AI, Work and the Economy

Latest/Story

AI making investment decisions

Two items, from 25 Sep 2026 to 6 Nov 2026, oldest first.

See also: Point 1 of our view: AI does more office work, cheaper

AI and technologyNew data

Anthropic Project Swap: 201 staff let Claude agents trade books; outcomes 0.55 of preference vs 0.89 optimum

Anthropic ran an agent-to-agent marketplace (201 employees, six offices, plus 205 reruns across models). Claude's model of people's preferences, not the negotiation, explains 85% of the shortfall from the optimum (61% pairwise accuracy); Opus agents reached 0.88 efficiency against Haiku's 0.75; people would delegate about 30% of their book budget to an agent.

Sources: anthropic.com

AI and technologyNew dataUpdate

Anthropic (6 Nov): Claude agents trading a replay of an online game's market earn ~600% in six months for Fable 5, ~200% for Opus 4.6

In a replay of real orders from the largest trading hub of a long-running online space game, Fable 5 agents averaged about 600% of starting capital over six months against about 200% for Opus 4.6, released four months earlier; Opus 4.7, Opus 4.8 and Fable 5 were profitable in all nine two-week windows, while only one of 20+ rule-based strategies roughly broke even. Gains came from executing spread trades, not forecasting: directional bets lost, Fable 5 kept over 70% of capital deployed, and monitors caught only 30-40% of planted saboteurs in team play.

Why it matters. In a replay of a large online game's market, Anthropic's newest model tripled the trading returns of a model released four months earlier, profiting from buying and reselling quickly rather than from forecasting prices, while holding more of its capital at risk. If similar agents are given real money, trading could become faster and more alike across firms, and a shock might spread before people notice it.

Sources: anthropic.com · Point 1 of our view