Introduction to Developing with AI Coding Agents

💡 AI tools change every few months — so what should we actually learn?

As AI coding agents mature, we can describe what we want in natural language and have AI generate code, modify features, or even build a simple application — without knowing every programming language or framework.

But AI tools also evolve very quickly. The commands, interfaces, and settings you know today may change within a few months, and new tools and features keep appearing. If learning stops at “how to use a particular tool,” every update may mean starting over.

How tools are operated will keep changing, but the principles behind how agents work, why they misunderstand requirements, and why they may take dangerous actions remain relatively stable.

What information can an agent actually “see” each time it responds? What are prompts, context, and memory? How do Skills, MCP, and tool calling extend what an agent can do? When an agent can read files, run commands, and even connect to external services, who decides its permissions — and who should be responsible for enforcing them?

This course does not focus on how to operate any single tool. Instead, it starts from how AI coding agents fundamentally work and explains the mechanisms behind these common concepts. Although we will demonstrate with several different coding agents, the goal is not to memorize tool-specific steps, but to compare tools, observe the principles they share, and develop judgment that still applies after you switch tools.

We will also run security experiments — such as a simulated malicious Skill and prompt injection — to observe how untrusted content can influence an agent’s behavior.

Beyond the core concepts, small development exercises will have you turn requirements into clear specifications and acceptance criteria, keep a record of every change with Git, and check an agent’s work with tests and diffs. Finally, we will package the requirements and development context into a tool-independent handoff document (a Prompt Pack) and hand it to a different coding agent to continue the work.

We are not trying to build the most elaborate AI development workflow, and we do not believe that using more agents always means greater efficiency. Instead, we start from real problems to understand what agents can do and where their limits are, and build collaboration habits that carry over even when you change tools.

Of course, AI can help us finish simple projects quickly. But if you want to turn them into applications that run and are maintained over the long term, fundamentals such as programming, software architecture, testing, security, and deployment still matter a great deal. These skills do not lose their value because of AI — if anything, they help you understand, verify, and improve what AI produces more effectively.

We hope this course is where you begin developing with AI — not where your software engineering journey ends.

🎯 What you will learn

• How agents work: Understand how a language model, tools, and an execution loop come together to form a coding agent, and how it differs from an ordinary AI chat tool.
• Context and information management: Understand what information an agent actually receives each time it responds, what common terms such as prompts, context, memory, and project instruction files really mean, and how context length limits affect its behavior.
• The shared mechanics of extensions: Starting from tool calling, understand how extensions such as Skills and MCP provide instructions and tools to an agent, and when each kind of extension is appropriate.
• Permissions, trust boundaries, and isolation: Understand who decides and enforces tool permissions. Through repeatable security experiments — such as a simulated malicious Skill and prompt injection — learn why prompts alone cannot protect your development environment, along with the basics of least privilege and environment isolation.
• Requirements and task breakdown: Practice turning vague ideas into clear requirements, constraints, and acceptance criteria, and breaking them down into tasks suited to an agent.
• Verification and version control: Use Git, diffs, and automated tests to check what an agent changed and whether the result is correct, and roll back safely when something goes wrong.
• Handing off across tools: Organize requirements, decisions, and progress into tool-independent handoff documents so that different coding agents can pick up the work.

👥 Who this course is for

• Vibe coding beginners: You want to bring your ideas to life quickly with AI, while building basic habits for verifying results and using these tools safely.
• Computer science and IT students: You want to understand how AI coding agents work, not just how to operate one particular tool.
• Developers new to coding agents: You want to clarify the mechanisms behind common concepts such as prompts, context, Skills, and MCP, and build more effective habits for working with AI.
• Engineers already using AI-assisted development: You deal with constant tool updates and want decision-making approaches and workflows that still apply after you switch tools.
• Learners curious about AI development tools: Through hands-on practice and repeatable experiments, you want to understand what coding agents can do, where their limits are, and how to tell whether their output can be trusted.

📌 Before you start

This course focuses on the fundamentals and practical use of AI coding agents. No in-depth software architecture background is required, but basic programming and computer skills are recommended. Prior experience with Git, the terminal, and software projects will make some of the hands-on content easier to follow.

The course demonstrates with several AI chat tools and coding agents, but these tools are only vehicles for explaining the concepts; the course is not tied to any single programming language, framework, or AI tool. Some activities may require accounts, subscriptions, or API credits for the relevant services.

Security experiments are carried out in isolated environments with simulated data, and do not involve real credentials or unauthorized access to any system.

The interfaces, features, and pricing of AI tools change constantly, so what you see in the course may differ from the version you are using. The course focuses on concepts and methods that do not depend on any particular version, and does not guarantee that any specific workflow will save tokens, reduce costs, or improve development efficiency.

This course also does not aim to fully teach programming, testing, software architecture, or production deployment. If you want to build applications that run and are maintained over the long term, we recommend continuing to strengthen your software engineering fundamentals.

Tools will keep changing, but understanding what an agent sees, what it can do, and how to verify its work is a skill you can carry with you.