Agent Loop

The canonical cognition cycle for EcoEchoSystem agents#

The agent loop defines how an agent experiences the world, updates itself, and acts back upon the system.
It is not a decision tree. It is a bounded, recursive process shaped by identity, learning, and social context.

Agents do not optimize outcomes.
They navigate constraints over time.

Purpose#

This module exists to:

define a reusable cognition cycle for all agent types
integrate perception, identity, learning, and interaction
enforce bounded rationality and temporal limits
enable feedback between agents and simulation layers
prevent scripted or omniscient behavior

The agent loop is the engine of emergence.

Canonical Agent Loop Overview#

Each agent executes the following loop at every simulation step:

Perceive
Interpret
Evaluate
Decide
Act
Learn
Update Identity
Update Social State

The loop is recursive and lossy — information degrades, bias accumulates, and timing matters.

1. Perception#

Agents receive signals from:

environment (resources, threats, opportunities)
institutions (rules, enforcement, legitimacy cues)
other agents (signals, behavior, reputation)

Perception is filtered by:

attention limits
salience bias
identity relevance

Agents never perceive the full state.

2. Interpretation#

Perceived signals are interpreted through:

existing beliefs
narrative identity
group affiliation
recent memory

Interpretation answers:

What does this mean for someone like me?

Meaning precedes accuracy.

3. Evaluation#

Agents evaluate options using:

bounded utility proxies (safety, status, belonging, purpose)
risk tolerance shaped by stress
learned heuristics
social expectations

Evaluation is comparative, not absolute.

4. Decision#

Agents select an action based on:

perceived best‑fit option
identity consistency
coordination expectations
time pressure

Decisions may be:

habitual
reactive
exploratory
defensive

Perfect rationality is impossible by design.

5. Action#

Actions may include:

resource use
communication
cooperation or conflict
institutional compliance or defiance

Actions modify:

local environment
social networks
institutional state

Every action feeds back into the system.

6. Learning#

Agents update internal models based on:

outcome feedback
social reinforcement
stress response

Learning follows the learning curves module:

uneven
delayed
identity‑constrained

Failure to learn is common.

7. Identity Update#

Identity is updated when:

actions conflict with self‑model
narratives fail
stress exceeds tolerance

Identity change may be:

gradual
defensive
fragmentary
crisis‑driven

Most loops reinforce identity rather than change it.

Agents update:

trust levels
reputation assessments
group alignment
coordination readiness

Social state shapes the next perception cycle.

Temporal Dynamics#

The agent loop operates across multiple time scales:

micro‑steps (attention, reaction)
meso‑steps (learning, trust)
macro‑steps (identity, institutional alignment)

Not all updates occur every tick.

Agent Types and Loop Variants#

The same loop applies to:

individuals
groups
institutions

Differences arise from:

memory depth
learning rate
identity inertia
action scope

Institutions loop slower but act wider.

Failure Modes#

Agent loop modeling fails when:

agents see everything
decisions ignore identity
learning is instantaneous
actions lack consequence

If outcomes feel scripted, the loop is broken.

Integration Notes#

The agent loop:

consumes identity, learning, and social modules
feeds city and civilization simulations
enables AI‑guided exploration
preserves substrate coherence

This loop is the heartbeat of EcoEchoSystem cognition.

Status#

Canonical agent loop for cognitive agent simulation.
Designed for extensibility, realism, and emergent behavior.