Workshops: Prompt Engineering

Context Engineering: Prompting for AI Agents Workshop

This workshop shows why agents succeed or fail less on single prompts than on the right context. You will learn to design instructions, tools, knowledge and memory so AI agents complete multi-step tasks reliably, from Copilot agents to agents you build yourself.

Who is this workshop for?

  • Developers building agents
  • AI solutions architects and technical product owners
  • Automation teams using n8n, Make or Power Automate
  • Power users configuring agents in ChatGPT, Claude or Copilot
  • Teams using coding agents such as Claude Code or Codex

What you will take away

  • You understand the difference between prompt engineering and context engineering.
  • You write agent instructions with goals, boundaries, stop criteria and escalation rules.
  • You design tool descriptions that agents use correctly and sparingly.
  • You manage context windows deliberately: what goes in, what gets summarised, what is offloaded.
  • You use project files such as CLAUDE.md or AGENTS.md and protocols such as MCP purposefully.
  • You assess the risks of autonomous agents and apply approvals and human oversight.

Agenda

  1. Module 1

    From prompt to context

    • What makes an agent: goal, tools, loop, memory
    • Why agents fail: context loss, overload, wrong tool choice
  2. Module 2

    Instructions for agents

    • System instructions, roles and success criteria
    • Project memory: CLAUDE.md, AGENTS.md, custom instructions
    • Exercise: specify a research agent
  3. Module 3

    Tools and knowledge

    • Writing clear tool descriptions and parameters
    • Model Context Protocol (MCP) and connectors
    • Retrieval: relevant documents instead of whole knowledge bases
  4. Module 4

    Managing the context window

    • Summaries, notes, sub-agents and hand-offs
    • Structuring long tasks in stages
  5. Module 5

    Control and security

    • Permissions, approvals, human oversight
    • Prompt injection via web pages, emails and files
    • Logging and traceability

Why context matters more than the single prompt

A chat prompt steers one answer. An agent works across many steps, calls tools, reads documents and makes intermediate decisions. What it knows at each moment, and what it does not, determines the quality of the result. Designing that entire information environment is called context engineering.

Our AI Code Engineers and AI Solutions Architects build agents and agentic workflows every day. In the workshop they pass on that practice as concrete patterns.

Agent platforms at a glance

We work across platforms and choose the practice environment to fit your participants.

  • Agent features in ChatGPT and Claude
  • Microsoft 365 Copilot agents and Copilot Studio
  • Google Gemini and Mistral Le Chat agents
  • Coding agents: Claude Code, OpenAI Codex, Cursor, GitHub Copilot
  • Workflow tools: n8n, Make and Zapier with AI steps

Securing autonomous systems

The more an agent may do on its own, the more boundaries matter. Prompt injection through external content, overly broad permissions and missing logs are real risks. We show safeguards and how to build in meaningful human oversight, also with a view to EU AI Act requirements.

Frequently asked questions

Do participants need to code?

Not for the concepts. For exercises with your own agents via APIs or MCP, basic coding skills help. With mixed groups we run separate exercise tracks.

How does this differ from your automation workshops?

Automation workshops focus on workflows in tools such as n8n or Make. This one is about how agents receive information and instructions, regardless of the tool.

Do you cover multi-agent systems?

Yes, in the form of sub-agents and hand-offs. For complex architectures we recommend accompanying consulting by our AI Solutions Architects.

Workshop

Get in touch

Shall we work through context engineering on your actual agent project?

Tell us a little about your team – we will get back to you within 24 hours with a proposal for content, format and dates.

Prefer to talk?

We only use your details to handle this request. See our privacy policy.