ChainCatcher “From Cryptocurrency to Smart Economy” Roundtable: AI Agent is shifting from conversational to executional, with trust and verifiability becoming key bottlenecks for scaling

ME News report, April 19 (UTC+8), at the “Crypto 2026: From Cryptocurrency to Smart Economy” themed forum held in Hong Kong, guests including Laughing, Head of Asia-Pacific at KiteAI; Anita, Head of Asia-Pacific at Sentient; Michael, Co-founder and CEO of Brevis; Jerry, Product Lead at Mentis; and “Ahuang” Ricardo, Asia-Pacific BD Lead at Predict.fun, held a roundtable discussion around the theme “AI × Crypto — the next-generation digital economy foundation.”

Regarding whether the boom in open-source AI Agents (such as “Lobster Farming”) is a short-term sentiment or a long-term trend, Laughing believes it marks the beginning of a long-term shift in AI—from a “conversational” paradigm to an “execution-based” paradigm. However, since Agents have no physical identity, merchants find it difficult to trust their trading behavior. Issues such as data leakage and uncontrollability remain obstacles that must be overcome for large-scale adoption.

Jerry, drawing on his own usage experience, pointed out that current Agents still face security risks such as vague memory, dead-loop behavior consuming Tokens, and accidental file deletion. Although the newly emerging Agents have been improved, they still have shortcomings.

“Ahuang” Ricardo stated that behind short-term sentiment are real user needs, but truly reproducible, profit-making trading Agents remain very rare, and there is still a long way to go.

Addressing key bottlenecks such as trust, security, and verifiability, Michael emphasized that the current Agent execution process is like a black box, and users cannot verify whether its decisions are executed as promised. Zero-knowledge proof technology can mathematically prove the source of an Agent’s outputs and whether its behavior complies, while not exposing privacy.

Anita added that Sentient is committed to building an open and collaborative network of AI agents, advancing the deep integration of AI and blockchain across the identity, data, and incentive layers, and providing underlying support for the next-generation digital economy.

As for the application scenarios that are most likely to scale first, the guests generally believe that the programming field is relatively mature. In prediction markets, Agents are expected to perform better than humans in certain tracks due to rationality and the lack of emotions. In the short term, they are optimistic about API Marketplaces and cross-platform consumer shopping Agents. In the mid term, they are optimistic about content-creation AI (such as AI influencers). Agent-to-Agent interactions and the Agent/Skill Marketplace are seen as important subsequent hotspots. (Source: ChainCatcher)