Practice · module 10 of 100
Work together as a team
Make AI usage discussable, repeatable, and accountable.
After this lesson
You create a small team plan with an owner, a trial, and agreements on quality.
In view · Working together towards one result
A team with shared agreements
Determine which tasks your team wants to support with AI.
Share usable prompts with purpose, fillable fields and an owner.
Record who checks, reports errors and updates the approach.
Part 1 of 3 · Choosing together
Your preference for less movement is on. Choose each part yourself.
Choose one shared problem
Start with a task that colleagues often lose time on and whose outcome is easy to assess. Ask the people performing the work what goes wrong and what data is needed. A useful trial solves a concrete problem.
Share methods, not just prompts
When saving a prompt, also save the goal, an example, boundaries, and control points. Agree on who updates the method when a tool or process changes. Discuss errors without assigning blame, so the team can learn from them.
Record when you stop
Determine in advance what success is and which errors are unacceptable. Give an employee responsibility for the assessment. If checking takes more time than the method saves, adjust the trial or stop it.
New practice example
From explanation to a usable result.
A team can only improve together if the method, owner, and control criteria are visible.
Fictional practice situationA fictitious team uses three different prompts for the same newsletter. No one knows which version has been approved.
View a possible solution
This is how it can be done
Maintain one shared base prompt with owner, version date, allowed source, and checklist. Test a change on a fixed sample text before replacing the base.
Why this works
This allows you to link a result to a version and rerun an improvement. Sharing without an owner is not yet management.
Make it slightly more difficult
Describe how a colleague reports an error without copying the full confidential input into the report.
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
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
This allows you to link a result to a version and rerun an improvement. Sharing without an owner is not yet management.
A step further
Describe how a colleague reports an error without copying the full confidential input into the report.
A prompt to get you started
Create a one-page trial plan for [fictional team] that wants to use AI for draft emails. Include: goal, permitted data, owner, five trial cases, quality criteria, control moment, and stop criteria. Ask open questions where a policy choice is required.
Apply it
A team wants to use AI for draft answers. There is no owner yet for the source FAQ and no one knows where to report errors.
Which agreement makes a small trial verifiable?
- Assign an FAQ owner and controller, establish the reporting point, and first test with fictitious questions.
- Everyone chooses their own tool and adjusts the FAQ according to their own judgment.
- Start broad; owners and error reports can be handled later.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
What belongs in a shared prompt library?
- Only a long list of prompts without explanation.
- Goal, example, boundaries, control points, and a responsible person.
- Copies of real customer files for everyone.
Extra practice
Note two applications for your team. Choose one and name the owner, a success criterion, and a reason to stop the trial.
Options: draft emails and preparing calendars. We choose draft emails. The team coach assesses five trial cases. Success: less total time with the same factual accuracy. Stop: the tool repeatedly adds unauthorised commitments or processes non-permitted data.
