Metaintelligence architectureMetaintelligence · Protocols · Relationships

A link as a first-class object: a relationship with state and history

In most systems the link between participants is invisible: there are calls, a queue, a shared channel. Make the relationship an object in its own right, with its own state and history, and things appear that do not exist otherwise: participants can reach an agreement, revise its terms and part cleanly.

Metaintelligence architecture · part 3 of 6

What is lost when the link is not an object

If an interaction comes down to a call, it has no memory. Every exchange starts from zero: who may do what, which context is needed, whether this pair works at all, everything is found out again. The system builds up no knowledge of its own structure.

A relationship as an object holds that knowledge: what the parties have agreed, how often it pays for them to communicate, what has already been passed on, how reliable the link is and whether it is worth keeping up.

What a link contains

Negotiation instead of assignment

A link comes into being by agreement between the parties; the environment cannot create it by order. The initiator proposes terms. The other side accepts, rejects, defers or makes a counter-offer, and then the roles swap. The result is an alternating negotiation protocol with a stored transcript.

The value of the transcript shows up later. When you need to understand why two participants work the way they do, the answer is in the history of their agreements and not in someone’s memory. Revising the terms goes through the same mechanism: renegotiating a link means taking the same path as concluding it.

Adapting to observed usefulness

The terms of exchange are not permanent. If a link brings little benefit, the parties lower the frequency and volume of exchange; if it brings a lot, they raise them. A link that has stopped being useful first weakens, then falls asleep, then retires.

It sounds like a biological metaphor, but the point is an engineering one. In a bounded environment, keeping a link alive costs resources, and weakening what is useless automatically keeps them from being spent on structure that no longer gives anything. Without such a mechanism a long-lived system accumulates links that nobody dares to delete.

Probation

A new link does not get full rights at once. It starts on probation: exchange goes on, but its volume is limited while reliability builds up. The link becomes a full one only after its usefulness is confirmed.

We apply the same principle to agent autonomy in applied systems: rights expand on accumulated statistics and are not granted in advance. The difference is that here the principle is built into the environment itself and does not depend on the policy of a particular project.

What this gives the system as a whole

  1. The system’s structure becomes observable: links can be listed, measured and explained.
  2. The design itself rules out conflicting edits; developers do not need an agreement for that.
  3. Ending a relationship becomes a routine operation. No code has to be deleted for it.
  4. The history of agreements replaces word-of-mouth knowledge of why everything works the way it does.
  5. The cost of maintaining the structure is visible, and it can be managed.

There is a limitation too. All of this is justified where relationships are numerous and long-lived. For a pair of components that exchange messages twice an hour, an ordinary queue is simpler and cheaper.

Code and artefacts

The “Metaintelligence architecture” series

  1. Metaintelligence: why the environment around the model sets a system’s ceiling
  2. The agent as a bounded local world
  3. A link as a first-class object: a relationship with state and history
  4. Living graph: topology shaped by work
  5. Consensus without a centre: deciding together without taking anyone’s word for it
  6. Agent economics: autonomy needs a budget

Frequently asked questions

How is this different from a message queue?

A queue carries data and knows nothing about the relationship between the parties. A link as an object holds the agreed terms, the rights to change the shared state, the revision history and its own estimate of usefulness. A queue can serve as the transport inside such a link, but it cannot replace it.

Would it not be simpler to define the link structure in advance?

It is simpler if the task is fully known and does not change. As soon as the set of tasks changes while the work goes on, a predefined structure starts getting in the way: nobody will create the links needed now, and nobody will delete the ones no longer needed.

How do you avoid endless negotiation?

Proposals have a lifetime and expire, and the number of counter-moves is limited. A deferred proposal comes back later; a rejected one closes the attempt. Negotiation here is built as a finite state machine, with no open-ended dialogue in it.

How we help with this

Read next

Follow the series

If you’re building a similar environment or want to discuss the research with us, write through the form on the homepage.

Email hello@xteam.pro