LinkedIn
Facebook
X (Twitter)
WhatsApp
Beschreibung
Agentic AI means a language model that plans its own steps, calls tools and works towards a goal instead of answering a single prompt. The course first shows what happens under the hood, then works at the level where the real decisions are made. Which degree of autonomy fits which problem? What ends up in the context window, and what does it cost? How do you know your system is reliable, and where can external tools, skills and servers attack it? Twelve sessions, hands-on every week, one project
Learning Goals:
- Building an agent in Python without a framework and connecting it to real systems via tools, skills and MCP.
- Assessing the reliability and cost of an agent system using evals and error analysis.
- Determining an agent's trust boundaries and attack surfaces and judging its exposure to prompt injection.
- Designing, building and presenting an agent, and justifying the degree of autonomy it was given.
What you get
By the end of this course you will have built agents that plan their own steps, call tools and work on real tasks. You will also be able to explain why you built them the way you did.
We start with the building blocks. What a tool actually is and how an agent chooses between them. Skills as reusable instructions that only load when they are needed. MCP for connecting an agent to real systems — calendars, databases, internal services — and what that connection does to your context. In the fundamentals you build your first agent loop yourself, in Python and without a framework, so that you understand what a harness takes off your hands later on.
From there it gets more interesting. Delegation to subagents, and what it really buys you once you count the cost. How to judge whether your system works at all, and how to notice when a model update quietly breaks something that used to run. Where prompt injection comes from and why it cannot simply be patched away.
Towards the end we look at long-lived agents such as Hermes and OpenClaw: systems that run continuously, build up memory over time and write their own skills. The point is to recognise their parts, and to ask what happens to a system that keeps correcting itself while nobody is watching.
One question runs through all twelve weeks: which degree of autonomy is appropriate here, and what do you give up when you turn it up?
What you should bring
You should have worked with an LLM assistant such as Claude or ChatGPT before — enough to know that the first answer is rarely the final one. Everything else builds on that.
Programming experience helps but is not a requirement. A good part of the course can be done entirely with agent tooling, where you describe what you want instead of implementing it yourself. If you do write code, Python will feel most familiar, since the reference material is built on it. The concepts carry over to other languages.
No machine learning background is needed. We do not cover model architectures, training, or the mathematics underneath. This course is about building and running systems, not about how the models themselves work.
What helps more than any of the above: a process of your own that annoys you. Something recurring from your work or your studies that you would rather hand over. That becomes your project from week 3 onwards, and the course is considerably more rewarding if you arrive with an idea in mind.
Technically you need a computer you are allowed to install software on, plus access to a model provider. We recommend an account with Anthropic or OpenAI and expect an additional €10–30 for API tokens (more, depending on your own use case). We will help with the setup in the first session; no prior knowledge is required.
Termine
Donnerstag, 29.10.2026
15:00 - 16:45
Donnerstag, 05.11.2026
15:00 - 16:45
Donnerstag, 12.11.2026
15:00 - 16:45
Donnerstag, 19.11.2026
15:00 - 16:45
Donnerstag, 26.11.2026
15:00 - 16:45
Donnerstag, 03.12.2026
15:00 - 16:45
Donnerstag, 10.12.2026
15:00 - 16:45
Donnerstag, 17.12.2026
15:00 - 16:45
Donnerstag, 07.01.2027
15:00 - 16:45
Donnerstag, 14.01.2027
15:00 - 16:45
Donnerstag, 21.01.2027
15:00 - 16:45
Donnerstag, 28.01.2027
15:00 - 16:45
Online
Buchungszeitraum:
01.09.2026
-
23.10.2026
Donnerstag, 29.10.2026
15:00 - 16:45
Donnerstag, 05.11.2026
15:00 - 16:45
Donnerstag, 12.11.2026
15:00 - 16:45
Donnerstag, 19.11.2026
15:00 - 16:45
Donnerstag, 26.11.2026
15:00 - 16:45
Donnerstag, 03.12.2026
15:00 - 16:45
Donnerstag, 10.12.2026
15:00 - 16:45
Donnerstag, 17.12.2026
15:00 - 16:45
Donnerstag, 07.01.2027
15:00 - 16:45
Donnerstag, 14.01.2027
15:00 - 16:45
Donnerstag, 21.01.2027
15:00 - 16:45
Donnerstag, 28.01.2027
15:00 - 16:45
Buchungszeitraum:
01.09.2026
-
23.10.2026