Course Outline
Training program
- Fundamentals of LLM and agent operations in developer work
What is an LLM from a programmer's perspective
Differences between a language model, code assistant, and coding agent
How the model analyzes context, code, instructions, and interaction history
Limitations of LLMs
The role of the developer as the person responsible for technical decisions
Typical applications of LLMs in development - Setting up Claude Code, Codex, and Cursor for daily AI work
Overview of tools used in agent-based code work
Claude Code, Codex, and Cursor – differences in workflow and typical use cases
Installation, configuration, and preparation of the development environment
Proper repository setup for AI-assisted work
Principles for preparing project structure, documentation, and instructions for agents
Working with the terminal, IDE, and repository in AI-supported mode
Controlling the scope of changes and minimizing the risk of unwanted modifications - Agents.md, CLAUDE.md, and skills.md as modules organizing agent work
What these are: the main agent markdown files and their role in LLM work
The difference between a one-time prompt and a persistent project instruction
How to create skills describing coding, testing, and documentation standards
Organizing skills for different types of tasks
Examples of good and bad instructions for agents - Plan Mode in practice
What is Plan Mode and when to use it
Planning before executing code changes
Analyzing risks, dependencies, and potential side effects
Translating the plan into specific actions within the repository
Iteratively guiding the agent
Working with larger changes
Evaluating the quality of plans generated by AI - Practical exercises on open repositories
Onboarding to an unknown codebase using LLMs
Identifying entry points, dependencies, and logic flow
Executing realistic team tasks utilizing agents
Refactoring a code segment for readability and maintainability
Generating or supplementing tests
Updating technical documentation and README - Sub-agents in practice
What are sub-agents and when to delegate tasks
Workload distribution among agents:
Designing the scope of responsibilities for sub-agents
Parallel work and control over result consistency
Applying sub-agents in larger refactoring tasks - MCP as a way to extend agent capabilities
What is MCP and its role in working with AI tools
MCP clients, servers, and context sources
Connecting agents with additional tools, data, and workflows
Examples of MCP applications in developer work - Safety, quality, and accountability in AI-assisted work
Risks of code, data, business logic, and architecture information leaks
Principles for working with production code and critical security fragments
Verifying changes generated by AI
AI in the code review process
Best practices for implementing agent-based code work within an organization - Summary and practical implementation
Key principles of effective developer-LLM collaboration
How to transfer training knowledge into daily workflows
The minimum set of practices to implement post-training
Requirements
Experience in software development is recommended, along with a knowledge of Git basics and the ability to navigate a code repository. Participants should know at least one programming language at a level that allows reading and modifying code; examples will be demonstrated using Python. The training does not require prior experience with Claude Code, Codex, Cursor, or MCP, although basic familiarity with working in a terminal will be helpful.
Target audience
· Developers at an intermediate to advanced level;
· Developers working with existing, complex, or poorly documented repositories;
· Software development team members who wish to streamline their AI usage practices in daily work;
· Tech leads and senior developers responsible for code quality, code review, and selecting development tools;
· Individuals seeking to consciously utilize LLMs for code analysis, refactoring, documentation, testing, and automation of technical tasks.
