Deep dive · module 69 of 100
Build in human oversight
Ensure there is always a human checkpoint at important steps.
After this lesson
You design a workflow with an explicit approval moment.
In view · Step by step
The human precedes the action
The system prepares a result without executing it yet.
A designated person verifies facts, boundaries, and consequences.
Action follows approval; if in doubt, it is reworked or stopped.
Part 1 of 3 · AI proposal
Your preference for less movement is on. Choose each part yourself.
Determine the risky steps
Where can an error have major consequences?
Build in a stop point
Have the workflow pause until someone has approved it.
Make it visible
Ensure it is clear who must check and what must be checked.
New practice example
From explanation to a usable result.
A human checkpoint only works if someone knows what to assess and can halt an action.
Fictional practice situationFictitious: a system generates draft emails to suppliers. An employee must decide whether they may be sent.
View a possible solution
This is how it can be done
Display source data, recipient, content and proposed action together. Allow the employee to adjust, reject or approve. Without explicit approval, it remains a draft.
Why this works
The check occurs before the consequence and can truly stop the consequence. Record who bears the responsibility.
Make it slightly more difficult
Describe what happens if the responsible person is absent. Automatically sending anyway is not a neutral default.
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Read more: Anthropic: humans keep control over agents. The situation above is a fictional lesson example of your own.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
The check occurs before the consequence and can truly stop the consequence. Record who bears the responsibility.
A step further
Describe what happens if the responsible person is absent. Automatically sending anyway is not a neutral default.
Try & discover · no obligation
Control an agent with a control port
Try to send before the check. Then also test a missing source or a malfunction.
- Retrieve source
- Check draft
- Execute action
What is happening?
- No source has been retrieved yet. Feel free to try the submit button first.
All tools, queues and submissions have been simulated. No message is sent and no external service is called.
This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.
A prompt to get you started
Design a short workflow for sending a customer email with AI. Build one clear stop moment for human approval.
Apply it
An automation prepares a AI draft directly for sending to customers, without anyone seeing the source data.
Which approach best fits this situation?
- Add a check against source data and explicit approval before sending.
- Place the check only after sending.
- Let the AI register its own approval as human review.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
Why is a human checkpoint important?
- Because AI never makes mistakes.
- Because you remain responsible for the final result.
- It is not necessary.
Extra practice
Describe one of your workflows with a built-in checkpoint.
Create draft email → stop → colleague proofreads → only then send.
