🧠 Athena

certified-claude-architect-masterclass-2026-glossary

10 min read

Glossary: Certified Claude Architect Masterclass 2026

This glossary defines key terms from the “Certified Claude Architect Masterclass 2026” course, focusing on agentic architecture and planning.

Task Decomposition

Definition: Breaking down a complex, high-level goal into a series of smaller, more manageable subtasks. Sources: Related Notes:

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:

Agentic Systems

Definition: Systems that can autonomously pursue goals by perceiving their context, selecting actions, invoking tools, and iterating on the results until the goal is completed. Sources:

Assume and Process

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

Authority Boundaries

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

CCAF Framework

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

Chatbots

Definition: Systems that can generate human-like text responses in a conversation but are fundamentally stateless and cannot take actions. Sources:

Context Capacity

Definition: The amount of information that a language model can hold in its context window at one time. Sources:

Context Isolation

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

Dynamic Replanning

Definition: The process of revising a plan mid-execution in response to a failed step, unexpected tool output, or a change in the environment’s state. Sources:

Error Threshold

Definition: A form of loop control where an agent is programmed to stop its execution loop after a certain number of errors have occurred. Sources:

Escalation (Agentic)

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

Evaluator Optimizer

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

False Parallelism

Definition: An anti-pattern where tasks that have a hidden dependency on each other are treated as independent and run in parallel. Sources:

Four-Phase Lifecycle of a Tool Call

Definition: The process an agent follows to use a tool: Decision, Execution, Observation, Feedback. Sources:

Goal Drift

Definition: A risk in adaptive planning where an agent strays from its original goals while revising its plan. Sources:

Handoff Messages

Definition: Structured payloads passed between agents, containing the necessary context for a sub-agent to perform its task. Sources:

Hub-and-Spoke Topology

Definition: A multi-agent orchestration pattern where a central orchestrator routes all tasks to peripheral expert agents. Sources:

Hybrid Topologies

Definition: Multi-agent orchestration patterns that combine different patterns (e.g., Hub-and-Spoke, Pipeline, Peer-to-Peer) to leverage the strengths of each. Sources:

Incremental Complexity

Definition: A design principle that emphasizes starting with the simplest possible architecture and only adding complexity when necessary. Sources:

Independent Testability

Definition: A characteristic of a well-designed sub-agent, meaning it can be tested in isolation. Sources:

Infinite Loops

Definition: A failure mode where an agent gets stuck in a cycle of repetitive actions that yield no meaningful progress. Sources:

Instruction Design Principles

Definition: Principles for designing instructions for sub-agents, including narrow scope, precise output formats, and explicit success criteria. Sources:

Iterative Refinement Loops

Definition: Generate-evaluate-refine cycles for closing quality gaps in an agent’s output. Sources:

Merging Strategies

Definition: Methods used at the fan-in stage to combine parallel outputs (e.g., voting, concatenation, structured aggregation). Sources:

Multi-Agent Topologies

Definition: Patterns of communication and coordination between agents in a multi-agent system. Sources:

Open-ended Goals

Definition: Objectives where the target is known, but the exact steps to achieve it cannot be fully pre-specified at design time. Sources:

Orchestrator Sub-Agent

Definition: A workflow pattern where a main orchestrator delegates a complex sub-task to another, more specialized orchestrator. Sources:

Over-Engineering

Definition: The anti-pattern of introducing unnecessary architectural complexity. Sources:

Over-specified

Definition: A handoff message that contains extraneous information not relevant to the sub-agent’s task. Sources:

Peer-to-Peer (P2P) Topology

Definition: A multi-agent orchestration pattern where agents communicate directly with each other without a central coordinator. Sources:

Perception, Reasoning, Action, and Observation (PRAO)

Definition: The fundamental process (agentic loop) that enables an agent to operate autonomously. Sources:

Pipeline Topology

Definition: A multi-agent orchestration pattern where agents are arranged in a linear sequence. Sources:

Progress vs. Spinning

Definition: In an agentic loop, progress is marked by new observations, while spinning consists of repetitive actions with no meaningful results. Sources:

Prompt Chaining

Definition: The simplest workflow pattern, where the output of one LLM call is passed directly as input to the next. Sources:

Replanning Triggers

Definition: Events that necessitate a revision of the current plan. Sources:

Result Aggregation

Definition: A core function of an orchestrator agent to validate, synthesize, and handoff results from sub-agents. Sources:

Role of the Orchestrator

Definition: To manage tasks and synthesize results in a workflow or agentic system. Sources:

Routing (Workflow Pattern)

Definition: A workflow pattern that uses a classifier to sort inputs before processing. Sources:

