What will your team work on?
A proposed sequence, adapted to your tools, experience and available time.
Draw the workflow
List triggers, inputs, decisions and outputs. Separate predictable rules from steps that require interpretation.
Build the smallest useful prototype
Connect a sample input to one model task and a draft output. Use test accounts and limited permissions.
Design human approval
Put approval before external messages, payments or irreversible changes. Show the reviewer the original evidence and intended action.
Test the awkward cases
Try missing fields, duplicate inputs and tool failures. Add a stop condition, an error log and a way to recover without repeating actions.
What does practice look like?
A supervised prototype and a test plan
Classify fictional supplier requests, extract a reference number and draft a response. Route uncertain cases to a reviewer. The prototype must never send the response itself.
A teaching example. It does not describe a completed client engagement.
| Work product | What makes it usable |
|---|---|
| Source material | Approved inputs, with the useful context retained. |
| Reusable instructions | A clear task, constraints and expected output format. |
| Review checklist | The checks a colleague must perform before using the result. |
| Next experiment | An owner, an example and criteria for deciding whether to continue. |
What should you bring?
Bring a concrete question, a representative example and the rules your team already follows. The purpose of preparation is to spend the session practising, with the right accounts, permissions and source material in place.
- A task that someone on the team does regularly.
- An approved AI tool, or a decision about which environment to use.
- A sample that can be shared safely, preferably stripped of sensitive data.
- A reviewer who knows what a good result should look like.
The final proposal sets the duration, group size, trainer, scope and fee. Tool subscriptions and any follow-up are specified separately.
