Agentic Systems Glossary

Last updated: 2026-06-24

Action Environment Gap

Definition: The difference between what a language model can describe in text and what it can actually execute in a real-world environment. Sources:

Related Notes:

Agent State Checkpointing

Definition: A technique to preserve state at critical points to prevent full workflow restarts after failures. Sources:

Related Notes:

Agent Substitutability

Definition: A design principle where agents conforming to a shared protocol can handle tasks interchangeably, promoting modularity. Sources:

Related Notes:

Agentic Systems

Definition: Systems that autonomously pursue goals by perceiving context, selecting actions, invoking tools, and iterating on results. Sources:

Related Notes:

Ambiguous Goal

Definition: A high-level objective lacking specific definition, creating multiple valid interpretations and misalignment risks. Sources:

Related Notes:

Assume and Process Strategy

Definition: A strategy for handling ambiguity where an agent makes a reasonable inference and continues execution without human clarification. Sources:

Related Notes:

Audit Logging

Definition: Structured logging of agent handoff events to provide the observability needed to diagnose failures. Sources:

Related Notes:

Authority Boundaries

Definition: The scope of a sub-agent’s permissions and access, guided by the Principle of Least Privilege. Sources:

Related Notes:

CCAF Framework

Definition: A method to mitigate risks associated with designing and building agentic systems. Sources:

Related Notes:

Chatbots

Definition: Stateless conversational systems that cannot perform actions in the real world. Sources:

Related Notes:

Clarify First Strategy

Definition: A cautious approach to handling ambiguity where an agent pauses execution to seek information from a human. Sources:

Related Notes:

Context Capacity

Definition: The amount of information (instructions, history, tool results) a language model can hold in its context window at one time. Sources:

Related Notes:

Context Isolation

Definition: A design principle where sub-agents start with a “blank” context and only see information explicitly handed over by the orchestrator. Sources:

Related Notes:

Decomposed Task

Definition: A sub-task broken down from a larger, more complex goal, designed to be independently retryable, delegatable, and testable. Sources:

Related Notes:

Dependency Graph

Definition: A directed acyclic graph (DAG) used to model the relationships between subtasks in a complex workflow, making the order of operations explicit. Sources:

Related Notes:

Deterministic Workflows

Definition: Fixed sequences of steps designed by developers, providing predictability and control. Sources:

Related Notes:

Dynamic Planning

Definition: The ability of an agent to adapt its plan during execution based on real-time observations from its environment. Sources:

Related Notes:

Dynamic Replanning

Definition: The process of revising a plan mid-execution in response to failures or environment changes. Sources:

Related Notes:

Error Propagation

Definition: The process of surfacing sub-agent failures back to the orchestrator to enable intelligent error management. Sources:

Related Notes:

Error Threshold

Definition: A safety mechanism that stops an agent’s execution loop after a certain number of errors occur. Sources:

Related Notes:

Escalation

Definition: A loop control mechanism where an agent passes a problem it cannot solve to a human operator or fallback system. Sources:

Related Notes:

Evaluator Optimizer

Definition: A workflow pattern using a generate-evaluate-refine cycle to close quality gaps in an agent’s output. Sources:

Related Notes:

False Parallelism

Definition: An anti-pattern where tasks with hidden dependencies are treated as independent and run in parallel, causing hazards. Sources:

Related Notes:

Four-Phase Lifecycle of a Tool Call

Definition: The fundamental mechanic of agent interaction: Decision, Execution, Observation, Feedback. Sources:

Related Notes:

Goal Drift

Definition: A risk where an agent revises its plan to overcome obstacles but strays from its original goals. Sources:

Related Notes: