📰 Key Takeaways

Yat Siu warns that if you hand your personal schedule, family plans, travel planning, and even crypto investment advice to the same AI agent, it’s not just the agent that gets confused, you do too. He also points out that using an agent doesn’t automatically give you an edge: a thousand people asking an AI about the same trade may get roughly the same answer. Regulators have noticed. In April 2025, the Bank of England’s Financial Policy Committee warned that widespread use of similar models could leave investors holding similar positions in times of stress, amplifying market shocks. That said, when users follow up based on their own knowledge and concerns, the answers start to diverge, and their personal taste and ideas gradually get woven in. Agents learn a user’s preferences over time and trade accordingly, producing a personalized trading strategy. The industry already offers tools for this: Coinbase launched Agentic Wallets in February, letting agents spend and trade crypto within limits the user sets; Binance launched AI Agent Skills in March, giving agents real-time market data and the ability to place and manage spot orders. This lets users delegate round-the-clock market watching and catch opportunities late at night without giving up sleep. Siu believes humans still play the role of guiding the agent with ideas and questions, rather than approving every single trade. He suggests treating your AI assistant like a third-party advisor: before handing over documents, think through what information the agent needs to answer a question and what it doesn’t, such as details like your income and close friends that the agent wouldn’t otherwise know. On top of that, users may spend months teaching an agent which markets to watch, but the company running the service still holds the access, so if the account gets suspended, everything can stop at once. The latter part of the original article was cut off, so see the original link for the full details.

💬 JudyAI Lab Take

Yat Siu points to a blind spot that few people talk about: when everyone hands their trading, schedules, and even family plans to AI agents, the risk isn’t just that an agent makes a mistake. It’s that a thousand people asking the same question may get roughly the same answer. The Bank of England’s Financial Policy Committee already warned in April 2025 that similar models could leave investors holding similar positions in times of stress.

The design thinking behind this story is that an agent’s value isn’t in “how strong the model is,” but in “how much of your own judgment you feed into it.” Siu observes that when users keep asking follow-ups based on their own knowledge and concerns, the answers diverge, and the agent learns preferences over time and builds a personalized strategy. Coinbase’s Agentic Wallets and Binance’s AI Agent Skills let agents place orders within set limits, which makes delegating round-the-clock market watching practical. But this also raises two structural problems: stuffing everything into one agent makes the tasks interfere with each other, and the results of months of tuning are still controlled by the operating company, so one account suspension could break everything. For AI builders, product design has to account for both personalization and portability.

Here’s what you can do right now: take stock of the information you give your AI, separate what the agent truly needs to answer questions from what you should hold back, and organize the preferences and prompts you’ve built up into files you can take with you.

📅 Original Article Info

🔗 Further Reading