AI-Assisted Software Development
Overview
This four-day (2 + 2) training covers the transition from traditional software development to an AI-native development model.
The training is divided into two modules:
- Module 1: AI-Assisted Software Development in Practice (2 days): using AI as part of daily development work.
- Module 2: AI in Applications, as Agents, and in Production (2 days): integrating AI into applications, building agents, and taking applications to production.
The training is hands-on. The agenda is preliminary and may change slightly depending on the needs of the participants.
Target Audience
The training is aimed at software developers who want to integrate AI into their daily development work.
Prerequisites
- Proficiency in at least one programming language, e.g. Python, Java, JavaScript, etc.
- Familiarity with version control and command-line environments
No prior experience with AI tools is required.
Learning Objectives
By the end of the training, participants will be able to:
- Use AI as a practical daily tool in software development
- Understand the strengths of different AI coding tools and choose the right one for the situation
- Generate, test, and review code with the help of AI
- Understand how AI is integrated into applications using different integration strategies
- Know the principles of agent and MCP architecture and be able to apply them
- Take an AI-assisted prototype toward production in a controlled manner
The training is delivered remotely. Participants must have access to the following tools before the training begins.
Claude
Although the training covers several AI tools (Codex, Claude, Gemini, etc.), the exercises are built around Claude. Participants need access to Claude during the training.
Option A: Company API Keys (Recommended)
The organizing company creates an account at console.anthropic.com, loads it with credits (pay-as-you-go), and creates API keys for participants. Participants configure Claude Code to use the provided API key instead of their own login. Usage is billed token-based from the company’s credit balance, and there are no per-user hourly limits, so work continues uninterrupted as long as credits are available.
Option B: Personal Subscription
Each participant uses their own Claude Pro or Team subscription (Team is recommended due to higher usage limits). Note that Claude Pro has usage limits that reset every few hours, and intensive hands-on work may exhaust the limit mid-session.
Company API keys ensure that exercises proceed without interruption throughout the training days, without the risk of quota-related interruptions.
GitHub
Participants must have a GitHub account, either personal or provided by their organization. The account is used for version control in exercises and CI/CD exercises.
Slack
Some communication during the training takes place via Slack. Participants must have access to the training Slack workspace. Login instructions will be provided before the training begins.
Module 1: AI-Assisted Software Development in Practice - 2 days
- What generative AI and LLMs mean in practical software development
- How models produce responses and why they make mistakes
- The changing role of the developer: from implementer to guide, evaluator, and quality assurer
- The AI-native development model
- Principles of prompting and the structure of a good prompt
- Making project context understandable to AI
- Context management in long tasks
- Main categories of AI tools: chat-based, IDE assistants, agentic development environments
- Example tools: Claude (Code), GitHub Copilot, Cursor, Gemini, ChatGPT/Codex-style use, IntelliJ IDEA AI
- Differences between models
- Claude Code CLI basics
Generating Code with AI
- Differences between building new systems and AI-assisted modernization of existing code
- Greenfield: AI in design, initial implementation, interpreting specifications, and layering tests
- Brownfield: AI in understanding legacy code, refactoring, documentation, and identifying change risks
- Claude Code: plan mode, change management, debugging, feature development
- AI-assisted code review
Specification- and Test-Driven Development with AI
- Generating tests from specifications with AI
- Tests as a driver in AI-assisted implementation
- Validating AI-generated code against acceptance criteria
- Claude: Test and Evaluate, Develop Tests, Eval Tool
Module 2: AI in Applications, as Agents, and in Production - 2 days
AI in Applications and Integration Strategies
- Comparing and selecting integration strategies: direct API calls, SDK-based integration, MCP-based tool connections
- Cloud vs. local models and hybrid strategies
- Installing and using a local cloud service
Retrieval-Augmented Generation (RAG)
- RAG thinking in practice
- Using your own documentation and databases in RAG
- When to use RAG vs. direct context
- Embeddings and Files
- Agent thinking in development
- Multi-agent model
- LLM Tool Calling
- Claude Tools
- OpenAI Function Calling
- MCP (Model Context Protocol)
CI/CD and AI in the Development Pipeline
- Traditional vs. AI-assisted pipeline
- From development to production: error handling, edge cases, performance
- AI’s role at every stage
- Code generation and review
- Automatic test writing
- Commit messages and PR descriptions
- Pipeline configuration maintenance
Workshop
- Developing a participant’s own idea into a prototype using AI
- Taking the prototype toward production in a controlled manner
- Testing, documentation, architecture
- Presenting your own prototype