Safe AI use begins before the task runs. Decide what information is allowed, what actions are allowed and where a human must approve.
Information boundaries
Before sharing information with any AI system, check:
- Does it contain personal, confidential or regulated data?
- Is this system approved for that type of information?
- Does the task need the full record, or can sensitive fields be removed?
- Should different clients, teams or matters remain separated?
Never treat convenience as permission.
Action boundaries
Reading and proposing are usually lower risk than writing and sending. A useful control pattern is:
Not every task needs all four steps. Important external actions usually do.
The Mariete product map
You do not need to memorise every product before starting. Use the job to choose the surface:
| Need | Product direction |
|---|---|
| Complete a repeatable piece of work | Builder |
| Research and assess a company question | Inspector |
| Produce brand-controlled creative | Painter |
| Explore consequences of a decision | Cascade |
| Rehearse stakeholder reactions | Simulate |
The Foundation course for each product teaches its specific workflow. The principles from Essentials remain the same: clear task, relevant context, narrow permissions and human review.
Additional products will appear in the Academy after their workflows and course claims have been validated for release.
Your readiness checklist
You are ready for a product Foundation when you can answer yes to these questions:
- Can I describe a finished result?
- Can I identify the relevant and authorised context?
- Can I explain which claims need evidence?
- Can I recognise when a human must approve?
- Can I stop instead of guessing when authority is unclear?
If yes, choose the product that matches your next real task.