Agentic Engineering Lab is a hands-on course for developers who already write code with an AI agent and want it to produce code that passes review. You’ll set up the agent to check its own work: it agrees on a specification with you, writes the code, and runs it through linters, static analysis and tests.
At the end of the course, you present a final project built with this process.
We work in a shared repository: you repeat every step after the trainer and take the finished setup into your own project. The trainer works in Claude Code, and every approach applies to any coding agent: Codex, Cursor, GitHub Copilot.
Who this Agentic Engineering Lab course is for:
- Developers who already use an AI agent but spend too much time fixing its code
- Those who have tried vibe coding and see that it doesn’t scale on a real project
- Tech Leads and Team Leads who want to make an AI workflow a shared team standard
- QA engineers with automation experience who want to use an agent for testing and finding defects
You will learn to:
- Understand how a coding agent works and manage its context
- Prepare a repository for working with an AI agent: AGENTS.md / CLAUDE.md, permissions, modes
- Build a workflow with specifications (spec-driven development) and automated checks
- Review AI-generated code and catch the agent’s typical errors and hallucinations
- Use the agent for testing, debugging and incident analysis
- Work with agents safely and keep them under control in a team
- Build your own project with an agent, from idea to demo
Course format:
- 8 online sessions of 1.5–2 hours: 7 hands-on sessions and a final project demo. Tuesdays and Thursdays at 18:00 (Kyiv time), November 24 – December 17.
- Every session is hands-on: you clone the same repository as the trainer, run the agent and repeat every step.
- Lab format: we experiment, compare agents’ results and discuss what worked
- Recordings of all sessions will be available to participants.
- Homework assignments are steps of your final project. The core part of each step takes 30–60 minutes, with extra tasks if you want to go deeper. You work at your own pace, and the trainer answers questions in the group chat.
- Final project: a small web app you build with the agent during the course, based on your own idea or one from the trainer’s list. Before the demo, you can book a 15-minute session with the trainer to discuss the project and your presentation.
Course Program
Session 1. How an AI coding agent works: installation and first run
- From LLM chat to agent: model, tools, agent loop
- Installing and configuring the agent (together, during the session)
- Context window: what the agent sees and why it “forgets”
- Practice: the agent explores an unfamiliar repository
Session 2. Project context and your first feature
- AGENTS.md / CLAUDE.md: giving the agent knowledge about the project
- Plan mode, operating modes, permissions
- Practice: your first feature with the agent, from plan to code
Session 3. The agent checks its own work
- Linters, typing and static analyzers as feedback for the agent
- Tests and hooks as mandatory checks
- Practice: the agent finds and fixes its own mistakes
Session 4. Spec-driven development: from specification to code
- Why vibe coding doesn’t scale
- OpenSpec: proposal → design → tasks → implementation
- Reviewing AI code: what to check, typical hallucinations
- Practice: a feature from specification to code
Session 5. QA and testing with an AI agent
- Skills: a QA methodology the team can reuse
- Unit, API and E2E tests, Playwright
- Practice: finding defects in a web application
Session 6. Debugging and incident analysis with an agent
- Code shows what should happen. Logs show what did happen
- Logs, the database and the browser as sources of evidence for the agent
- Practice: investigating incidents in a running application
Session 7. AI code security and agent control
- Vulnerabilities in AI code, secrets, and what’s safe to share with an agent
- Prompt injection, risks of MCP servers and third-party skills
- Monitoring agents instead of banning them: permissions, hooks, logs
- Team rules for the agent: shared settings, AGENTS.md, skills
- Practice: security analysis of AI-generated code
Session 8. Final project demos
- Participants present their projects: what the agent did, which checks caught errors, what they take back to their team
- Feedback from the trainer and the group
- Wrap-up: how to honestly assess the impact of AI on your work
After the course:
You’ll have a working setup you can move straight into your project:
- an AI agent configured with your project context (AGENTS.md / CLAUDE.md)
- automated checks: linters, static analysis, tests, hooks
- a development process from specification to code
- team rules for the agent: shared settings, AGENTS.md, skills
- plus your own app, built with an agent from idea to demo, with feedback from the trainer and the group
What background do you need?
- Commercial development experience, Git, basic terminal skills
- A laptop on which you can install software
- A coding agent. Participants pay for their own subscription; we install it together in the first session:
- recommended: Claude Code (Claude Pro subscription or higher) or Codex (ChatGPT Plus or higher), about $20 per month
- free option: Codex on the free ChatGPT plan. Its limits are low, so you’ll likely watch part of the practice rather than repeat it
FAQ
Will the course work for me if I use Cursor or GitHub Copilot?
Yes. The trainer demonstrates everything in Claude Code, but the approaches (context, specifications, checks, hooks) carry over to any coding agent.
Will there be homework?
Yes. Homework assignments are steps of your final project: the core part of each takes 30–60 minutes. You do them at your own pace, the trainer answers questions in the group chat, and before the demo you can book a 15-minute session with the trainer.
What if it's not a good fit?
100% refund before the start, 70% after the first class.
What if I miss a session?
You can watch the recording. Each session is self-contained, so missing one doesn’t stop you from working in the next ones.
How much theory is there?
Exactly as much as you need to understand the practice. Every session is built around working with the repository.
How is this different from vibe coding?
Vibe coding works for prototypes. In this course you build a process where the agent follows a specification and checks its own code with linters and tests, so the result is ready for review.
Can a team take the course together?
Yes, the program can be adapted for corporate training.
What language is the course taught in?
Ukrainian, with technical terms in English.
