An AI Agent without a well-defined system message answers however it likes — formal one time, casual the next, sometimes off-topic. In this episode I show how to give the agent a fixed role and rules.
What you will learn
- How a system message differs from the user prompt and why the split matters.
- How to define the agent's role, tone and constraints in the AI Agent node.
- How to make every query follow the same rules.
- Common mistakes when writing system messages and how to avoid them.
Good practices from the episode
In short: role, scope, response format and prohibitions — those four elements in a system message solve most problems with unpredictable agent replies. The rest you refine iteratively based on tests.
Watch the episode
The video is available on YouTube: youtube.com/watch?v=ycHHYCyZsrk.
All episodes are collected on the Automation Academy page (in Polish). Subscribe to my YouTube channel so you do not miss new material.
The video is in Polish; this post summarizes what it covers.












Comments