Talk presented by Marisa Ferrara Boston at Collective Intelligence 2026, held jointly with HComp. Adapted into the companion visual essay below.
When experts correct an agent’s work, the accepted answer often survives while the explanation disappears. Was the agent wrong? Was information missing? Did the procedure change? Each possibility calls for a different response, but the corrected output alone may not tell us which one applies.
In my talk at HComp/Collective Intelligence 2026, I explored how agents might help preserve those distinctions—and what organizations could learn if observations remained connected to the work and reached people with the expertise to interpret them.
The visual essay
Read “Agents Are Instruments, Not Pilots” →
This short visual essay adapts the talk into a reader-paced story. It follows a corrected audit workpaper through the questions the correction leaves unanswered, then explores a design proposal: keeping the agent’s attempt, evidence, expert intervention, local context, and disposition together as the work moves through the organization.
The proposal treats agents as instruments as well as task performers. An agent might be useful at observing work before it can reliably finish the task itself. The record it helps create could support expert judgment and future investigation.
Related work
Navigational Assurance explores the broader idea of agents as instruments that help people orient themselves, preserve what they learn, and navigate unfamiliar work.
