Coinbase: Understand the current development status, risks and future of artificial intelligence agents

Author: David Han, David Duong, Coinbase; Translated by: Deng Tong, Golden Finance

summary

• Agent AI has become one of the most promising narratives in the field of AI x encryption, but due to the rapid development of technology and the massive spread of agents, the field is still in its infancy and difficult to navigate.

• Investors’ interest in AI agents is mainly reflected in two aspects—the core infrastructure of launching and hosting agents, and the single agent itself.

• We believe that due to the continuous enrichment of functions in the field of proxy artificial intelligence in the field of encryption, there is huge room for growth in the future. However, at least for now, expectations for AI agents may outweigh the development of the technology itself.

AI agents have also become a revolutionary theme not only in the field of encryption, but also in the wider technology field. The concept of autonomous entities that can analyze market news sources or other external data and make real-time decisions has attracted the imagination of many institutional investors. Elsewhere, some tech industry leaders believe that proxy AI could eventually replace the vast software-as-a-service (SaaS) industry. That is, a trained AI agent can in principle perform any task involving a digital interface.

We believe that the encryption track can play a core role in the future and become the main value transfer mechanism for AI agents. The inherent programmatic nature of encryption and permissionless designs support the expansion and deployment of almost unlimited numbers of such agents, enabling a wider range of use cases from managing on-chain portfolios to paying offline services. Given this potential opportunity, the market cap of proxy AI crypto assets peaked at more than $20 billion in early January, and then approached $8 billion at the end of the month.

The recent decline may indicate that expectations for AI agents have begun to outweigh the actual development of the technology itself. While we expect this topic to change the crypto-x AI landscape in the long run, its short-term applications are limited by the challenges of integration and proxy differentiation and the uncertain long-term utility of proxy tokens. In other words, the full potential of realizing AI agents may be longer than many people think.

Understanding the Differences: Agents vs. Models

Many popular AI tools, including chat interfaces such as ChatGPT or image generators such as Stable Diffusion, are wrappers for generative AI models. They are defined by a bounded set of inputs and outputs, usually in the form of text, audio, and images. AI agents extend the direct functionality of these models by introducing a new class of applications that represent a “combination of reasoning, logic, and access to external information” (as defined by Google).

Specifically, AI agents can access and interact with a wider range of data and tools, allowing them to drive more complex behaviors, from searching multiple databases to planning trips and booking flights. With on-chain wallet integration, the scope of AI agent activity has expanded greatly by incorporating payment services into its toolkit.

Crucially, AI agents can also leverage their reasoning capabilities to act autonomously in dynamic environments. The triggers of AI agents are not limited to manual user prompts—they can be concurrent based on a variety of data streams, including X (formerly known as Twitter) posts or Twitch chats. Likewise, their responses can include multi-step output, such as placing an order, making payments, and sending confirmations to the parties involved.

Agents are typically made by (1) the core LLM model as their inference engine, (2) short-term and long-term memory components, (3) potential roles or personality frameworks, and (4) most importantly, to be able to access more through the Application Programming Interface (API). Comprised of the capabilities of a wide range of Internet and other tools. Therefore, decisions made by agents can directly affect the real world.

Artificial intelligence agents are ushering in a big explosion

Infrastructure and tools around artificial intelligence agents are rapidly evolving, becoming one of the most watched technology trends over the past year. Multiple developer frameworks for building AI agents (including but not limited to CrewAI, LangGraph, AutoGen, phidata, Atomic Agents, AgentGPT and AutoGPT) are competing for market adoption, with the top 15 code repositories on Github in January 2025 They are also related to artificial intelligence.

The attention and excitement about artificial intelligence and its proxy applications have continued to AI-related tokens in cryptocurrencies. Most of the recent growth in token value since the fourth quarter of 2024 has been associated with proxy AI topics, which currently account for 29% of all AI- related crypto tokens. In the proxy AI ecosystem, proxy tokens account for the majority of their valuations, accounting for $4.5 billion in market capitalization, and are related to launchpads and frameworks based on data from cookie.fun, a platform that tracks AI agents in crypto spaces. The market value of the token is US$2.9 billion.

We believe that over-attention to agents is partly driven by the meme nature of many “agents” relative to their underlying infrastructure, which coincides with the increased meme coin activity observed for most of 2024. In fact, one of the first AI-related tokens to achieve virality is famous because it is recognized by the AI ​​agent that runs the famous X account truth_terminal (now with more than 250,000 followers), rather than having it related. the underlying project or governance structure.

