Key Takeaways
- • Yasanji Ratnaike built a $1M paper book at Interactive Brokers run by eight defined AI agent roles, four performing research and four acting as deterministic control gates before any order reaches the market.
- • The first live order was rejected by Interactive Brokers because the account is a UK retail account and European regulation bars retail clients from buying US-domiciled funds lacking a specified disclosure document.
- • The validator agent Erik Lindqvist reversed a near-discarded trend strategy: a sample without a crisis showed a Sharpe ratio of 0.38 versus 0.70 for buy-and-hold, but a longer test including 2008 and 2020 showed the trend rule losing far less in both crashes.
- • Capital preservation is enforced by coded limits: a 5% drawdown suspends new orders and a 7% drawdown liquidates to cash, with no discretion given to the strategy.
- • The author explicitly draws no conclusion on the strategy's merit, noting the live book has run only a short time.
Summary
Yasanji Ratnaike describes an experiment, published on DataDrivenInvestor via Medium in November 2026, in which a team of AI agents runs a one million dollar paper book at Interactive Brokers. The desk separates research from control: four research agents, including head of research Arjun Nair, strategist Daniel Weiss, validator Erik Lindqvist and CIO Thomas Fischer, propose, backtest and challenge ideas, while four control functions (operations, compliance, risk and execution, led by Lars Berg, James Okoye, Marco Russo and Marcus Cole) are performed by deterministic code rather than AI. The author argues that a constraint expressed in code is a genuine control, whereas agents inclined to agree with one another are not.
The strategy is deliberately simple: it holds an asset while it trends upward, sizes positions by volatility and rebalances weekly. The desk runs on a single cloud server with version-controlled rules, scheduled jobs and a single gateway process as the only component permitted to place orders. Research runs Monday, strategy and validation Wednesday, stress testing Thursday, and the CIO review and trade placement Friday. The author receives notifications on his phone and a weekly letter from Fischer.
The first live order was rejected immediately by Interactive Brokers, which cited the account's status as a UK retail account and European rules barring retail clients from buying US-domiciled funds without a specified disclosure document. The response was a permanent control: compliance now verifies instrument eligibility before every order, and the universe was moved to European funds.
The author highlights a near-mistake as the clearest case for the validator's role. On data from 2022 to 2026, trend timing produced a Sharpe ratio of 0.38 against 0.70 for staying invested. Erik noted the sample contained no crisis, and when the test was extended to include 2008 and 2020, the trend rule returned about 7.6% through the financial crisis versus a 14.8% loss for staying invested, and fell 3.9% in 2020 versus a 12.3% loss. The author frames the timing as insurance rather than a redundant mechanism, while cautioning that a short live track record settles nothing.
Capital preservation is the stated priority, enforced through coded drawdown limits, a weekly stress test covering 2008, 2020 and an interest rate shock, and a manual kill-switch.
The strategy is deliberately simple: it holds an asset while it trends upward, sizes positions by volatility and rebalances weekly. The desk runs on a single cloud server with version-controlled rules, scheduled jobs and a single gateway process as the only component permitted to place orders. Research runs Monday, strategy and validation Wednesday, stress testing Thursday, and the CIO review and trade placement Friday. The author receives notifications on his phone and a weekly letter from Fischer.
The first live order was rejected immediately by Interactive Brokers, which cited the account's status as a UK retail account and European rules barring retail clients from buying US-domiciled funds without a specified disclosure document. The response was a permanent control: compliance now verifies instrument eligibility before every order, and the universe was moved to European funds.
The author highlights a near-mistake as the clearest case for the validator's role. On data from 2022 to 2026, trend timing produced a Sharpe ratio of 0.38 against 0.70 for staying invested. Erik noted the sample contained no crisis, and when the test was extended to include 2008 and 2020, the trend rule returned about 7.6% through the financial crisis versus a 14.8% loss for staying invested, and fell 3.9% in 2020 versus a 12.3% loss. The author frames the timing as insurance rather than a redundant mechanism, while cautioning that a short live track record settles nothing.
Capital preservation is the stated priority, enforced through coded drawdown limits, a weekly stress test covering 2008, 2020 and an interest rate shock, and a manual kill-switch.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.