Schema Versioning

Definition: The practice of assigning version numbers to handoff message schemas to ensure forward compatibility. Sources:

Separation of Responsibilities (Agentic)

Definition: The principle of dividing the work between Coordination (the orchestrator) and Execution (the sub-agents). Sources:

Sequential Execution

Definition: A workflow where tasks are performed one after another in a fixed, deterministic order. Sources:

State Accumulation

Definition: A condition in sequential pipelines where each step builds on the context or state produced by the earlier steps. Sources:

Stop Signal

Definition: A form of loop control where a command from the orchestrator or a human user instructs an agent to cease its operations. Sources:

Subagent Definition

Definition: A distinct agent instance created by an orchestrator for a specific, delegated task. Sources:

Task Assignment

Definition: A core function of an orchestrator agent that involves routing and matching subtasks to the appropriate sub-agents. Sources:

Task Completion

Definition: A termination condition for an agentic loop, where the agent stops its execution because its goal has been successfully achieved. Sources:

Termination Conditions

Definition: Explicit criteria that indicate when an agent should end its execution loop. Sources:

Tool Use (Agentic)

Definition: The mechanism by which an agentic system takes action in the real world. Sources:

Turn Budgets

Definition: A strict limit on the number of iterations an agent can execute. Sources:

Under-specified

Definition: A handoff message that lacks the necessary context for a sub-agent to perform its task correctly. Sources:

Unpredictable Tool Sequences

Definition: When the specific series of tools an agent needs to call cannot be known at design time. Sources:

Well-Defined Subtasks

Definition: A subtask with a single responsibility, defined inputs, a bounded scope, and clean, structured outputs. Sources:

Definition: The critical boundary in a multi-agent workflow where control and context are passed from one agent to another. Sources:

Orchestrator

Definition: A controlling agent or service responsible for managing the entire workflow, including interpreting the goal, decomposing tasks, and coordinating execution. Sources:

Monolithic Task

Definition: A task where one agent handles everything from start to finish. It’s simple to assign but hard to debug and difficult to parallelize. Sources:

Decomposed Task

Definition: A task that has been split into well-defined, independently retryable, delegatable, and testable subtasks. Sources:

Hierarchical Decomposition

Definition: A pattern of recursively breaking down goals into sub-goals, creating a tree structure where each level manages its own tier of complexity. Sources:

Sequential Decomposition

Definition: A pattern where tasks are executed in a defined linear order, with each task’s output serving as the input for the next. Sources:

Parallel Decomposition

Definition: A pattern where independent tasks are run simultaneously to reduce overall execution time, requiring a synchronization step to converge results. Sources:

Sequential Pipeline

Definition: An agentic workflow pattern where multiple specialized agents are arranged in a linear chain, processing tasks in a fixed, predetermined order. Sources:

Dependency Graph

Definition: A directed acyclic graph (DAG) used to model the relationships between subtasks, where nodes represent tasks and edges represent true dependencies. Sources:

Error Cascade

Definition: A critical failure mode where a minor, localized error in one agent propagates and amplifies through a chain of subsequent agents. Sources:

True vs. Artificial Dependency

Definition: A true dependency is a structural requirement for one task to have the output of another. An artificial dependency is a sequence imposed by convention rather than necessity. Sources:

Over-decomposition

Definition: An anti-pattern where a task is broken down into an excessive number of granular subtasks, causing coordination overhead to outweigh the benefits. Sources:

Under-decomposition

Definition: An anti-pattern where tasks are left too large and monolithic, making them difficult to retry, delegate reliably, or debug. Sources:

Parallel Execution

Definition: An orchestration pattern where multiple independent tasks are run concurrently to reduce total end-to-end latency. Sources:

Fan-out

Definition: The process within a parallel execution pattern where a single task is split into multiple, independent subtasks that are executed concurrently. Sources:

Fan-in

Definition: The synchronization step that follows a fan-out, responsible for collecting, handling failures, ordering, and merging the results from parallel branches. Sources:

Static Planning

Definition: An approach where all steps in a plan are defined before execution starts. It is predictable and auditable. Sources:

Dynamic Planning

Definition: An approach where an agent adapts its plan during execution based on real-time observations and tool outputs. Sources:

Replanning

Definition: The process of revising the current plan mid-execution in response to a failed step, unexpected tool output, or a change in the environment state. Sources:

Ambiguous Goal

Definition: A goal that lacks a clear, single definition, leaving multiple valid interpretations for an agent to choose from. Sources:

Clarify First Strategy

Definition: A strategy for handling ambiguity where the agent pauses execution to request more information from a human before proceeding, used when the cost of a wrong assumption is high. Sources:

Graph View