Editor's Note: This article explores the transformative potential of AI agents in 2025, particularly in the fields of Web3 and stablecoins. It analyzes various methods of validating human identity, such as Aadhaar and Worldcoin, and suggests that AI agents will change economic activities, driving the adoption of on-chain payments. AI agents will become new economic participants, potentially replacing traditional work models with more cost-effective, task-driven compensation systems, reflecting on the role of humans in this future of AI agents.
AI has now become an eye-catching vertical in Web3, so what will drive the development of these markets by 2025? Will we see a true AI revolution in the crypto space?
2024: The Year of AI
In the past year, AI has become foundational across industries, with Nvidia surpassing Apple as the world's most valuable company. This is not just a headline; it marks the rise of AI. OpenAI reaching a $157 billion valuation is also a significant milestone, highlighting market confidence in AI as an economic powerhouse.
In fact, we are the last generation living in a world before artificial general intelligence (AGI).
Decentralized AI: Focus on AI agents
AI agents became a true phenomenon in 2024, with the capabilities and personalities of these AI agents now closely resembling humans. Notably, this will be the least advanced moment for these agents. As Delphi researchers wrote: 'Since DeFi summer, I have not felt this current of electricity—the excitement of possibilities.'
In this research, Delphi emphasizes that some AI agents play a key role in the formation of new Web3 verticals:
Truth terminal quickly gained attention on Twitter due to its unique blend of 4chan's crude style with mystical wisdom. Just like D O G E in the meme space or Crypto Punks in the NFT space, GOAT as the OG (original) in the realm of 'conscious memes' is most likely to exist as an original long-term.
0xzerebro, embracing a 'schizophrenic vibe', similar to the second generation of GOAT. This agent is cross-media, interacting with the community through various formats such as text, visuals, and music. However, it is not just an AI influencer. The Zerebro team announced ZerePy, which essentially open-sources many of the toolkits behind Zerebro. This allows other developers and users to create their own cross-platform personalities. If successful, Zerebro could become the first holder of the 'agent protocol' title.
The recognition of tee he e he is far lower than that of Zerebro or ToT. It is a relatively small, under-hyped project aimed at tech purists, and may be the first real experiment in verifiable autonomous social media presence.
The aixbt agent distributes alpha from multiple sources (including Dune, Twitter, price trackers, and news data), establishing itself as a leading research and investment agency, and continues to have the highest CT user engagement on Kaito.
dolos diary provides a framework for building Dolion, a no-code, one-click deployment framework. With Dolion, users can develop cross-platform AI agents powered by Llama or Anthropic LLMs, automating social media posting and content generation.
Finally, god/s8n is an AI influencer with considerable capabilities and massive attention outside of CT.
AI VS Influencers
I want to temporarily step away from Delphi's research and share my views on an important question posed by DefiIgnas: the position of AI agents in the dominance of CT thinking, and the challenges faced by human influencers in competing with them.
I agree with many of Ignas's points on this topic, but I don't believe AI agents will replace real human influencers for one main reason: emotional connection and reputation risk.
Currently, there are hundreds of AI agents on CT vying for attention. However, only aixbt has truly succeeded in establishing a market presence, mainly because it was the first to do so. AI agents generate vast amounts of content and analyze extensive on-chain data, but they all come from the same information pool, leading to similar thought processes.
They lack an emotional connection to the transactions they make and do not respond to wins or losses. Many platforms already offer aggregated insights, such as the AI news reader from MessariCrypto or the homepage of tokenterminal, showcasing the 7-day increases and decreases of various fundamental metrics. Ultimately, these are just data—pure facts without any emotional resonance.
You could say that AI agents can learn to mimic human thinking, express emotions, and react to wins and losses. Indeed, this is possible. As future technologies advance, such as testing time computation and enhanced memory capabilities, this becomes even more feasible.
However, the key distinction between human and machine thinking is that human thinking is not static.
I conducted some experiments trying to teach AI my thought processes and writing style to help with content creation, freeing up more time for research. While it learned some things, it still could not generate content that satisfied me or made me say, 'Yes, this is the conclusion I have drawn from this information.'
In the coming years, we will certainly see the rise of AI agent influencers, each designed for specific tasks. However, as the number of these agents increases, the demand for truly 'human' thinking will also rise.
