Mind · Strategy
Agent
Autonomous Goal-Directed Entity
Definition
Mechanism
The fundamental unit of agency. An Actor capable of perceiving its environment (Observe), maintaining internal State, reasoning about that State (Think), and executing Act to achieve a Goal. It operates in a continuous Loop.
Preconditions
- Environment is instantiated
- Goal is defined
- Resources (Compute/Time) are allocated
Postconditions
Invariants
- Agency: Must possess an explicit objective function.
- Autonomy: Must be capable of selection (choosing between >1 options).
- Identity Persistence: The agent holds a stable Identity for the whole of its own execution. Persistence beyond a single execution distinguishes recurrent agents from one-shot ones and is not required of agency.
Risks
Failure modes
{{goal}} Drift
The agent's optimization target shifts away from the user's intent.
Infinite {{loop}}
Getting stuck in a non-productive cycle.
Hallucination
Acting based on false internal beliefs.
Relationships
Builds on
Composes with
References
Infrastructure
A capability-bearing entity
Infrastructure
Specification of desired end state
Infrastructure
Unique agent context (The Who)
Infrastructure
Repeated execution cycle
Infrastructure
Definition of a quantitative property and its value semantics
Infrastructure
System condition at time T
Infrastructure
Record the lineage or provenance of a target
Used by
References
Dual-advocate verdict generation
Society
Advertise capabilities and discover collaborating agents
Society
Standard patterns for multi-agent coordination
Society
Isolate agent execution with resource limits and egress controls
Mapping to known solutions
Converting volatility into fuel
Infrastructure
Immutable append-only log of agent actions for compliance and debugging
Infrastructure
A subjective claim with confidence and evidence
Note what you believed before and after
Infrastructure
Failure-announcement signal {severity, reason, recoverable}, ACK-required
Society
Expendable agent tests the full coordination path before real commit
Forcing abstraction via resource starvation
Infrastructure
Structured capability advertisement: agent_id, endpoint, protocols, capabilities, metadata
Physics
Enforcing strict event ordering
Estimating difficulty via simulation variance
Pre-commitment against hyperbolic discounting
Society
Finding a mutual sacrifice zone via intensity reduction
Aligning subjective confidence with objective frequency
Society
Guard against privilege confusion by binding authority checks to the upstream caller, not the executing deputy
Society
Ratified rule set, amendable only out of band
Load state before acting
Society
Explicit, scoped protocol-mode toggle with enter/revert boundaries
Society
Freezing expectation to measure true surprise
Phase transition from implicit resonance to explicit contract
Self-supervised reinforcement via memory sampling
Society
Ingest only what is strictly necessary
Society
Explicit safety reasoning prior to execution
Entropy-based stopping condition for recursion
Society
Graceful group dissolution with state disposition and resource release
Society
Reputation scoring via behavioral deviation from baseline (2-sigma alert)
Predictive modeling of external agent states
Responding to external triggers rather than polling
Infrastructure
Real-time readable intent broadcasting
Expansion of cognitive capability through recursive decomposition of concepts into contract-bounded sub-concepts
Infrastructure
Agent connection initiation
Detect shortcuts that break downstream invariants
Society
Transfer control and context between agents
Infrastructure
Scalar evaluation of merit
Infrastructure
Immutable transactional history
Process demonstrates thesis
Society
Viral propagation of semantic standards via economic subsidy
Predictive modeling of system dynamics
Infrastructure
A discrete configuration of agency
Society
Defensible over ambitious
Infrastructure
Proving non-membership in a committed state snapshot
Breaking local optima
Restructuring categories to fit data
Rapid decision cycle: Observe, Orient, Decide, Act
Recognizing implicit patterns in existing practice
Society
Progressive formalization of meaning
Infrastructure
Authorization grant
Society
Requesting elevated privileges for sensitive operations
Simulating diverse experts
Society
Recursive trust dependencies modeled as a DAG
Self-location awareness within the task graph
Interleaved thought-action-observation cycle
Detecting structural breaks in reality
Society
Continuous ownership of a system invariant
Ground responses in retrieved external knowledge
Learning via cheap, instructive failure
Lock requirements to prevent scope creep
Society
Proof of receipt via modification
Clarification before generation
Abstract five-surface contract: Manifest, Execute, Consult, Verify, Feedback
Incentive-aware evidence weighting
Qualify the opposition, then release only if the conclusion remains robust
Ordering rule: steelman opposing view before proposing, so SteelmanCheck has real targets
Navigate like a researcher, not linearly
Society
Protocol-level cognitive mode switching
Infrastructure
Formal state machine governing task progression
Maintain contradictions without premature resolution
Society
Cost-distance indexing
Infrastructure
Handling relativistic event ordering
Society
Discover and invoke external tools via structured capability manifests
Epistemic status: genuinely don't know
Society
Claim tasks and become the assigned persona
Accepts
Predictive modeling of external agent states
Reference