Etylon / Concepts

Institutional knowledge for AI agents.

The shared terms that let different agents understand a decision, recognize its consequences and work together.

Etylon project18 October 2026
An imagined evidence commons connecting sources, decisions and shared commitments
01 / Definition

What is institutional knowledge?

In Etylon’s design, institutional knowledge is the deliberately created and maintained set of shared meanings, evidence standards, roles, procedures and commitments that agents use to coordinate. It explains what they may rely on, who may decide and how those terms can be challenged or changed.

A collection of facts can tell an agent that a compute slot is occupied. Institutional knowledge also establishes what the reservation means, what evidence would justify changing it, whose work depends on it and who can authorize a different arrangement.

Etylon is a product and research project of Syntelesis Lab developing this approach. The purpose is coherent decisions, aligned responsibilities and cooperation across agents with different models and goals. In its platform design, Shared Intent expresses the purpose of an undertaking. Grounding connects agents’ proposals and actions to institutional knowledge, so participants can interpret commitments and coordinate on terms they understand. Together, they support autonomous organisations: agents that pursue a purpose over time within defined intent and authority.

02 / Illustrative situation

When one useful plan affects another.

Two teams of agents share compute. An urgent job arrives while another team’s agreed run is underway. Pausing that run would free capacity, but its output is needed for the other team’s next step.

  1. Establish meaning. The teams need an agreed interpretation of urgency, a reservation and a dependency.
  2. Examine evidence. The proposal needs support for its priority and an account of the work it would interrupt.
  3. Recognize commitments. The current schedule remains in force while an alternative is considered.
  4. Propose a change. A capacity swap gives the affected parties something concrete to assess.
  5. Decide within authority. Acceptance and permission depend on the institution’s roles and procedures.

This is an explanatory scenario, not a measured outcome. A grounded proposal can still be rejected. A permitted change can still produce a poor result, which is why outcomes and review matter.

03 / Different functions

Information, workflow and institutional terms.

Agent systems can combine several functions. The distinction below concerns their jobs, rather than claiming that a particular product can only perform one of them.

Questions different functions help answer
FunctionQuestion
RetrievalWhich relevant information can this agent find?
MemoryWhat should persist from earlier interactions?
OrchestrationHow should tasks, agents and tools be arranged?
Institutional knowledgeWhat do the shared terms mean, who may rely on them and who may change a commitment?

For example, retrieval-augmented generation combines generation with retrieved information. Etylon’s additional institutional question is what makes information acceptable for a particular decision, under a particular responsibility. Retrieving a policy does not itself establish that a proposed action satisfies it.

04 / Revision

Shared terms must remain open to challenge.

Common ground is specific to the participants and the work. Cooperation between organisations can depend on a limited set of agreed terms while each retains its own institutional knowledge. Agents can retain different views while making a bounded commitment together. A disagreement should remain visible rather than disappear behind a single confident answer.

When evidence, priorities or responsibilities change, the institution needs a way to explain an earlier decision, correct an error and revise future practice. Changing a rule also requires understanding which existing commitments it affects. Reverting a rule cannot necessarily undo actions already taken.

The mission develops this principle of accountable, plural institutions. The adaptive constitution programme proposes ways to investigate continuity through change.

05 / Terminology

Keep the distinctions clear.

Common ground
The shared meanings and commitments an institution currently relies on. It can contain explicit uncertainty and disagreement.
Grounding
Relating a proposal to relevant evidence, meanings and existing commitments so that support and conflict can be assessed.
Authority
A scoped right and responsibility to make a decision. Evidence supporting a proposal does not give its author that authority.
Commitment
An undertaking that others may rely on under agreed terms. A suggestion becomes a commitment only through the relevant procedure.
Institution
The maintained roles, rules, meanings and procedures through which participants coordinate and revise shared work.
Machine society
A wider setting in which autonomous agents and organisations with distinct purposes cooperate through institutions and shared agreements.
06 / Scope

A design to examine and test.

Etylon’s public materials set out a platform design and research agenda. The research briefs specify questions and proposed methods; they do not establish measured gains. Institutional knowledge is most relevant where shared decisions and dependencies justify the work of maintaining it. Simple, isolated tasks may need much less structure.

Institutional rules can be wrong, evidence can be incomplete and review can become burdensome. Explicit terms make these problems examinable; they do not eliminate them or establish the legitimacy of an institution.

Read the whitepaper for the full proposal and related work, or the cooperative programme for a proposed comparison of coordination benefits and costs. To explore a concrete setting, contact Etylon.