AI in the work of developers and engineers

Key information

Language of the training: English

The training shows how to use AI in practice in everyday engineering work. Participants learn prompt engineering techniques, learn to use GitHub Copilot and Microsoft Copilot in practice (coding, code review, testing, documentation) and see real AI deployments at technology companies. A workshop format with hands-on exercises.

During the training, participants will find out, among other things:

• How to build effective prompts and avoid typical mistakes when communicating with AI
• How to use GitHub Copilot to generate code, tests and documentation in the editor
• How to use Microsoft Copilot to support productivity and work with documentation
• How to conduct AI-assisted code review with a critical evaluation of the results
• What real AI deployments at technology companies look like and what lessons can be drawn from them

The live demo is run in a TypeScript/Node.js stack. The techniques themselves (prompt engineering, working with Copilot, editor integration) are universal: participants practise them in parallel on their own repositories in their team’s stack.

Objective

The objective of the training is to introduce participants to the practical use of AI in everyday engineering work. The training covers prompt engineering techniques, hands-on work with GitHub Copilot and Microsoft Copilot (code generation, tests, code review, documentation) and real examples of AI deployments at technology companies. The emphasis is on the workshop format: every module contains hands-on exercises that participants complete independently.

Programme

Module 1: Prompt engineering for engineers

• How large language models work and why prompt quality determines the quality of the result
• Practical prompting patterns: zero-shot, few-shot, chain-of-thought, role prompting
• Good and bad prompting practices with real engineering examples
• Building and correcting prompts iteratively
• Workshop: participants build prompts for tasks from their own area using demo materials

Module 2: GitHub Copilot and Microsoft Copilot in practice

• GitHub Copilot in the editor (VS Code): tab completion, inline chat, Copilot Chat
• Code generation: from a task description to a working solution
• AI-assisted code review and refactoring suggestions
• Generating and maintaining tests with Copilot
• Technical documentation with AI: README files, code comments, API specifications
• Copilot for CLI and productivity in the terminal
• Microsoft Copilot: productivity support outside the code editor
• Critical evaluation of AI output: what AI does well and what needs a human

Module 3: AI in software testing

• Browser agents in testing: Claude Computer Use, Playwright MCP
• Generating unit and integration tests with GitHub Copilot
• AI-assisted debugging: stack trace analysis, reproducing bugs, fix suggestions
• Automating end-to-end tests with AI agents
• Critical evaluation of test coverage: when AI generates tests that look correct but are in fact worthless

Module 4: Real AI deployments at technology companies

• Case studies: how technology companies deploy AI across the software development lifecycle
• Measuring productivity and ROI of AI tools in engineering teams
• Strategies for adopting AI in a team: top-down vs bottom-up
• AI security and governance in an organisation
• Lessons from failures and deployment anti-patterns
• Q&A session

Learning outcomes

After completing the training, participants:

• Build effective prompts using proven prompt engineering patterns
• Use GitHub Copilot to generate code, tests and documentation in their daily work
• Use Microsoft Copilot to support productivity
• Conduct AI-assisted code review with a critical evaluation of the results
• Use AI agents and Playwright to automate tests
• Recognise the limits of AI tools and know when a human decision is needed
• Know real case studies of AI deployments at technology companies and can plan adoption in their own team

Organisational requirements on the client’s side

• Every participant has an active GitHub Copilot licence (Individual, Business or Enterprise)
• Every participant has their own laptop with VS Code installed and the GitHub Copilot extension
• A GitHub account (personal or company)
• Every participant has an active Claude Pro subscription (Claude Max is recommended for the full capabilities of the AI agent)
• Stable Wi-Fi in the room, sufficient for 20 people working in parallel with cloud-based AI tools
• A projector or large screen for the trainer’s presentation
• A room suitable for workshop-style work (tables, space for notes)

Key information

The training shows how AI agents such as Claude Code and Cline can genuinely speed up the work of a development team. We focus on the tasks developers dislike (task descriptions, tests, logs, documentation) and on embedding the agent in the team’s existing workflows (version control, code review, integrations with everyday tools). Participants learn advanced agentic techniques (subagents, MCP, parallel processing) and leave with a plan for introducing an agentic workflow in their team.

During the training, participants will find out, among other things:

• How to set up Claude Code with a Claude Max or Pro subscription and when to consider alternatives in the client’s cloud architecture
• How to manage the agent’s context effectively when working with an existing codebase
• How to use the agent for task descriptions, tests, logs and documentation
• How to bring Claude Code into the Git and GitHub workflow (pull requests, code review, GitHub Actions)
• How to extend the agent with your own tools and team knowledge bases through MCP
• How to delegate tasks to subagents and use parallel processing

The live demo is run in a TypeScript/Node.js stack (or another agreed with the client before the training, if the team works exclusively in a different ecosystem). The agentic techniques themselves (context management, subagents, MCP, Git and editor integration) are universal: participants practise them in parallel on their own repositories in their team’s stack.

Objective

The objective of the training is to introduce participants to the practice of working with AI agents in a development team’s daily workflow. The training covers using Claude Code for tasks such as task descriptions, generating and maintaining tests, log analysis, creating documentation and AI-supported work with version control. The emphasis is on real team workflows and on integrating the agent with everyday tools (Git, GitHub, code editor, team documentation), as well as on advanced agentic techniques (subagents, Model Context Protocol, parallel processing). All techniques are demonstrated in a way that is independent of any specific technology stack.

Programme

Day 1: Fundamentals of working with AI agents in a team

Module 1: Introduction to agentic programming with Claude Code

• How Claude Code and Cline differ from assistants such as Copilot and Cursor
• Setting up Claude Code locally with a Claude Max or Pro subscription, and when to use AWS Bedrock or GCP Vertex AI instead
• Models and costs: Haiku vs Sonnet vs Opus, when to use which, and how not to burn through your budget
• The agentic programming mindset (delegation vs supervision, the trust boundary)
• Non-determinism of results and how to deal with it in a team
• The most common pitfalls of the first days with an AI agent

Module 2: Managing the agent’s context and memory

• How an AI agent understands the structure of a project
• The Claude memory file (CLAUDE.md) as a contract with the agent and with the team
• Style guides, decision records and team conventions as drivers of AI output
• Integrating Claude Code with the code editor
• Prompt patterns for working with an unfamiliar repository

Module 3: AI for the tasks developers dislike

• Task descriptions, tickets and acceptance criteria: the agent as an aid in refining requirements
• Technical documentation: README files, ADRs (Architecture Decision Records), code comments, API specifications
• Working with logs: analysis, summarisation, reproducing bugs from a stack trace
• Generating and maintaining tests: the test level (unit, integration, end-to-end) and approach chosen to suit the participant’s team’s stack
• Critical evaluation of AI output: what AI does well and what categorically requires a human

Day 2: Team workflow and advanced techniques

Module 4: Version control with an AI agent

• Integrating Claude Code with Git and GitHub
• The agent in the pull request process: auto-review, suggested changes, commit descriptions
• Configuring Claude Code with GitHub Actions
• Automatic replies and initial analysis of reported issues
• Parallel work on branches and how the agent can support it

Module 5: Integrating AI with the team’s everyday tools (MCP)

• Model Context Protocol (MCP): the concept and its uses
• Connecting the team’s own tools and knowledge bases to the agent
• Working with internal documentation, the backlog and the issue database through MCP
• Real workflows: backlog → ticket → code → review → merge with the agent’s involvement
• Security hygiene when connecting an agent to company tools

Module 6: Advanced agentic techniques and team adoption

• Subagents: delegating tasks and specialisation
• Parallel processing and orchestrating multiple agents
• Custom slash commands and automating repetitive tasks
• Strategies for adopting AI in a development team: top-down vs bottom-up
• Industry insights: how large engineering teams scale their work with AI agents

Learning outcomes

After completing the training, participants:

• Set up Claude Code with a Claude Max or Pro subscription and manage the costs of working with an AI agent consciously
• Use an AI agent for task descriptions, tests, logs and documentation in their daily work
• Manage the agent’s context through a memory file, style guides and team conventions
• Integrate Claude Code with Git, GitHub Actions and the team’s pull request process
• Connect their own tools and knowledge bases to the agent through MCP
• Use subagents and parallel processing for real tasks in their own repository
• Recognise the limits of an AI agent and know at which point in the workflow a human decision is needed

Organisational requirements on the client’s side

• Every participant has an active Claude Max or Pro subscription (a fixed monthly fee, with no per-token cost risk). For teams whose company security policy prohibits the use of public AI services, the alternative is Claude through AWS Bedrock or GCP Vertex AI in the client’s own cloud infrastructure (billed per token, requires a budget on the client’s side)
• Every participant has their own laptop with Git and a code editor installed, and administrator rights that allow them to install Claude Code
• A GitHub account (personal or company) with the ability to create repositories and Actions
• Stable Wi-Fi in the room, sufficient for 12 people working in parallel with an AI agent in the cloud
• A projector or large screen for the trainer’s presentation
• A room that allows workshop-style work in pairs and small groups (tables, space for notes)

The live demo is run in a TypeScript/Node.js stack. If the client’s team works exclusively in a different ecosystem (Python, Java, .NET etc.) and wants the demo in their stack, we agree this with the client before the training. Regardless of the demo stack, every participant practises the techniques on their own repository in their team’s language.

