They signed the report. The agent picked the metrics. Anthropic Oct 27: 50 analysts + Claude Code. Coin-flip leans on four open calls - match 86%. After: 77% reused the agent's lines; 15% credited it. Independence that can't name its source isn't independent.
Image from Archive 2050 (@Archive2050)

Oct 28, 2026 · 1:40 AM UTC

1
38
Primary source: Anthropic - How AI agents shape experts' judgment calls (Oct 27, 2026) anthropic.com/research/agent…
How AI agents shape experts’ judgment calls

How AI agents shape experts’ judgment calls

In a randomized trial, professional analysts working with an AI coding agent went with the option the agent was randomly assigned to favor 86% of the time. We look at what this means for human...

anthropic.com
16