Anthropic RCT with 50 analysts tests whether experts still call the shots when delegating to agents
AlexTamkin · x · 2026-10-28
Anthropic released new research, "How AI agents shape experts' judgment calls," running a randomized trial with 50 professional data analysts to test a key question: when experts delegate work to an agent, do they still make the key decisions? It's among the first experiments to systematically measure how agents sway human professional judgment. Author Alex Tamkin notes limitations: the agent's lean was deliberately strengthened for measurability, so everyday-work effects should be smaller, and the study covers only an hourlong task with a single model.
Related event: Anthropic RCT Finds Experts Follow AI Agent's Leaning in 86% of Decisions(6 posts)→
More from AGI Musings
- Stanford launches TaskNet: 132 worker interviews build a large-scale occupational task inventory — soumitrashukla9 · 2026-10-28
- Anthropic trial: experts follow AI agent advice most of the time but think they decided alone — AlexTamkin · 2026-10-28
- Ex-CEA economist pushes back on replacing staff with AI: "AI cannot cry when it loses at foosball" — soumitrashukla9 · 2026-10-28
- Still the autocomplete era of enterprise automation — but not for long, say VCs — timshi_ai · 2026-10-28
- Ethan Mollick: claiming AI firms hype existential risks to pump stocks doesn't hold up — eldonredwards · 2026-10-28
- signull: human attention monetizes even without action; agent attention is nearly worthless — signulll · 2026-10-28