Use AI agents safely for stewardship work

Learning outcome

Scope, test, review, and verify AI-assisted remediation without treating AI output as automatically correct.

Scenario

A tagged quality issue affects enough records that manual remediation would be expensive. The condition is well understood, but you still need controlled review before applying changes broadly.

Read

Exercise

  1. Choose a safe, clearly scoped tagged population.
  2. Create the documented AI-assisted remediation job.
  3. Review the populated data scope and instructions before running anything broadly.
  4. Test the job and inspect the proposed or resulting changes.
  5. Run only when the test evidence is acceptable.
  6. Return to Tag Monitoring or the original population and verify the outcome.
  7. Record one case where AI is appropriate and one where manual, source, or architect-led remediation is safer.

Deliverable

An AI remediation record containing scope, instructions, test evidence, review decision, result, and verification.

Complete when

  • You can explain what the AI job is allowed to change.
  • You test and review before trusting a broader run.
  • You verify the effect on the original quality population.

Next

Continue to Understand streams, downstream impact, and architect handoffs.