📰 Key Takeaways

Coinbase launches in-app AI agent “Coinbase Advisor” registered as a licensed financial advisor with US SEC and CFTC, capable of automatically executing recurring purchases and fund allocation tasks without continuous human monitoring. Using ETH dollar-cost averaging as example, the agent automatically fetches 30 days of hourly price data, identifies the lowest historical price window, sets up $20 daily market purchases and runs for two consecutive weeks. However, after studying over 925,000 token holders jointly by Pantera Capital, Stanford University, Ava Labs and the Crypto Council for Innovation, they found AI agent treasury shows $30 million in paper gains, while token holders collectively lost $191.7 million, and most projects lack clear evidence of autonomous trade execution — they’re still just basic API integrations. Meanwhile, stablecoin issuer Circle has launched tools enabling AI agents to use wallets and initiate programmable payments, with its CEO predicting billions of AI agents will use stablecoins within five years; infrastructure provider Crossmint launched a service authorizing AI agents to pay with Visa credit cards; Keyrock reports AI agents settled $73 million across 176 million transactions between May 2025 and April 2026.

💬 JudyAI Lab Perspective

Coinbase launching licensed AI financial agents marks the正式 intersection of “AI agents + financial compliance”. This isn’t just a product iteration — it’s a key milestone for AI agents moving from concept to regulated entities.

However, the numbers uncovered in the original article deserve serious attention from all AI builders: after studying over 925,000 token holders jointly by Pantera Capital and other institutions, they found AI agent treasury shows $30 million in paper gains while holders collectively lost $191.7 million, and most projects are still just basic API integrations without truly autonomous execution capabilities. This shows “automation shell” doesn’t equal “user-beneficial outcomes” — the core challenge lies in aligning the agent’s objective function design with user interests. Around the same time, infrastructure providers like Circle and Crossmint have already seized the opportunity to lay out AI agent payment rails, and Keyrock data shows related transactions reached $73 million across 176 million transactions in one year — demand is real, but who’s protecting end-user interests remains a gap.

Our recommendation when evaluating any AI agent service: ask first — whose interests does this agent actually optimize for? Whether the objective function is well-designed determines whether this agent is a tool or a trap.

📅 Source Info

🔗 Further Reading