How do you turn an AI experiment into a reliable way of working? In this two-day course, you will learn to assess language models and AI agents, select suitable tasks and verify results transparently. Examples from software development, engineering and service connect the fundamentals to your everyday work.
The course spans technologies and vendors. We teach approaches that transfer across different AI assistants and coding agents. Exercises use tools approved by your company. Features such as project knowledge, persistent instructions, tool connections and effort levels are covered according to the capabilities of the chosen environment; not every vendor supports them in the same way.
Course objectives
- Distinguish between language models, AI applications and agents, and assess which tasks suit which working environment.
- Distinguish training knowledge, supplied context and tool results, and recognise hallucinations and missing domain knowledge.
- Formulate requests with context, task, supporting material, output format and verification instructions, then improve them through dialogue.
- Verify results using original sources, calculations, tests and expert review.
- Use coding agents in small, controlled steps and trace and reverse changes with Git.
- Deliberately design permissions, data access and approvals, and respect limits involving production and safety-critical systems.
- Establish shared project rules and documentation, and evaluate your own use cases by effort, quality and cost.
Target audience
Day 1 is for staff in software development, engineering and service, as well as technical project leads, AI coordinators and internal champions. Programming skills are not required for this day of fundamentals and applications.
Day 2 is primarily for software developers and technical teams who want to use coding agents with their development environment, version control and toolchain. Technical tasks and examples may also come from automation and service. The participant groups are agreed during preparation.
Day 1 – AI fundamentals and everyday applications
- Understand language models: tokens, text generation, training knowledge and limitations. Distinguish plausible answers from substantiated statements; handle uncertainty, invented sources and calculation errors.
- Prepare context deliberately: supply your own documents, examples and company knowledge; choose suitable text and table formats. Distinguish chat history, context windows, project knowledge and, where available, memory; structure long tasks and save interim progress.
- Write effective instructions: apply the five components of context, task, supporting material, format and verification. Refine results through dialogue, use clarifying questions and check statements against original documents.
- Compare working environments: chat, Office assistants, IDEs, terminals and, where applicable, cloud workspaces. Understand which files and systems are accessible and how generating answers differs from taking tool actions.
- Protect data and access: company rules, approved accounts and tools, confidential information and data minimisation. Recognise prompt injection in third-party files, emails and websites and assess its risk for agent actions.
- Practise with your own cases: summarise technical documents, compare materials, extract information with sources and create structured analyses. Each participant describes two or three suitable use cases with expected benefits and necessary checks.
Practical phase between training days
A practical phase is planned between the two days. Participants try selected tasks with approved data and document their approach, result quality and effort with and without AI. These experiences form the starting point for day 2. Timing and tasks are agreed together.
Day 2 – using coding agents in projects with control
- Prepare the workspace and toolchain: set up an approved coding agent in the project folder and connect it to the terminal, IDE, compiler and tests. Distinguish local and hosted workspaces and their access capabilities.
- Use Git as a safety net: understand working copies, staging, commits, remotes, branches, merges and worktrees. Start from a saved baseline, review diffs and selectively reverse changes; control commits, pushes and approvals deliberately.
- Practise the agent workflow: formulate the task and acceptance criteria, analyse the existing project, review a plan, have small changes implemented and verify results through builds, tests and review. Analysis, documentation and verification provide an entry point before larger code changes.
- Use technical safeguards: assess read and write permissions, command and network access, clarification and approval steps, and available sandbox or container solutions. Exclude confidential files and separate exercise environments from production machines, controllers and data.
- Demonstrate quality: understand unit, module, integration and, where relevant, UI tests; challenge test claims with deliberate faults and independently verify checks. Combine rule-based checks and compilers with expert or additional AI review; responsibility stays with the team.
- Manage context, models and budget: select models suited to the task, use available effort or reasoning settings and monitor context size, tool calls, runtime and costs. Use additional agents only for clearly bounded tasks and consolidate their results.
- Organise project knowledge for the long term: version concise project instructions and coding guidelines, maintain technical and business documentation and decision and development logs. Capture recurring workflows as templates or skills; use integrations, such as MCP, only within approved boundaries.
Practical exercises and examples
- Analyse a technical fault list or manual: compare answers with and without supplied material and trace each statement to a source.
- Compare a component discontinuation notice against a machine or component list and create a verifiable results table.
- Have an existing function or project explained and documented before proposing changes.
- Fix a bug or add a small feature in an isolated demo project: task, Git diff, tests, review and documented completion form a complete workflow.
- Design a reusable project scaffold with instructions, documentation and checks, and evaluate your own use case for further application.
Tasks are adapted to participants and available tools. We work with demo data or explicitly approved documents. No production machines or controllers are connected; safety-critical functions are neither created nor modified in exercises.
Standards and transfer to the team
To conclude, we distinguish organisational rules for people, technically enforced settings, shared agent instructions and project-specific guidance. Owners, storage locations and maintenance are defined for reusable rules. A simple AI application register records purpose, owner, data, status and approval.
The exercises produce an initial action plan for your team: suitable pilot tasks, outstanding prerequisites, required quality checks and criteria for moving from experiments to production. Further support or deeper work on standards and testing can be agreed separately.
Prerequisites and preparation
No prior AI knowledge is required. Day 2 assumes basic familiarity with the relevant development environment and source code; the necessary Git fundamentals are taught in the course. We agree the language and toolchain for practical tasks in advance.
A suitable computer and access to the agreed, company-approved AI tools are required. The technical practice day also needs a development environment, Git and a suitable test or build environment. Installations, licences, access and permitted data are clarified together before the course; no particular AI vendor is required.
Teaching approach and supporting materials
Short explanations, live demonstrations, supervised exercises and joint reviews alternate. Task cards, checklists, a demo project and templates for project instructions and documentation support practical transfer. Vendor-specific commands and settings are explained for the tools actually used.
Course duration
2 days
Two training days with an agreed practical phase in between. Day 1 provides shared fundamentals and applications; day 2 deepens the use of coding agents in technical projects.