Agentic Architecture & Orchestration
Key terms and definitions for Domain 1. Each entry includes a concise definition, exam context, and links to the relevant lesson.
12 of 12 terms
Agentic Loop
A repeating cycle where Claude receives input, reasons about it, optionally invokes tools, observes results, and decides whether to continue or stop. The loop is governed by stop_reason values returned by the API.
Exam Context
Know the three stop_reason values (end_turn, tool_use, max_tokens) and how each affects loop continuation. Expect questions on when to break the loop versus when to continue.
Claude Agent SDK
A Python framework for building production agentic systems with Claude. Provides primitives for agents, tools, guardrails, handoffs between agents, and tracing. Designed for multi-agent orchestration with built-in safety features.
Exam Context
Know the core primitives: Agent, Runner, tool decorator, guardrail decorator, and handoff. Understand how Runner.run() manages the agentic loop internally.
end_turn
A stop_reason value indicating Claude has voluntarily finished its response. In an agentic loop, this signals that Claude believes the task is complete and no further tool calls are needed.
Exam Context
Distinguish from max_tokens (involuntary stop). In agentic loops, end_turn is the primary signal to exit the loop.
Error Recovery
Strategies for handling failures in agentic systems, including retry with exponential backoff, graceful degradation, fallback tools, and providing error context back to Claude so it can self-correct.
Exam Context
Know that feeding tool errors back to Claude often enables self-correction. Understand retry strategies and when to escalate versus retry.
Fan-Out/Fan-In
A parallelization pattern where a task is split into independent subtasks (fan-out), each processed concurrently, and results are aggregated (fan-in). Reduces latency for tasks with independent components.
Exam Context
Know when parallelization is safe (independent subtasks with no shared state) versus when sequential processing is required (dependent steps).
Guardrail
A safety mechanism that constrains Claude's behavior within defined boundaries. Guardrails can operate on input (pre-processing validation), output (post-processing filtering), or tool use (restricting which tools can be called and with what parameters).
Exam Context
Understand the difference between input guardrails, output guardrails, and tool-use guardrails. Know that guardrails in the Claude Agent SDK run as parallel checks that can halt execution.
Human-in-the-Loop
A pattern where the agentic system pauses execution to request human approval before proceeding with sensitive operations. The human can approve, reject, or modify the proposed action.
Exam Context
Understand when human-in-the-loop is required (irreversible actions, high-stakes decisions) versus when it adds unnecessary friction. Know how to implement approval gates in the agentic loop.
Multi-Agent System
An architecture where multiple specialized agents collaborate on a task. Each agent has its own system prompt, tools, and guardrails. Agents communicate through handoffs or a central orchestrator.
Exam Context
Know when to use multi-agent versus single-agent. Multi-agent adds complexity; use it when tasks require fundamentally different tool sets or expertise domains.
Orchestration Pattern
An architecture for coordinating multiple Claude API calls or agents. Common patterns include prompt chaining (sequential), routing (conditional dispatch), parallelization (fan-out/fan-in), and evaluator-optimizer loops.
Exam Context
Be able to identify which pattern fits a given scenario. Routing is for classification-based dispatch; chaining is for sequential transformation; parallelization is for independent subtasks.
Routing Pattern
An orchestration pattern where an initial classification step determines which specialized agent or prompt chain handles the request. Useful when different input types require fundamentally different processing.
Exam Context
Routing is the right choice when you have distinct categories of input. Use a fast, cheap classification call to route before invoking specialized (potentially expensive) processing.
stop_reason
A field in the Claude API response indicating why generation stopped. Values: end_turn (Claude finished voluntarily), tool_use (Claude wants to call a tool), max_tokens (output limit reached).
Exam Context
Critical for agentic loop implementation. tool_use means continue the loop with tool results. end_turn means the task is complete. max_tokens may require continuation logic.
tool_use
A content block type in the Claude API response indicating Claude wants to invoke a tool. Contains the tool name, a unique ID, and the input parameters matching the tool's JSON schema.
Exam Context
Know the structure: each tool_use block has id, name, and input fields. The tool result must reference the same id when returning results to Claude.