Agent Ages

Compliance Traceability

The capability for every critical agent action to be mapped back to compliance requirements, rule bases, and approval records. Not post-hoc documentation, but automatic recording of 'why this was allowed' during execution. Regulatory audits can directly export evidence chains.

Legal Personhood Interface

An interface layer between agent systems and legal personhood frameworks. Defines whether and in what sense agents can be legally recognized, where responsibility boundaries lie, and what rights they hold. This is the必经之门 (necessary gateway) for agents transitioning from technical systems into social systems.

Knowledge Drift

The phenomenon where knowledge relied upon by agents gradually becomes mismatched with the real world. Usually not sudden failure, but slow accuracy decline. Trigger conditions include policy changes, API modifications, and business rule adjustments. Requires establishing 'knowledge validity period' mechanisms.

Capability Decay

Overall decline in execution quality, stability, and judgment accuracy after long-term agent operation. Causes include environmental changes, knowledge aging, and feedback poisoning. Requires periodic 'capability refresh' to counteract, rather than waiting for problems to occur before fixing.

Coordination Bandwidth

The total capacity for effective coordination between humans, agents, and systems. Many scaling failures are not due to insufficient compute, but coordination bandwidth exhaustion first - too many tasks requiring human approval, too many conflicts needing manual arbitration, too many states requiring synchronization.

Accountability Chain

A complete responsibility mapping structure from task initiation, authorization, execution, review, to result delivery. Each link must have explicit 'who made what decision at what time'. When problems occur, they can be traced to specific links, not vaguely attributed to 'the agent was wrong'.

Supervision Decay

The natural phenomenon where human supervision frequency decreases as agent performance stabilizes. Positive effect: efficiency improvement. Negative effect: expansion of risk blind spots. Requires proactive design of 'mandatory review checkpoints' to counteract decay, rather than relying on vigilance.

Progressive Authorization

Agent permissions are not granted all at once, but progressively expanded based on performance and verification results. Applicable to high-risk scenarios: start with read-only permissions, then write permissions after verification, finally autonomous permissions. Each upgrade requires explicit verification checkpoints.

Trust Delegation

The decision to hand over both execution authority and judgment authority to an agent for a task. The core question is not 'whether to trust', but 'under what conditions, which powers, and in what manner to delegate'. Once delegation is established, humans transition from executors to

Agent Router

A mechanism that dynamically assigns models, agents, or execution paths to different tasks based on risk, cost, capability, and context. Core questions: Which agent/model for which task? When is human intervention needed? When should degradation occur?

Kill Switch

A global switch that immediately stops agent execution in emergencies. Trigger conditions include: detected attacks, abnormal behavior, permission violations, budget exhaustion. Core requirements: must be faster than agent execution, must be independent of agent control.

Discoverable Agent API

A self-describing interface layer that agents can automatically understand, query, and invoke. Supports schema discovery, self-describing schemas, and GET /schema endpoints. Core goal: agents don't need to learn APIs in advance, they can discover and understand them at runtime.

Resource-Action Architecture

An architectural paradigm for agent-native systems: resources + actions, not UI flows + form steps. Examples: article/page/product/user + create/update/publish/delete. Agents understand resources and actions, not pages and buttons.

Agent Object Model

Core resource object abstractions in the agent world. Includes: task, state, evidence, capability, policy, scope, receipt, exception. Each object has explicit schema, lifecycle, and executable operations.

Agent Native Protocol

A native interaction protocol designed for intelligent agents, not making AI adapt to human UIs, but designing native Intent Interface, Action Protocol, and Object Model for agents. Core idea: agents should not click buttons, they should invoke intents.

Verifiable Agent Runtime

An agent runtime that is auditable, replayable, and verifiable. Every execution produces event logs, causal chains, and receipts that support post-hoc replay and verification. The core goal is not monitoring, but making agent behavior verifiable by third parties.

Agent Immune System

A defense system against prompt injection, context pollution, attack patterns, and cognitive drifting. Includes: input isolation, contamination detection, quarantine zones, cognitive circuit breakers, self-recovery, and immune memory. Core philosophy: assume attacks will happen, design defenses in advance.

AI Agent / Intelligent Agent

A software entity capable of perceiving goals, reasoning about tasks, and executing actions across digital systems. In the agent economy, agents transform intelligence from a passive advisory function into an active execution layer. This entry is part of the Agent Ages Concept Dictionary.

Agent Workflows

A structured sequence of tasks executed by one or more AI agents to transform inputs into outputs. Agent workflows replace many traditional organisational processes with programmable automation. This entry is part of the Agent Ages Concept Dictionary.

Workflow Asset

A repeatable, automated workflow that generates value independently of constant human labour. Workflow assets are the foundational productive units of the agent economy. This entry is part of the Agent Ages Concept Dictionary.

Agent Economy

An economic system in which intelligent agents participate directly in production, coordination, and service delivery, fundamentally altering labour, capital, and organisational structures. This entry is part of the Agent Ages Concept Dictionary.

Execution Collapse

The rapid decline in the cost and effort required to perform complex tasks due to automation through AI agents and programmable workflows. This entry is part of the Agent Ages Concept Dictionary.

Intelligence Capital (I)

A new form of capital representing deployable machine intelligence capable of performing tasks traditionally done by human labour. This entry is part of the Agent Ages Concept Dictionary.

Accountability Capital

The scarce economic value associated with human judgement, responsibility, and decision-making in systems where execution can be automated. This entry is part of the Agent Ages Concept Dictionary.

Agent Routing

A mechanism that dynamically assigns models, agents, or execution paths to different tasks based on risk, cost, capability, and context. Core questions: Which agent/model for which task? When is human intervention needed? When should degradation occur? This entry is part of the Agent Ages Concept Dictionary. Chinese version available.