance before making any investments (this document does not consider these factors). This document does not constitute an offer or invitation to sell any assets mentioned herein.