Advanced AI Coding Training – AI agents in a developer’s work
Level
AdvancedDuration
16h / 2 daysDate
Individually arrangedPrice
Individually arrangedAdvanced AI Coding Training – AI agents in a developer’s work
Advanced AI Coding is a practical course in advanced work with coding agents such as Claude Code, Codex, GitHub Copilot and Cursor. Participants work on an existing repository and set up an environment in which AI draws on persistent project instructions, tools, MCP, skills and automatic validation mechanisms. The course covers delegating complex multi-file changes, managing context, working with subagents, controlling autonomy and assessing the quality of results. Particular emphasis is placed on the safety of command execution, minimising permissions, protecting data and secrets, resistance to prompt injection and the mandatory verification of changes through tests and code review.
Who is the Advanced AI Coding training for?
For developers who work comfortably with Git, the terminal, an IDE, automated tests and existing repositories, and who have already used at least one AI assistant or agent.-
For graduates of the „AI Tools Supporting Developers” training or people with equivalent practical experience.
-
For senior developers and tech leads who want to standardise agentic workflows in a repository and widen the range of tasks that can be safely delegated to AI.
-
For development teams that want to move from single prompts and generated code snippets to repeatable, controlled processes for working with AI agents.
What will you learn during the Advanced AI Coding training?
- You will prepare a repository for work with AI agents by creating persistent project instructions, rules for using tools and criteria for validating changes.
- You will configure an advanced workflow in Claude Code, Codex, GitHub Copilot or Cursor using MCP, skills, hooks or the equivalent mechanisms available in a given tool.
- You will carry out an agentic multi-file change from analysis and plan through implementation to tests, code review and assessment of the diff.
- You will split a complex task between a main agent and specialised subagents, and assess the quality of the results before integrating the changes.
- You will define automatic checkpoints covering tests, linting, security analysis or other quality mechanisms required by the project.
- You will assess the risks connected with agent autonomy, terminal access, MCP, data and secrets, and apply the principle of least privilege and control over high-risk actions.
Training program
Day 1 (8h)
Module 1: Context engineering and repository instructions
- Architecture of an agent’s context: code, documentation, change history, instructions and tool output
- Repository instructions: AGENTS.md, CLAUDE.md, Cursor rules and GitHub Copilot instructions
- Context strategies for large repositories: scope, hierarchy, selection and noise reduction
Module 2: Agentic workflows for complex changes
- The agent’s working cycle: analysis, plan, implementation, tests, diagnostics and diff review
- Multi-file changes: dependencies, migrations, refactoring and acceptance criteria
- Controlling autonomy: permissions, checkpoints, Git, environment isolation and approval points
Module 3: Extending agents — MCP, skills and hooks
- Model Context Protocol: tools, data sources, configuration and the scope of permissions
- Agent skills and reusable instructions: specialisation, resources and team standards
- Hooks and automatic checkpoints: tests, linting, formatting, scanning and execution policies
Day 2 (8h)
Module 4: Subagents and orchestrating work
- Agent-subagent architecture: delegation, context isolation and specialisation
- Parallelisation strategies: code analysis, implementation, tests and code review
- Coordinating results: conflicts, integrating changes, context budget and cost of execution
Module 5: Quality and security of agentic programming
- Validating AI results: tests, static code analysis, acceptance criteria and change review
- Threat model for agents: prompt injection, sensitive data, secrets, dependencies and unauthorised operations
- Guardrails for an agent’s work: least privilege, allowlists, sandboxing, action approval and an audit trail
Module 6: Advanced workshop — an agentic change in a repository
- Task specification: requirements, constraints, technical plan and definition of done
- Multi-agent workflow: Claude Code, Codex, GitHub Copilot or Cursor, subagents, MCP and automatic validation
- Final assessment of the solution: diff, tests, security, code quality, cost of the agent’s work and readiness for review
