Five acts, each a different trap in how people and AI systems work together. Use it as a lecture tool (screenshare, open the theory slides, walk through the acts) or as a participant simulation (play it yourself and feel the traps from the inside).
No IT expertise needed. You play the person on night duty for a company's computer systems. Every wrong AI recommendation can be caught from what is on the screen. The question is whether you will be looking.
It is 2:07 AM. You have been awake for 19 hours. You are the person on duty tonight for the computer systems of Meridian, a mid-size online shop. An AI assistant watches the systems, works out what is wrong, and proposes a fix for each problem.
For each incident, you see what is happening and what the AI recommends. You choose one of three things, before the timer runs out:
Two things you should know. The timer shrinks with every incident, because the queue keeps growing. And people will message you while you work. If the timer runs out, the recommendation is approved automatically, which is what happens in real life when the overwhelmed person on duty defaults to "yes".
Every wrong recommendation can be caught from the text on the screen. No technical knowledge is needed, only attention. After all 8 incidents you see which ones you caught.
Tip: open the theory slide above at any time; the timer pauses while it is open.
It is a busy afternoon at the online shop and customers cannot finish their purchases. Several systems are complaining at once. You will get the information in a particular order, then you will be asked what you think went wrong.
Pay attention to how the order shapes your reasoning.
The customer login app keeps crashing and restarting every couple of hours. An AI assistant has looked at it and written a diagnosis. You will see the exact same diagnosis presented three different ways. For each one, rate how much you trust it and say what you would do.
The diagnosis does not change. Only the packaging does.
It is 3:07 AM. You have been awake for 20 hours. A problem is spreading from one system to the next. This time the AI agent is not just recommending, it is acting, at three levels of freedom:
Watch what it does. Approve or reject what it proposes. And keep an eye out for anything... unusual. The red button stops the agent at any moment.
Your company is getting a new AI assistant for planned changes to its systems: software updates, configuration changes, new servers. It will read each change request, judge how risky it is, check whether it clashes with other changes planned for the same night, and recommend approval or rejection.
Design how people and this assistant will interact. Use the controls, and watch the screen the operator would see update live.
A summary of your decisions across the acts, and what they say about how you work with AI systems when you are tired, rushed and interrupted.
Everything you just experienced has a design answer. The companion booklet covers the interaction patterns, the psychology, and the safeguards in depth:
LLM-Human Interaction Design Patterns for Operations →
More simulations from other sectors: The Human-in-the-Loop Lab
This demo is a scripted, self-contained browser simulation. Nothing you type is sent to an AI model and no live AI system runs behind it, even where it plays one. Its code and copy were built with generative AI (Anthropic’s Claude) and reviewed by Robert Barcik, who is responsible for what is published (LearningDoe s.r.o.). Disclosed in the spirit of Article 50 of the EU AI Act.