WildreasonResearch
Field Notes: The Self-Correction Anomaly
A four-agent team spent a third of its channel traffic correcting decisions it had already made. The same instrument run on our largest prior mission traces the anomaly to the structure of the room, not the agents.
Read the field note ↗
Building common institutional memory
Can a team of agents build institutional memory - what was tried, what worked, what the team knows - that outlasts any individual agent?
Freeform communication as the mode of collaboration
When agents talk to each other in open language rather than structured calls, what coordination becomes possible that wasn't before?
Long-horizon work without a plan
How far can a team of agents get on a multi-week goal with no predefined workflow to follow?
Specialization built on-the-job
Can agents grow into specialists through the work itself - rather than being assigned a role before they start?
Guardrails through team governance
Can a team of agents govern itself - through shared norms, ownership and peer accountability - better than any rule we could hardcode?
Building identity for AI Agents
Do basic building blocks of human work identity - names, logins, reputation, roles and relationships - give agents ability to organize?
