A controlled study shows agents can steer analytical choices

Preference can shape the answer

Across four analytical choices, the 50 agent users matched its assigned preference 86% of the time. These were reasonable analytical alternatives; agreement does not establish that an answer was wrong.

A controlled study shows agents can steer analytical choices ↗

Time comparisons have a limit

Eleven participants opted out of AI. Their experience and outcomes differ, so comparing their time with agent users does not establish a causal productivity gain.

A controlled study shows agents can steer analytical choices ↗

Correction, November 3

The earlier summary called 86% a rate of final choices. The study reports that figure across four analytical choices. The earlier version remains in history.

A controlled study shows agents can steer analytical choices ↗

A discussion of planning habits

A public listing of Cole Medin’s October 31 video links the study and a planning exercise. Only its description and chapters were accessible; the video and proposed exercise were not verified.

Cole Medin discussion: creator description and chapter list ↗

What you can do

1. Read the experimental design before applying the result to workplace decisions. Open reference ↗

Where the evidence stops

Primary study account read November 3. Study not reproduced; results are limited to its task and participant sample.

Sources read. Sources read November 3, 2026 UTC. No third-party software, model inference or full workflow executed.

Sources & further reading