That being said, some AI proxy tokens provide more practicality by granting access to token-gated chat terminals or services where agents can provide a differentiated perspective on various topics, such as crypto market conditions. Meanwhile, in our view, AI infrastructure tokens are often based on revenues for specific projects, often used to pay for fees and governance.

So far, most AI agents, launchpads, and other infrastructure have found their homes on high throughput and low-cost blockchains—especially on Solana and Base. Solana accounts for $4.2 billion in proxy AI token market capitalization, Base accounts for $3 billion, and the remaining chains account for $1.5 billion in market capitalization. We believe that this is partly because low-cost architectures are necessary to support the widespread popularity of AI agents. Furthermore, we believe that the strong developer ecosystem formed on the leading chain promotes a virtuous cycle of idea sharing and adoption.

Fighting for Investor Psychological Share: Social Influence

There are several leading agents in the current field of proxy AI and they have begun to dominate this field. By far the most prominent AI agent aixbt has gained attention by operating an X account (now with over 465K fans), which is dedicated to interacting with crypto audiences on the platform. It has a token gated terminal where users holding a sufficient number of project tokens can access a dedicated chat space with the agent, allowing private access to the real-time "ideas".

Other leading agents, such as zerobro (with 119K fans), follow a similar pattern, gaining attention in the field through prominent social media profiles. Zerebro specializes in on-chain art generation. Its native tokens can be used to pay for creating images, while its chat terminal is controlled by NFT.

However, not all proxy AI tokens have a practical appearance. One of the top AI tokens (tracked by cookie.fun) is Fartcoin, whose idea was conceived through conversations with the above truth_terminal AI agent. That is, the origin of the token itself is related to AI proxy, although the long-term utility of the token itself is still unclear except for the meme importance.

We believe that the interaction between memes and the token utility of AI agents attracts traders in all areas, from speculators to value investors. That being said, given the rapid development of the field, we believe that the final range and capacity of any single token remains largely unknown. That said, it remains an open question as to whether meme coins associated with AI agents will develop and/or show any utility beyond pure speculation or access gating. We will discuss this further in the Risks and Future section below.

Follow the Fund: Launch Platforms and Infrastructure

After becoming one of the best performing cryptocurrency sectors in November and December 2024, AI proxy tokens suffered a significant hit in January 2025, partly because the market became severely saturated in such a short period of time. This has led to some market consolidation. Since many of these tokens compete directly with memecoin in the cryptocurrency’s attention economy, it is still difficult to plan for value accumulation in this area. In the short term, we found that direct protocol revenue tends to be concentrated on the trading interface and launchpads for deploying AI proxy tokens rather than the tokens themselves, although their total market capitalization is relatively small.

On Base, Virtuals Protocol has been the leading AI startup platform, supporting simplifying the launch of AI agents and tokens in gaming and various application fields. (Note: Virtuals announced its extension to Solana on January 25.) Create Virtuals agents without any encoding. Users simply need to fill out a simple form and spend the required amount of Virtuals platform tokens. After submission, the baseline agent will be initialized on the Virtuals infrastructure, while the relevant tokens are minted on the chain. Initially, the tokens are deployed in the binding curve and once a certain liquidity threshold is reached, it will be transferred to the Uniswap pool. (Note: This has some similarities to the token release on pump.fun and its transfer to the Raydium pool.)

To date, nearly 16,000 proxy tokens have been launched on Virtuals, generating more than 20 million Virtuals tokens for fees. That being said, the number of agents launched has decreased over the past few weeks, falling from a peak of 1181 a day to an average of 31 in the last week of January. Additionally, the number of tokens with sufficient liquidity has dropped to an average of one to two per day. Overall, of the 15,985 tokens launched, only 334 (2%) achieved sufficient liquidity to move to the Uniswap pool, indicating that competition for attention and capital is very intense.

We believe that the main reason for this decline is the difficulty in creating new agents that are sufficiently different from existing ones. While Virtuals agents can be customized in their cognitive, voice, and visual cores, parsing changes between agents launched on Virtuals has become an increasingly difficult task—as is the same as mind-sharing competition like memecoins. Nevertheless, we believe that as AI proxy integration is extended and use cases further explored, launchpads like Virtuals will play a key role in the proliferation of proxy in this ecosystem. In fact, the aixbt mentioned above is launched on Virtuals.

