Agentic Coding with Claude Code

The Shifting Role of the Developer

The developer's job is changing: away from writing every line by hand and toward defining requirements, orchestrating agents and reviewing the code they produce. This chapter frames that shift, explains what an agentic coding tool is and positions Claude Code within the broader landscape of available tools.

• From Code Writer to Analyst, Agent Architect and Reviewer
• From autocomplete and chat to Agentic Coding
• The agentic loop: reason, use tools, observe, iterate
• Agentic coding tool landscape: Claude Code, GitHub Copilot, Cursor and OpenAI Codex

Introduction to Claude Code and LLMs

This module explains how Large Language Models work, the difference between static training data and dynamic context, and how Claude Code sits as an agentic orchestration layer on top of these models. Learn where Claude Code runs, how to choose the right model for the job, and how to stay cost- and token-aware while understanding its capabilities and limitations.

• Understanding Large Language Models
• LLM Static Knowledge vs. Dynamic Knowledge
• Training data cut-offs and the Context Window
• Where it runs: terminal CLI, IDE extensions, desktop and web
• Model Selection: Choosing between Opus, Sonnet and Haiku
• Plans, pricing and cost-control habits
• Understanding privacy and security considerations
• LAB: Installing and authenticating Claude Code and a first run

Fundamentals of Claude Code

In this module, we focus on the out-of-the-box experience: how to interact with Claude Code in an interactive session, how to use its built-in tools, and how to manage its most important resource - the context window - so it keeps performing as work grows.

• Keybindings and the slash command palette
• Built-in tools
• Feeding context
• Resetting and compacting context
• Checkpointing and rewinding conversations
• The permission model
• LAB: Complete a guided first task

AI Across the SDLC: Analysis and Design

Discover how Claude Code transforms every phase of the Software Development Lifecycle (SDLC), from analysis through verification. This module focuses on turning fuzzy requirements into a clear, agent-readable plan the agent can build on, using plan mode and extended thinking before any code is written.

• Scoping a task and extracting requirements with Claude
• Letting Claude interview to author a self-contained spec
• Asking questions to onboard to unfamiliar code
• Plan mode and Extended Thinking
• Reviewing and editing the plan before execution
• LAB: Designing the E-Commerce Platform: specs and plan

Integrations with MCP and CLI Tools

In this module, you'll learn how to extend Claude Code by connecting it to external systems through the Model Context Protocol (MCP), a standard way to connect AI models to tools and data sources. We'll also explore how command-line (CLI) tools can be used to enhance the agent, and compare this approach with MCP-based integrations.

• What MCP is and how it fits the agentic loop
• Connecting and scoping MCP servers in Claude Code
• Common MCP servers: GitHub, databases, Figma and browser control
• MCP Server Security Considerations
• Tool search and keeping MCP context cost low
• The Skill plus CLI tools pattern as an efficient alternative
• LAB: Create a modern UI using Claude Code and the Playwright CLI

Skills, Agents, Dynamic Workflows and Hooks

This module shows how to tailor Claude Code to your team's way of working. You'll learn how to use CLAUDE.md to apply shared standards and project memory, Skills to standardize recurring workflows, Subagents to delegate work to isolated contexts, and Hooks to enforce guardrails that an instruction alone cannot provide. Finally, when a task needs more agents than a single conversation can coordinate, you'll see how dynamic workflows move the plan into a script Claude writes and a runtime runs in the background.

• Project memory and instructions with CLAUDE.md
• Defining reusable knowledge and workflows with Skills
• Using pre-built skills, plugins and marketplaces
• Subagents and custom agents for isolated contexts
• Parallel workflows with Git worktrees and agent teams
• Automating clean-up and guardrails with Hooks
• Orchestrating agents at scale with dynamic workflows
• When to use CLAUDE.md vs. Skills vs. Agents vs. Hooks vs. Workflows?
• LAB: Building custom CLAUDE.md, Skills, Subagents and Hooks

Automating Claude Code in CI/CD

Claude Code is not limited to interactive sessions: it can run non-interactively as part of your automation. This module shows how to drive Claude Code in headless mode and wire it into CI/CD pipelines such as GitHub Actions and Azure Pipelines, so it can review pull requests, triage issues and perform scheduled maintenance without a developer at the keyboard.

• Headless mode: running Claude Code non-interactively and output formats
• Authenticating in CI
• Controlling tools and permissions non-interactively
• GitHub Actions: @claude mentions, issue-to-PR and automated PR review
• Running Claude Code in Azure Pipelines
• Use cases: review, triage, documentation and scheduled jobs
• LAB: Build a pipeline that runs Claude Code

AI Across the SDLC: Implementation and Testing

Once the plan is ready, this module shows how to let the agent handle the implementation. You'll learn how to give Claude room to run with auto mode and sandboxing, and how to use automated tests, code quality checks, and command-line tools to give the agent clear feedback, helping it produce high-quality code instead of making assumptions or guesses.

• Executing the plan with Claude Code
• Auto mode and permission allowlists
• Sandboxing for unattended runs
• Using test suites as guardrails for the agent
• The Test-Driven Development loop with Claude
• Verification signals: tests, builds, linters and screenshots
• Gating completion with /goal conditions and a Stop hook
• LAB: Building and testing the E-Commerce Platform with Claude Code

AI Across the SDLC: Reviewing and Verifying AI-Generated Code

AI can generate a lot of code in a short time, but that doesn't mean the code is always correct. This module teaches you how to review and validate AI-generated code, spot bugs and incorrect suggestions, and use automated tools - including subagents dedicated to review - to verify that the code works as intended before it is deployed.

• Reviewing large AI-produced diffs efficiently
• Static analysis and security scanning as objective review signals
• Verifying behavior with tests, Playwright and benchmarking tools
• Steering the agent to produce reviewable, incremental changes
• LAB: Reviewing, verifying and hardening the E-Commerce Platform

Building Custom Applications with the Claude Agent SDK

Everything so far has driven Claude Code interactively or through CI. The Claude Agent SDK exposes the very same agent loop, tools, and context management as a programmable library for TypeScript and Python, so you can embed Claude in your own scripts, pipelines, and applications, coordinating tools and MCP servers to build intelligent solutions that go beyond the IDE and command line.

• The Agent SDK as a TypeScript and Python library
• The agentic core and execution loop
• Configuring agent runs and authenticating
• Giving the agent custom tools
• Connecting external MCP servers programmatically
• Controlling autonomy
• Use cases: batch automation, CI/CD and building your own agentic apps
• LAB: Build a small custom agent with the Agent SDK