Key information

Day 1: AI and agent basics

Objective: to gain the knowledge base needed to create agents. To understand the basics of how AI works, learn to write high-quality prompts and use Microsoft Copilot safely in everyday work, including in Microsoft 365 tools (according to participants’ needs: Word, Excel, PowerPoint, Teams, Outlook).

Section 1: Microsoft Copilot and prompt engineering

Key concepts: prompt, context, context window, AI persona

The context window and tokens: strategies for improving the quality of responses

AI hallucinations: how to recognise and verify them

Microsoft Copilot: an overview of the interface and capabilities at work

Anatomy of a good prompt: task, context, examples, format, tone

The most useful prompts

Practical workshop: working on participants’ examples

Discussion

Section 2: Microsoft Copilot and Office tools (a choice of 2-3 tools according to participants’ needs)

Microsoft Word

Microsoft Excel

Microsoft PowerPoint

Microsoft Teams

Microsoft Outlook

Section 3: Introduction to AI agents

The 3 elements of an assistant: knowledge, system prompt, user prompt

Assistant vs agent: differences and uses

Examples of uses in a company: case studies

Practical workshops

Discussion

After this module, participants will be able to:

create high-quality prompts

apply strategies for avoiding AI hallucinations

use Microsoft Copilot when working with Office tools, including:

edit and summarise text, change the tone of a message and create tables in Microsoft Word

format spreadsheet content, create formulas and charts in Microsoft Excel

create presentations, format images and use style suggestions in Microsoft PowerPoint

check participant availability, search for meetings and summarise meetings in Microsoft Teams

generate email content, search emails and summarise threads in Microsoft Outlook

distinguish an assistant from an AI agent

create an AI assistant

Day 2: Building agents in Microsoft Copilot

Block A: A RAG agent in Copilot Studio

Objective: to build a knowledge-based (RAG) agent that answers on the basis of documents and links and can hold a conversation thanks to topics and triggers.

Section 1: Getting started and configuring the agent

Introduction to the module: building a RAG agent in Copilot Studio (Knowledge Only)

Building, describing and configuring a RAG agent in Copilot Studio

Launching the RAG agent

Practical workshops

Section 2: Feeding in knowledge

Adding knowledge to the agent (1): Microsoft Word documents

Adding knowledge to the agent (2): external web links

Practical workshops

Section 3: Testing and evaluation

Testing and evaluating a RAG-based agent

Practical workshops

Section 4: Conversation and orchestration

Topics, triggers and conversation orchestration

Customising the conversation’s starting topic

Creating and configuring a new topic

Practical workshops

After this module, participants will be able to:

create a working RAG agent

add knowledge from external and internal sources

design a conversation using topics and triggers

test the quality of the agent

Block B: Autonomous AI agents with tools

Objective: to build an autonomous agent that performs a multi-step task (form → Excel → Outlook → email), with a trigger and automation.

Section 1: Introduction

Introduction to the module: building your first autonomous AI agent in Copilot Studio

Launching the autonomous agent: live demo + tests

Building, describing and configuring the AI agent’s instructions

Section 2: The form and starting the process

Creating a Microsoft Forms form using Microsoft Copilot

Building a tool (1): Microsoft Forms and trigger configuration

Section 3: Tools: Excel and Outlook

Building a tool (2): retrieving rows in Microsoft Excel

Building a tool (3): an email draft in Microsoft 365 Outlook

Building a tool (4): sending an email in Microsoft 365 Outlook

Building a tool (5): adding rows in Microsoft Excel

Section 4: Tests and a Teams exercise

Testing and evaluating an agentic automation workflow

Practical exercise: an agent tool for publishing messages in Teams

After this module, participants will be able to:

create a Microsoft Forms form using Microsoft Copilot

build and configure the instructions of an autonomous AI agent in Copilot Studio

build a tool with a trigger based on Microsoft Forms

retrieve rows from Microsoft Excel as part of the agent’s operation

create a draft email in Microsoft 365 Outlook as part of the agent’s operation

send emails in Microsoft 365 Outlook as part of the agent’s operation

add rows to Microsoft Excel as part of the agent’s operation

test and evaluate an agentic automation workflow

build an agent tool for publishing messages in Microsoft Teams