JPMorgan’s AI Portfolio Bet Echoes Jack Dorsey’s Vision, But With a Big Warning

JPMorgan’s artificial intelligence (AI) agents beat a traditional 60/40 portfolio across two decades of backtests. The bank celebrated the result, then warned investors not to trust it.

The test asks whether AI can move from assisting analysts to allocating capital itself. It lands as Jack Dorsey champions a similar shift in how people work with machines.

How JPMorgan’s AI Agents Beat the 60/40 Portfolio

JPMorgan’s cross-asset strategy team built eight AI agents that move between stocks and bonds as conditions change. The strategists, led by Thomas Salopek, shared the results in a July 9 note. The system reads four macro regimes set by growth and inflation.

The benchmark is fair and meaningful. The 60/40 split anchored balanced portfolios for decades. In 2022 it had its worst year since 1937, when stocks and bonds sank together.

The agents favored stocks when growth looked strong and bonds when it weakened. Over 20 years of backtests, the best agent topped the 60/40 portfolio by 0.7 percentage point a year.

It did so with 2.8% lower annual volatility. All eight agents won on a risk-adjusted basis, with Sharpe ratios of 0.74 to 0.95 against the portfolio’s 0.61.

The agents ran on off-the-shelf models from OpenAI and Anthropic, yet beat JPMorgan’s own rules-based regime model. That extends the bank’s recent AI calls into riskier territory.

Why the Bet Echoes Dorsey’s Agent-First Vision

The approach mirrors a philosophy Jack Dorsey described. The Block chief executive now defers to AI agents rather than directing them.

i’ve shifted from telling agents what to do, to asking them what to do, and pulling the best thread.

— jack (@jack) July 10, 2026

Dorsey has already bet his company on it, cutting over 4,000 jobs at Block in February and crediting AI. That was about 40% of staff. JPMorgan’s agents apply the same logic to markets, part of a wider push toward AI agents handling money.

The Warning Veteran Quants Know Well

JPMorgan was clear about the limits. The results come from historical simulations, not live trading, and the bank cautioned against over-reading them.

Richard Bernstein, a veteran Wall Street quant, put it more sharply. New strategies, he noted, rarely publish backtests that lose.

As one of Wall Street’s original #quants I would caution about getting too excited about #AI outperforming benchmarks. Have you ever seen a new strategy’s publicly disclosed #backtest that underperformed?https://t.co/w3NmdoGDMc

— Richard Bernstein Advisors (@RBAdvisors) July 10, 2026

His point is publication bias. Flexible AI models can fit past noise, then fade when live costs and unseen regimes hit.

JPMorgan also warned that crowded AI trades could amplify market stress, echoing broader cracks in AI spending. Backtests have flattered many strategies that later stumbled. Whether these agents survive live markets is the real question.