The Path to Agentic Engineering

Working Code

So far you've used Claude Code in a single session. Now let's use multiple sessions simultaneously.

Open 3 terminals and start Claude Code in each:

Terminal 1 — API development:

cd my-project
claude
> Implement the user authentication API

Terminal 2 — Testing:

cd my-project
claude
> Improve test coverage for the existing APIs

Terminal 3 — Documentation:

cd my-project
claude
> Write the API documentation

Each session works independently while sharing the same codebase. Use Git branches to manage conflicts.

Try It Yourself

Build an automation script with --print mode:

#!/bin/bash
# daily-review.sh

echo "=== Daily Code Review ==="
claude --print "Review the files changed since yesterday and summarize any issues"

echo "=== Test Status ==="
claude --print "Run the tests and analyze the cause of any failures"

echo "=== TODO Summary ==="
claude --print "Collect all TODO comments in the code and prioritize them"

Run this script every morning to automatically get a snapshot of your project's health.

"Why?" — From Coder to Orchestrator

Agentic engineering is the shift from writing code directly to directing AI agents to build systems.

Traditional vs Agentic Development

| Aspect | Traditional | Agentic | | ------------ | ------------------------ | ----------------------------------------- | | Writing code | Developer types directly | AI generates, developer reviews | | Debugging | Developer traces | AI analyzes, developer judges | | Testing | Developer writes | AI generates, developer validates | | Design | Developer decides | AI analyzes + suggests, developer decides | | Role | Implementer | Designer + reviewer |

Claude Code SDK

You can also call Claude Code programmatically:

import { claude } from "@anthropic-ai/claude-code";

const result = await claude({
  prompt: "Analyze security vulnerabilities in this file",
  workingDirectory: "./my-project",
});

console.log(result.stdout);

This enables integration with CI/CD pipelines, monitoring systems, and automation workflows.

Principles for Effective Agentic Workflows

1. Decompose tasks — break large tasks into independent units for parallel processing
2. Create verification points — mix automation and manual review appropriately
3. Manage context — maintain consistency with CLAUDE.md, custom commands, and MCP
4. Keep things reversible — ensure safety nets with Git branches and checkpoints
5. Automate incrementally — when you spot repeating patterns, turn them into commands or scripts

Next Steps

Here's a summary of the techniques covered in this course:

Basics:   Install → Basic commands → Context management
Daily:    Code writing → Debugging → Testing → Git
Setup:    CLAUDE.md → Custom commands → Multi-file
Advanced: Plan/Think → MCP → GitHub automation
Extend:   Hooks → Skills → Custom agents → Worktree → Headless/SDK
Practice: Project build → Agentic workflow

Combine these tools to create your own development workflow. There's no single right answer — the key is finding what works for your project and your team.

Deep Dive