When the bots arrive in the home office
A Bloomberg profile this summer described a software executive spending a year building an options bot at his kitchen table, only to match an index fund. Olga Magomedova on what the new retail AI trading wave changes, and what it does not.
The home office bot and the index fund
The story Bloomberg published in August is the kind of story that will keep being published. A software sales executive in Los Angeles spent more than a year of nights and weekends trying to build an automated options program of the sort that quantitative hedge funds run at scale. He wrote the Python by hand, ran simulations, and tracked hundreds of stocks. The result was a system whose performance was not materially different from a passive S&P 500 index fund.
The detail worth holding is not that the effort failed. It is that the effort was serious, technically competent, and produced a result the trader could have obtained without any of the machinery. That is the shape of most retail AI trading experiments right now.
What the new tooling actually changes
A year ago the retail trader who wanted a machine to help was mostly on their own. Today there is a broader shelf of agentic trading platforms marketed to individuals, and institutional adoption is running well ahead of the retail layer. A hedge fund manager profiled by CNBC in September described building a firm, Bracket22, staffed entirely by AI agents and claimed he is at least ten times more productive than at his previous firm. Whatever the eventual outcome of that experiment, the direction of travel is clear.
For Magomedova, the interesting question is not whether the tooling works. Some of it will. The interesting question is what the tooling changes about the trader who uses it. A tool that removes friction from execution does not remove the responsibility for the decision. It only makes the decision faster to regret.
The regulator's warning that belongs on the desk
The Bank of England's July Financial Stability Report set out a sober assessment of the same trend from the other end. The Financial Policy Committee judged that rapid progress in frontier AI capabilities since December had produced a significant increase in the risks to financial stability from cyber and operational vulnerabilities, and that frontier models are increasingly capable of identifying and exploiting software vulnerabilities at scale. The report also noted a substantial increase in the use of leverage in equity markets.
That combination matters. A market with more leverage and a technology stack with faster attack surfaces is a market in which the tolerance for procedural drift is smaller than it was a year ago. The retail trader running a machine-generated strategy through a broker's API is not outside that perimeter. They are inside it, on the thinnest edge.
The failure mode that shows up first
The first failure mode of a retail AI trading practice is rarely the model itself. It is the position size the trader accepts because the model produced the signal. The size feels justified by the sophistication of the source. The plan that would have called for half the exposure gets quietly overridden. When the drawdown arrives, the trader has no post-mortem to write, because the trade was not really theirs.
The engineering habit worth borrowing is the same one that applies without any AI in the loop. Every position must be defensible in the trader's own words, at the size the plan specified, before the model is consulted. The model is a second opinion. The plan is the constraint.
People are attracted to the idea that software can remove emotion. But discipline is not something you outsource.
The principle that does not move
The retail AI wave changes the surface of trading. It does not change the structure. A trader who kept a written journal, sized positions inside a fixed risk budget, and reviewed a weekly post-mortem before any of the current tooling existed will keep doing the same things after it. A trader who did not will find that the tooling accelerates the pattern that was already there.
For Magomedova, the household consequence is the one that matters most. The independence a trader is building is not for the good weeks. It is for the week the strategy underperforms, the platform has an outage, the model produces a signal the trader does not understand, and the household is still counting on the account. The plan built in the quiet months is the one that pays in the loud ones. The tooling that arrives after that plan is complementary. The tooling that arrives instead of it is a liability.