The main alternative to Virtuals launchpad is the ElizaOS proxy framework. Unlike the lean deployment of mature launchpads, a proxy developer framework like ElizaOS provides only the technical support needed for proxy building. That is, more technically oriented proxy creators can use ElizaOS to launch highly customized proxy on different blockchain networks, as model hosting, verification, and other engineering tasks are left to the creators. That being said, AI proxy hosting companies like Fleek also support codeless deployments based on ElizaOS models.

As a pure AI proxy development framework, ElizaOS does not have native tokens. However, the ai16z governance token on Solana (now renamed ElizaOS token) is often seen as the proxy for the technology adoption, as Shaw Walters, the creator of ai16z Decentralized Autonomous Organization (DAO), is the founder of Eliza Labs. The laboratory oversees the development of ElizaOS. ai16z DAO itself manages investments on and off-chain, and AI fund managers (built using the ElizaOS framework) manage transactions and positioning.

It is particularly worth mentioning that the ElizaOS framework has been attracting much attention since its release - once becoming the hottest repository on Github. If a developer finds a project particularly influential, he can choose to "star" the repository, just like like a photo or post on social media. The stars obtained by ElizaOS make it quite competitive in many other leading AI proxy frameworks, including those introduced by tech giants such as Microsoft. (See Figure 3.) We believe that this suggests a real interest in the intersection of AI proxy and on-chain activity, a core differentiation feature of ElizaOS as the AI ​​proxy framework.

In addition to Virtuals and Eliza, there are many other AI proxy frameworks and launchpads that have surfaced and found their own niche. For example, Griffain aims to create a proxy network tailored to DeFi activities. At the same time, what makes the Arc proxy framework unique is that it is designed to be lightweight and modular in addition to being built with Rust. We expect this field to develop rapidly as these frameworks develop and new frameworks are adopted.

Risk and the future

Additionally, we believe that the decline in AI agent performance in January may herald the industry will mature ahead of schedule as more capital flows begin to shift to DeFAI (Decentralized Finance + Artificial Intelligence) and/or other infrastructure.

DeFAI represents the combination of artificial intelligence and encryption technology to enhance various DeFi capabilities. The benefits include the ability to run automated revenue farming strategies, using prediction algorithms to enable better risk management and fraud prevention. As many DeFi protocols begin to become rigid, the convergence of AI capabilities with existing ecosystems can drive new innovations. Over time, the industry is expected to foster new financial products and scale many DeFi platforms by using the computing power of artificial intelligence.

However, at present, the long-term utility of AI proxy tokens is unclear in addition to gated access and promoting governance. The reality is that while AI agents have made significant progress very quickly, we have not yet reached the point where fully autonomous AI agents can handle complex real-world tasks without any supervision. Their current reliability is still limited and their costs are still ridiculously high. Many AI agents also fail to handle data verification issues consistently, which can cause legal issues or affect user confidence.

Still, persistent breakthroughs like the emergence of the DeepSeek R1 model—which itself focuses on advanced “reasoning” tasks—can overturn concerns about speed-cost ratios. In fact, these models are growing rapidly, accounting firm Deloitte predicts that within two years, half of all companies currently using generative AI will likely launch AI agents.

Ultimately, the transformative vision is that we may have a multi-agent system where autonomous AI agents strategically collaborate and/or compete to optimize results that are more complex than they may be at present. However, the new developments in this field make predictions challenging. Additionally, big tech companies like OpenAI have only recently started releasing their early AI agents, and we expect more companies to follow up soon. The development of centrally managed AI agents—which may be integrated into the traditional payment track—can also be disruptive in the adoption of on-chain AI agents. We believe that developments in this field will rely heavily on pioneers and early adoption of flywheels.

in conclusion

Agentic AI has been one of the topics of the most attention and trading opportunities in liquidity tokens over the past few months. While many of the major tokens have fallen sharply from their all-time highs (watched in early January), we believe that in the long run, developer interest and capital inflows in the sector could bring a significant boost to the industry as a whole. force.

But at the same time, we believe that it may be too early to predict the long-term value capture of AI tokens, as both cryptocurrencies and the broader technology sectors may interfere with this area. Furthermore, we believe that the on-chain utility of the agents may not be sufficient to guarantee that thousands of highly-used agents will be available in the short term. Nevertheless, we believe that the field remains an important topic to be watched due to the fast pace of innovation and the long-term potential.