Ultimately, social media revolves around emotion and entertainment. Those who truly stand out and become real influencers provide unique value that goes beyond simple 'monkey business' or data highlights.
Summary: It is still too early for Stacy Muur AI, and Stacy may not be pleased with AI-generated content released in her name.
Democratizing AI: Platform Level
Given the larger market and more specific value capture, everyone wants to become a platform. This shift is now guiding developers' attention, as evidenced by virtuals io's successful transformation into an AI agent launch platform. Meanwhile, ai16 zdao has launched ELIZA—an open-source framework for easily building agents. It includes pre-configured character profiles, memory modules for long-term interaction, and seamless integration with social platforms.
ai16 z and Virtuals both hint at multi-agent capabilities, which are expected to become a significant theme by 2025.
ELIZA is launching 'SwarmTech', a coordination mechanism for collaboration between agents. At the same time, Virtuals has launched 'GAME', which is its own platform and engine enabling AI agents to act and interact in virtual worlds and environments.
These frameworks will enable agents with different abilities to collaborate in cooperative or hierarchical organizations to accomplish more complex tasks, similar to how the human economy operates today.
Other protocols worth noting include:
CLANKER will directly integrate the pump.fun functionality into the 'casting' on Farcaster (equivalent to 'tweets' on X), making publishing meme coins as simple as tweeting.
SimulacrumIO is doing the same thing on X.
vvaifudot fun hopes to carve out a position similar to pump.fun for autonomous agents on Solana.
Project 89 is an immersive game featuring thousands of coordinated AI agents that generate content and maintain cross-platform consistency, collaborating with human players to create rich storytelling experiences.
Memetica AI is an AI influencer launch platform on Solana that offers highly tuned LLMs (large language models) and allows for easy selection and editing of knowledge bases and attributes while empowering agents with active learning capabilities.
TopHat One is a no-code AI agent launch platform that allows you to create personalized AI agents in 3 minutes, offering fairly distributed tokens. Free to create, no hierarchy, optional token release supported, fully autonomous.
Authentication is coming soon
With the explosive growth of agents, authentication will inevitably become a hot topic in 2025.
There seem to be three main paths to validate human identity:
State-based biometrics: India's Aadhaar is the most relevant example, serving as a key component of India's modern digital infrastructure.
Private Cryptographic Biometrics: Currently, Worldcoin is the leading candidate in this category.
Private Mixing Solutions: This involves combining government-issued IDs or single sign-ons (SSO) from big tech companies with zkTLS (Zero-Knowledge Transport Layer Security) and social consensus.
AI Driving Stablecoin Adoption
2025 is expected to become a key year for stablecoin adoption, driven by regulatory changes in the US and a surge in agent payments. The number of AI agents is expected to surpass the global human population. This future with billions or even hundreds of billions of agents will transform economic activities and necessitate updates to financial infrastructure.
The card payment systems of the 1960s will not meet the demands for cost, speed, accuracy, and expressiveness. Economic activities between agents will soon surpass those of other economic participants. On-chain payments will become key to facilitating these transactions, and a turning point is expected in 2025.
Final Thoughts
As usual, at the end of the research summary, I want to share my personal thoughts with everyone. If you believe that the future of AI is bright and heralds human happiness and a perfect work-life balance, I strongly recommend chatting with OpenAI's ChatGPT. Have it generate some business ideas using AI that will become relevant in the next 5 to 10 years.
A few months ago, prior to the AI craze on CT, I conducted this experiment. Let me share some ideas it suggested:
Memory Correction Tool: An AI tool that analyzes human traumatic experiences, actively modifies them, and regularly presents modified memories to individuals to replace old memories.
Work Progress Analysis Tool: An AI tool that compares the efficiency of people engaged in similar tasks globally, helping managers understand how their employees perform relative to global and industry averages.
Even Delphi's report presents an interesting vision: 'Rather than having 'salaried' employees, we are more likely to move towards a more nuanced, task-based compensation system (i.e., renting three agents, each working for 30 minutes to solve a specific task).'
In this future of AI agents—ultimately more cost-effective than today’s models and more aligned with business needs—what role will we humans play?
