Compiled | Odaily (@OdailyChina)
Translator | Azuma (@azuma_eth)
AI is the main theme of this cycle, and many believe it is a track that will continue to exist and evolve. Conversely, there are also many reasonable criticisms that most of the current AI Agents are perfunctory, and it will take 3-5 years for the technology to become more meaningful.
As a preface, the project categories of Crypto x AI span multiple levels. The true technological prospects of Crypto x AI are mostly related to using the Crypto economy to guide better reasoning or provide decentralized access to computation. Delphi's article is a good starting point to understand the entire stack.
However, this article focuses on the current state of Agents. While there are exciting innovations at the lower levels of the stack, Agents seem to have attracted the attention of the mainstream cryptocurrency community. As Crypto x AI Agents continue to evolve, here are 5 trends worth noting.
Trend 1 | Framework + launchpad: Value accumulation is starting to become important, and frameworks may continue to exist
Value accumulation is starting to become important. Why is this the case?
To take a step back, why do people hold assets? There are two main reasons why people exchange dollars for assets:
They can sell the asset to someone else at a higher price due to catching up with the narrative rotation that will bring new buyers;
The asset will generate more cash flow for them over time.
As @izebel_eth wrote in his article "Old Coins Worse, New Coins Better", the only thing that matters is the flow of funds. The following two reasons reflect two bullish flows of funds:
Growth in new holders;
Token sinking (tending towards deflation);
However, in most cases, we have not seen anything resembling true token deflation or value accumulation. Using tokens to power agent terminals (like AIXBT) is more akin to staking than traditional value accumulation.
This is also why frameworks like Virtuals, ai16z, zerebro, and arc have become very popular recently. ai16z was initially primarily an investment DAO, but it has become a leader in such protocols since releasing details on its upcoming launchpad and token value accumulation.
The framework + launchpad space is now very saturated, and the first-mover advantage has driven their success. There are many reasonable doubts about the utility of the launchpad around these, as many agents themselves are useless. However, those frameworks (eliza V2 + launchpad, zentients, arc and its handshake program) have not yet launched their main products. If they succeed in attracting developers and users, they may continue to lead the entire track.
Why will frameworks continue to exist?
Regardless of whether the Agents have real value, the frameworks used to launch the Agents will perform well, as they still have "product-market fit" (PMF) with speculation. Frameworks + launchpads allow users to have both a factory and a casino. To some extent, Virtuals has replaced pump.fun's position in the Base ecosystem.
More optimistically, as technology progresses, leading frameworks may launch more advanced Agents, and the development of open-source software libraries like Eliza may accelerate. Many of these launch platforms will also become the coordination layer for communication between groups and Agents, using their tokens for some form of value transfer. For more in-depth articles on Eliza value capture, such as the one Teng Yan wrote last week here.
Virtuals, ai16z, Zerebro, and arc are currently the main players in this track, but the launchpad space is becoming increasingly saturated. The frameworks with the fastest iteration speed, strongest scalability, and most unique functionality are worth watching.
Trend 2 | The next generation of Agents will prioritize utility and value accumulation, and DeFAI (DeFi x AI) may be the first class of Agents to achieve product-market fit
Most Agents only have meme tokens without utility. To grow the Agent industry, the next generation of Agents must further improve use cases. New opportunities will come from Agents that can truly accumulate value and take action. I believe that within the next year, first-tier Agents will leap to third-tier Agents.
We will see this first in the DeFi domain. We will see more terminals that allow people to express desired outcomes in natural language or voice, and Agents that can execute tasks behind the scenes. Existing wallets and protocols will also integrate Agents to improve the user experience.
Representative projects include Wayfinder, griffain, Hey Anon, Limitus, and neur.
You can learn more about this from 0xJeff's article.
Trend 3 | Consumer layer: Entertaining Agents, autonomous worlds, and the revival of games
Attention-driven Agents will develop more complex personalities and multimedia interactivity. This may trigger the next wave of game and metaverse technology.
One problem with existing Agents is that they are starting to become increasingly similar, becoming commoditized and exaggerated chatbots. While some Agent projects will pivot to infrastructure (many already are), some are starting to think about useful products/applications, while others may continue down the attention token path. However, the next generation of attention Agents will be better, developing more complex personalities and providing greater interactive possibilities. This can be achieved through audiovisual forms of expression, or by endowing Agents with a three-dimensional presence and physical embodiment.
There have been early attempts in this area, with Jeffy writing an article about implanting a body into Zerebro, and some high-profile KOL Agents directly launching through audiovisual forms like short videos. Slopfather and Ropirito are two early Agents that have adopted a video mode.
