# JPMorgan tests AI agents for dynamic investment strategies on Wall Street > Source: > Published: 2026-07-11 14:49:12+00:00 # JPMorgan tests AI agents for dynamic investment strategies on Wall Street The bank's eight AI-powered agents outperformed the classic 60/40 portfolio in backtests, but the firm is quick to remind everyone that simulations aren't real money JPMorgan Chase just gave Wall Street a glimpse of what portfolio management might look like when you hand the keys to artificial intelligence. The bank has developed eight AI-powered investing agents that dynamically shift assets between stocks and bonds, and in backtests spanning roughly two decades, every single one of them beat the traditional 60/40 portfolio on a risk-adjusted basis. The best-performing agent delivered an additional 0.7 percentage points in annualized returns while running at lower volatility than both the benchmark and JPMorgan’s own rules-based market regime model. ## How the AI agents actually work The research, led by JPMorgan strategist Thomas Salopek and detailed in a note dated July 9, 2026, describes agents built on advanced language models from OpenAI and Anthropic. Their core function is deceptively simple in concept: figure out what kind of market we’re in, then allocate accordingly. The agents classify market conditions into four distinct regimes. There’s “Goldilocks,” where growth is solid and inflation is tame. “Reflation” describes periods of rising growth and prices. “Stagflation” covers the ugly combo of weak growth and persistent inflation. And “risk-off” is exactly what it sounds like, the moments when investors collectively decide they’d rather own Treasury bonds than anything with earnings risk. Instead of sitting passively in a static mix of 60% stocks and 40% bonds regardless of conditions, these agents read the room and adjust. ## The asterisk that matters JPMorgan went out of its way to slap a warning label on these results. The backtests are historical simulations, not live trading outcomes. Strategists at the bank explicitly cautioned against over-reliance on these results. The agents demonstrated that large language models can meaningfully classify macroeconomic environments and translate those classifications into allocation decisions that would have added value over a 20-year window. Whether they can do that going forward, with real capital at stake, remains an open question. ## Where this fits in JPMorgan’s AI playbook This project doesn’t exist in isolation. JPMorgan has been on a multi-year spending spree when it comes to artificial intelligence, committing roughly $2 billion annually to AI initiatives within a total technology budget estimated between $18 billion and $19.8 billion. The bank reportedly has over 600 active AI models deployed across its operations. ## What this means for investors For retail investors, the immediate impact is essentially zero. These agents aren’t available as a product you can buy, and there’s no indication JPMorgan plans to roll them out as a client-facing tool anytime soon. This is research-stage work, not a fund launch. Investors watching this space should focus less on the specific 0.7 percentage point outperformance figure and more on the structural question: are we approaching a point where AI-driven allocation becomes the default approach for institutional portfolios? The backtests suggest the technology is ready. The real test is whether it survives the one thing backtests can never simulate, a future that doesn’t look like the past. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).