What it does
A 9 July 2026 research note described eight agents, built on OpenAI and Anthropic models, that classify markets into Goldilocks, reflation, stagflation, and risk-off, then shift allocations. In backtests spanning roughly two decades, every agent beat a 60/40 portfolio on a risk-adjusted basis; the best added 0.7 percentage points of annualized return at lower volatility.
Architecture
A supervisor classifies the regime; specialist agents propose weights; the published work is research-grade, not a production order-management hook.
Evidence
InvestmentNews, 13 July 2026, citing the JPMorgan research note. Evident separately reported the same 0.7pp figure.
Caveat
Backtests are not live trading. Transaction costs, capacity, and model drift are not independently audited here. Treat this as a published research architecture, not a production P&L claim.