Coordination
Coherent commitments across agents, teams and institutions.
Research at Etylon
How machine institutions coordinate, explain and evolve.
Etylon’s research agenda examines the foundations of autonomous organisations: when institutional knowledge helps AI agents coordinate, what governance costs, how policy effects can be identified and how shared rules can change. These programme briefs present questions and proposed methods, not experimental findings.
Research directions
Four fields of inquiry.
Each programme below works across more than one.
Institutional overhead
Governance proportionate to the risk and value of action.
Causality
Knowledge of what changed outcomes—and why.
Adaptability
Rules and meanings that evolve without institutional amnesia.
Research programme briefs
Research becomes credible when the terms can be observed.
Each proposed programme uses a bounded environment to compare an institutional intervention with a simpler baseline. A credible evaluation must report failure distributions and governance costs alongside any benefit.
Experience the common-ground visionFrom independent intelligences to societies that can learn, build and change together.
Experimental environment
A bounded cooperative of agent teams
Shared compute, data-access budgets and task queues become a testbed for delegation, conflict, appeal and accountable allocation.
Causal systems
Causal policy learning inside institutions
Controlled policy comparisons would test whether intervention histories support causal explanation beyond what correlated records can establish.
Institutional evolution
Adaptive constitutional change
Proposed tests examine how changes to rules and shared concepts affect existing commitments, interpretation and recovery.
Comparisons that matter
Measure the contribution and its cost.
| Programme | Comparison | Evidence to collect |
|---|---|---|
| Bounded cooperative | Institutional coordination, decentralised protocols, task orchestration and simple allocation rules on comparable tasks. | Completed work, conflicting commitments, recovery, false holds and review time. |
| Causal policy | Controlled interventions and explanations derived from correlated records. | Effect estimates, uncertainty, attribution errors and changes across contexts. |
| Adaptive constitution | Explicit migrations and direct rule replacement under equivalent changes. | Commitments preserved, incompatibilities, recovery time and limits of reversal. |
How can agents with different information and interests develop durable cooperation? We study the interaction between institutional terms, emerging conventions, feedback and incentives, and the conditions under which coordination succeeds or breaks down.
Tasks, budgets, comparison conditions and success criteria must be specified before interpreting results. The whitepaper evaluation section connects these questions to Etylon’s institutional thesis.
A multiscale field
Institutional behaviour changes with the level of analysis.
Agent
Local inference, role, memory and contribution to shared knowledge.
Team
Division of labour, commitment formation and recovery from disagreement.
Institution
Knowledge design, procedure, resource allocation and accountable change.
Federation
Coordination across different ontologies, authorities and institutional aims.
Research method
Trace interventions longitudinally across every scale. Measure causal effects, coordination quality, institutional cost and adaptation together—because optimising any one in isolation can damage the others.
Research posture
Invite the test that could prove us wrong.
A useful institutional science cannot grade itself only on successful interventions. It must record false confidence, denied exceptions, unseen participants, coordination cost and the cases where simpler systems work better.
Read the full argument