CCDV-F Deep Dive 11/25: Agent Construction with Claude (Domain 1, 5.3%) | Claude Certified Developer

Agent Construction with Claude explained for the CCDV-F exam: a Claude agent is a loop in which Claude calls tools and reacts to real results until the task is done. Choose the Client SDK to write the loop yourself (or use its beta tool runner), the Agent SDK to embed Claude Code's tools, permissions, hooks and sessions in a process you operate, or Managed Agents to have Anthropic host the loop. Always add stop conditions, and enforce permissions in code with deny rules, hooks and sandboxing, never in the prompt. Agent Construction with Claude carries 5.3% of the Claude Certified Developer Foundations (CCDV-F) exam weight, in Domain 1: Agents and Workflows (14.7%). This exam-prep deep dive is for developers preparing for the Claude Certified Developer Foundations (CCDV-F) exam who want more than the short Agent Construction with Claude lesson. It covers the agent loop, the four documented ways to build (Client SDK, Agent SDK, Managed Agents and the Claude Code CLI) and how to choose between them, what a hand-written loop must handle, stop conditions, custom loops versus the tool runner versus a full harness, how the Agent SDK loop and its result subtypes work, what the Agent SDK adds, the permission evaluation order and permission modes, hooks for deterministic actions, five design levers, and self-hosted versus Anthropic-hosted deployment, including Managed Agents self-hosted sandboxes. Features change, so check the current documentation. It is an independent study video, not official Anthropic material, and the practice questions are original. What you will learn: • Why an agent is a loop, and when an agent beats a workflow • The four ways to build: Client SDK, Agent SDK, Managed Agents, Claude Code CLI • What a hand-written tool loop must handle, and why stop conditions are mandatory • Custom loop, beta tool runner, or a full harness like the Agent SDK • How the Agent SDK loop works: turns, compaction, and result subtypes • Where permissions are enforced: the evaluation order and permission modes • Hooks for deterministic actions: PreToolUse, PostToolUse, Stop and more • Self-hosted versus Anthropic-hosted deployment, with a worked example CHAPTERS 00:00 Agent Construction with Claude deep dive for the CCDV-F exam: scope and exam weight 00:48 What is an AI agent: Claude using tools in a loop, agents versus workflows 01:36 Four ways to build: Client SDK, Agent SDK, Managed Agents, Claude Code CLI 02:25 Client SDK vs Agent SDK vs Managed Agents: how to choose a build option 03:11 Writing your own agent loop: stop_reason, tool_result, tool_use_id, step limits 03:57 Agent stop conditions: max turns, budget caps, success checks, human checkpoints 04:42 Custom agent loop vs beta tool runner vs Agent SDK harness 05:33 How the Claude Agent SDK loop works: turns, compaction, and result subtypes 06:25 Claude Agent SDK features: built-in tools, hooks, subagents, MCP, sessions 07:18 Agent SDK permission evaluation order: hooks, deny rules, modes, canUseTool 08:08 Claude Agent SDK permission modes: default, dontAsk, acceptEdits, plan, auto, bypass 08:58 Agent SDK hooks for deterministic actions: PreToolUse, PostToolUse, Stop 09:50 Agent design levers: instructions, tool descriptions, permissions, context, stop conditions 10:43 Self-hosted vs Anthropic-hosted agents: Agent SDK, Managed Agents, self-hosted sandboxes 11:34 Worked example: choosing the Agent SDK, Managed Agents, or the Client SDK 12:22 Practice questions and answers: Agent Construction with Claude CCDV-F deep-dive series: video 11 of 25. Each deep dive goes further than the matching short lesson in the CCDV-F full course; watch that lesson first if the topic is new. Previous: CCDV-F Deep Dive 10/25, Model Selection and Trade-offs:    • CCDV-F Deep Dive 10/25: Model Selection an...   Go deeper on this channel: CCDV-F 11/25: Agent Construction with Claude (Domain 1, 5.3%) | Claude Certified Developer:    • CCDV-F 11/25: Agent Construction with Clau...   Exam format: 53 multiple-choice and multiple-response questions, 120 minutes, pass mark 720 of 1000 (scaled). Weights and format come from third-party summaries of Anthropic's CCDV-F Exam Guide v1.0 (July 2026); confirm them against Anthropic's official guide before you sit the exam. Subscribe to Core Concept Learning for clear, visual explanations of AI and software engineering concepts. #CCDVF #ClaudeCertifiedDeveloper #ClaudeAgentSDK