However, I believe more Agent projects will incorporate these features to make Agents more lively. The product-market fit (PMF) in this domain will be achieved through the consumer layer, where individual users may want to create personalized Agent companions or interact with KOL Agents. This is already gradually spreading beyond cryptocurrencies, with some users conversing with AI companions for hundreds of hours on sites like character.ai, which Google recently acquired for $2.7 billion.
Furthermore, these 3D Agents may find product-market fit in existing consumer layers, particularly games and the metaverse. Agents can add deeper backstories to the worlds of stories. Imagine having an Agent act as an NPC that can independently complete tasks or play games, with constantly evolving memories and personalities. True autonomous game worlds may be realized through Agents.
Representative projects include:
Soulgraph aims to provide tools to give Agents more customized personalities and memories;
Holoworld AI has established Web2 partnerships with companies like L'Oreal, Bilibili, and Fox to create digital avatars.
There are also representative projects in the areas of games, metaverse, virtual worlds, and autonomous Agents:
Hyperfy is a metaverse platform that allows anyone to create virtual worlds and uses the Eliza plugin;
Parallel Colony is an AI simulation game where the Agents are the players, and you can converse with them;
Digimon is a pet-like creature game where the creatures evolve spiritually based on interactions.
SMOL also uses the Eliza framework, and the characters in the game are Agent driven by LLM, who can take actions in the virtual world based on your instructions.
Trend 4 | Agentic Organizations: The Return of DAOs
Decentralized Agentic Organizations are the next evolution of DAOs. Swarm or multi-Agent systems are exciting because they can coordinate and execute more advanced strategies, similar to running a company. A heterogeneous swarm composed of many different types of specialized Agents/models may perform better than a single large model.
While fully autonomous Agents and swarms may still be far away, the next iteration of DAOs is likely to be about the interaction between humans and collectives. This will reduce the inefficiency of bureaucracy and lower the cost and time of human execution. In the context of capital flows, the next step for income-earning Agents is to become full-fledged income-earning organizations.
Representative projects include:
Agent swarm infrastructure: Projects like SwarmNode.ai and FXN are specialized infrastructure projects focused on multi-Agent frameworks and coordination, while more mainstream Agent proxy frameworks like Zerebro and ai16z have also expressed interest in building at this layer.
DAO launchpads: We've seen the majority of early traffic concentrate on investment-focused DAOs. I believe daos.fun will be the first major DAO launchpad, which incubated ai16z. There are also some newer launch platforms like daos.world, where AI-driven funds like 3BC are gaining attention.
Trend 5 | Verifiable Agents: Current Agents will evolve towards greater autonomy and true ownership of their own liquidity
Currently, most Agents require a high degree of human intervention. The next wave of Agents will evolve towards true autonomy, starting with the ability to manage their own capital.
The intersection of Agents and cryptocurrencies lies in cryptocurrencies providing a financial path for truly economic Agents. However, most Agents do not control their own treasuries, or have their treasuries managed by human teams. To achieve truly economic Agents, Agents must be able to autonomously manage their own capital. This can allow Agent behavior to start evolving, as you can impose economic constraints on Agents, forcing them to pay for their own reasoning costs, introducing Darwinism into the Agent world where they must earn income to survive.
Representative projects include:
Freysa is one of the early Agents that controls its own capital and has performed well (including winning Musk's attention). They recently announced they are building a framework to give Agents verifiable autonomy within a TEE + Agent-controlled key system;
Lit Protocol also has an Agent framework that allows autonomous Agents to conduct on-chain transactions through a system of private key storage and execution;
Galadriel has launched a "Proof of Sentience" SDK to allow developers to comprehensively verify Agents on-chain.
Conclusion: Before we see 1 billion people on-chain, we will see 1 billion Agents on-chain
The user experience of cryptocurrencies is already not user-friendly for human users. But Agents won't mind this friction. We will start with human-Agent interaction, but the trajectory of crypto-AI is Agent-Agent interaction, where swarms of autonomous Agents will interact and transact on-chain, managing their own economic status.
To empower Agents with economic agency, allowing them to incentivize behavior (pay service fees) and coordinate real-world activities, they need the ability to control and deploy capital. Cryptocurrencies are the "home planet" for these Agents - blockchains will enable Agents to participate in permissionless financial activities. Stablecoins and high-performance Layer 1s are the ideal tools to achieve cost-effective, 24/7, global transactions.
Beyond the current hype and narratives, we have ample reason to be excited about the long-term prospects of on-chain Agent economies. Many real use cases, including DAOs and income-generating Agents, are much closer than we imagine.
