Eight JPMorgan AI agents outperformed a traditional 60% stocks/40% bonds portfolio on a risk-adjusted basis in simulations spanning about 20 years, with the best agent ahead by 0.7 percentage points a year.
The agents shifted money between stocks and bonds across four economic regimes—Goldilocks, reflation, stagflation and risk-off—rather than trying to pick individual winning stocks.
The result comes from a backtest, not live trading, and the report stresses that such models can overfit past data and often break down when confronted with unforeseen real-time events.
That caution is reinforced by broader academic research finding AI investing strategies often fail across full market cycles, turning too defensive in bull markets and too aggressive in downturns.
PwC's 2026 survey shows why the idea still resonates: 84% of consumers are managing finances more cautiously, 24% have used or considered AI tools, yet 87% still prefer human guidance.