Brokerages, startups, and even individual investors are all developing AI agent programs to monitor assets and automate various investment transactions that previously required manual operation.
Devin Ryan, head of financial technology research at Citizens Bank, stated, "Everyone essentially has their own private family office, working around the clock for them." These AI agents could soon independently execute stock trades on behalf of users. Imagine setting your acceptable risk level, retirement plans, and your children's college enrollment timeline for an AI, then letting it fully manage your portfolio while you sleep.
This agent-based trading model, where AI not only provides financial advice but also directly executes trades, is moving from concept to reality. Brokerages, startups, and retail investors are all creating AI agent tools to manage holdings and automate investment tasks that were previously done manually.
Ryan noted, "It's like everyone now has a 24/7 personal family office working for them, awake or asleep. This change isn't ten years away; it will become widespread in the next few years." Ryan believes the future capabilities of AI agents will go far beyond buying and selling securities. In the long term, AI could coordinate tax planning, cash reserves, loans, mortgages, and all investment portfolios around the clock, all customized to personal financial goals.
Fully autonomous investing is still being refined, but the industry's development race is already underway.
The Path to Implementation: Gradual Iteration
Most institutions are not building fully autonomous trading systems in one go but are opting for a gradual rollout. Fintech startup Podium Markets AI focuses on developing AI tools for investing. Its intelligent assistant, Ivy, can integrate all holdings from a user's multiple brokerage accounts, combining the investor's risk tolerance and financial goals to generate allocation suggestions.
However, this AI does not place orders automatically. The final decision on whether to adopt the plan and execute trades remains with the user. Dirk Muller-Englander, co-founder and CEO of Podium Markets AI, said, "The AI provides information; humans hold the final decision-making power. Ordinary investors must maintain control over the final choice... We are building a resident financial AI partner, always there for the user, rather than taking full control of the account."
Major brokerages are adopting a similar approach. In May, Robinhood launched a feature allowing third-party AI agents to access user securities accounts. Online broker Public developed its own AI tool that automates the entire investment process within its platform.
Leif Abraham, co-founder and co-CEO of Public, stated, "The era of intelligent agents has completely changed the old model. In the past, investors had to research on their own, analyze independently, and trade manually. Now the entire process can be automated, and AI agents can trade directly for users according to their strategies."
Ryan estimates that overall trading volume in the intelligent finance era will increase at least tenfold. Ordinary retail investors currently trade an average of two times per month; with AI agents, future daily trading frequency could reach 20 times. "By next year, the majority of trades on many platforms will be executed by AI agents. This trend is incredible but is happening."
Retail Experiments: From General AI Models to Trading Agents
While Wall Street institutions focus on professional investment AI, many individual investors have spent the past three years trying to use general-purpose large language models for investing. After ChatGPT became popular in late 2022, many investors used tools like ChatGPT and Anthropic's Claude to organize financial reports, research stocks, and screen for investment ideas. The results have been mixed: some find AI an efficient research assistant, while others discover AI's unreliability for investment decisions.
Obi Okafore, a 29-year-old tech consultant in Georgia, USA, who also runs a financial education platform, used Claude to build an AI program specifically to search for undervalued stocks and options opportunities. "Essentially, I have Claude play the role of a hedge fund analyst to find bargains, but I verify every recommendation before placing a trade. I always believe AI is just a tool, never a replacement for a person."
Thomas Schlosmacher, a 31-year-old individual investor who runs a business building AI systems for merchants, saw online claims that AI could capture profitable market patterns. He tested a trading AI himself but suffered continuous losses. "If you want to rely on a fully autonomous AI system and hand complete control to an intelligent agent, it's best to use a professional institution. Blindly letting AI try to make money for you is not realistic."
Risk Control Barriers: Solving the Instruction Comprehension Problem
The industry's biggest challenge is emerging: teaching an AI to buy and sell stocks is not difficult, but making it accurately understand the investor's true intentions is extremely hard. For example, a user might simply tell the AI to "aggressively grow value." However, this instruction is open to vast interpretation: does it mean accepting higher volatility, concentrating holdings, using options, or tolerating a higher probability of loss? If the AI mechanically executes the command, the final result could completely contradict the investor's original intent.
Therefore, various institutions are choosing to first establish risk control limits before gradually expanding the AI's permissions. Public's platform requires that before the AI executes any action, the user must manually review and approve the entire operational logic.
Abraham explained, "The final decision always rests with the user. The AI agent has no subjective thoughts of its own; it is only responsible for executing the plan."
The more operational authority the AI takes on, the more companies need to ensure its operation perfectly aligns with expectations. Citizens Bank's Ryan emphasized, "Everything must prioritize the client's interests. Once the AI deviates from the preset model and expectations, the platform bears significant risk."
Comments