Agentic Coding Course

Overview

This class is available for teams on request live online (2-4 hours). Please contact us to receive a proposal.

This course is for professionals who understand basic software development and want to use coding agents as engineers rather than as casual chat users. The central claim is that the quality of an agentic workflow depends less on a clever prompt than on the runtime around the model: instruction files, available tools, permission boundaries, feedback loops, and orchestration patterns.

The course uses Claude Code and OpenAI Codex as reference tools, but the patterns apply broadly. A team may use Claude Code for local interactive work, Codex for delegated tasks, MCP servers for shared tools, and a custom harness for production workflows. The course teaches portable design principles rather than one vendor-specific trick.

By the end of the course, trainees should be able to design a small Explorer -> Implementer -> Reviewer pipeline with explicit permissions, file-based handoffs, test verification, and stop rules.

What you will learn

  • How to move from prompts to runtime systems
  • The five-layer architecture of an agentic coding system: instructions, the agent loop, tools, permissions,  extensions
  • The instruction layer: CLAUDE.md and AGENTS.md
  • SPEC.md: the task contract
  • Build simple AI-enabled automations or agents for tasks such as routing, summarising, drafting, and notifications using accessible no-code tools

Who is it for

  • IT officers and developers

Tools used during the class

  • Claude Code
  • OpenAI codex