Cobo co-founder Shen Yu expects artificial intelligence agents to become a more consequential source of on-chain activity than human users, with Ethereum among the networks he is watching for signs of mass adoption. In a discussion with Polylink Technology founder Jin Ming, Yu argued that AI could place blockchain infrastructure behind the interface, allowing software agents to handle tasks that today require users to navigate wallets, transaction fees and security decisions themselves.
The idea carries implications for how crypto products are designed. Many current on-chain applications assume users understand private-key security, token approvals, bridging and transaction execution. Yu said that requirement remains a major obstacle to mainstream use. Agents could become an intermediary layer that interprets a person’s instructions and carries out selected actions across decentralized networks.
Yu did not argue that Ethereum would be the sole destination for these systems. Multiple networks could support agents, he said. But Ethereum’s decentralized infrastructure could make it suitable for functions such as agent registries and related coordination components, particularly if developers build services that agents can use without relying on centralized operators.
Ai changes the cost of building
Yu framed the shift through the falling cost of execution. AI systems can compress, organize and retrieve large bodies of human knowledge at relatively low cost, he said, reducing the distance between an idea and a working product.
That changes the economics for individual builders. Yu said he has created a series of small tools and dashboards since AI products became broadly available, using the technology to extend what one person can produce. He has yet to settle on a larger system-level project and said he remains in an exploratory phase.
His account reflects a growing divide in technology development: AI can make it easier to build prototypes and specialized tools, while identifying a durable product with a defensible business model remains much harder. The availability of coding assistants and research tools does not automatically solve distribution, security, product-market fit or regulatory questions, especially in financial applications.
Yu’s own daily workflow has moved heavily toward AI-filtered information. He said he uses three separate agents: one for instrumental decisions, one aimed at reflection, and a third that reviews the previous day’s activity, identifies possible blind spots and proposes short reading material.
He also described a note-taking system that connects earlier areas of interest with new subjects. Rather than reading large amounts of unprocessed source material, Yu said he typically reviews AI-generated digests and sometimes converts them into slides for rapid scanning. The approach treats AI less as a search engine than as a personal research and review layer.
From mining infrastructure to agent networks
Yu connected his interest in execution tools with his earlier work in crypto infrastructure, which dates to 2011. He said mining presented practical problems in 2013, including international network latency and packet loss that could reduce operating efficiency. He also saw the settlement practices then used in the mining industry as inefficient.
Those early infrastructure challenges differed from today’s discussion around agents, but they share a focus on reducing friction in systems that operate across borders and without conventional intermediaries. Mining operators needed more reliable network connections and settlement processes; agent-driven applications would need secure permissions, reliable execution and clear limits on what software can do with a user’s assets.
The security question is central. An agent that can submit transactions on behalf of a person may simplify usage, but it would also need constrained authority. Product designers would have to decide whether an agent can move funds freely, spend only within preset limits, access specific applications, or require additional approval for sensitive actions. Yu’s view places that design challenge close to the center of the next generation of Web3 products.
Bitcoin remains the core custody thesis
Yu said his view of Bitcoin has remained largely unchanged for more than a decade. He described it as “digital gold” and emphasized the ability to hold an asset directly without depending on layered intermediaries.
He linked that feature to the idea of the “sovereign individual,” a concept also discussed by Jin. In Yu’s framing, AI expands an individual’s productive capacity, while Bitcoin provides a form of property control that can be exercised through direct custody.
Yu said Bitcoin’s role becomes most visible in periods of stress, even if attention can fade during broad market rallies. That assessment is consistent with a long-term ownership thesis rather than a strategy built around reacting to each short-term macroeconomic event.
Jin outlined a similarly long-horizon approach to Bitcoin purchases, although he attached it to specific price levels. He said he would be less willing to add above $100,000 and more inclined below $90,000, scaling purchases at $9,000 intervals down to $60,000. He presented the approach as gradual accumulation rather than short-term trading.
A narrow framework for core holdings
Yu said he does not prioritize real-time event trading and prefers to study how financial products are created and why demand develops. The DeFi cycle pushed him to formalize that framework, he said, before the growth of tokenized representations of traditional assets encouraged him to study conventional markets more closely.
His process begins with small watch positions, usually no more than 2% of a portfolio and sometimes only a few tenths of a percent. A larger allocation requires further research. Yu said an asset moving from roughly 10% to 20% of a portfolio must demonstrate an exceptionally large total addressable market, coherent business logic and defensibility.
He summarized the approach as “monopoly + growth.” Bitcoin, Ethereum and Tesla fit within that framework for him, while SpaceX remains under observation. Yu also said there had been internal discussion about whether Ethereum should retain its standing in that group, underscoring that even a long-held thesis can be reassessed.
The discussion also challenged conventional diversification assumptions. Referencing economist James Tobin’s work on portfolio construction, Yu said assets that appear weakly correlated in ordinary conditions may move together during systemic stress. His conclusion was that negative correlation cannot be assumed to provide dependable protection in the most severe market scenarios.
Jin, whose work centers on video AI technology and who is also a Chinese national freediving team athlete, compared risk management in markets with the discipline required underwater. He described the danger of blackout during breath-hold diving, where a delayed loss of consciousness can turn an ordinary reflex to inhale into a life-threatening event beneath the surface.
That analogy fits the conversation’s wider caution: AI may make individuals and software systems more capable, but automated execution does not remove risk. As agents take on more on-chain activity, the most durable products may be those that combine machine-speed operation with clear controls over custody, permissions and exposure.
Explore how AI and blockchain converge to power software agents, automate crypto decisions, and reshape on-chain interaction and adoption.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
