Automate a knowledge-base update proposal with review

An automation may identify stale guidance and propose a patch, but the knowledge owner should see the source, change and affected version before publication.

Keep the proposal unpublished

The automation lab creates a synthetic patch against KB-7 version 3.

from automation_cases import kb_proposal_case

result = kb_proposal_case()
assert result["base_version_matches"] is True
assert result["source_present"] is True
assert result["proposal"]["status"] == "pending_review"
assert result["published"] is False

No knowledge-base system or model ran.

Produce a reviewable diff

Trigger from an approved policy source or owner request. Retrieve the current article and version, identify affected claims, and create a line-level patch with source ID, effective date and uncertainty. Validate links, terminology, required fields and cross-article dependencies.

Route to the subject owner with current text, proposed text and authoritative source. Bind approval to the base version and patch digest. If the article changes while pending, reject the stale proposal and regenerate. Publication uses a narrow credential and preserves revision history.

Test conflicting sources, future-effective policy, deleted articles, duplicate proposals and approval timeout. After publication, schedule link and retrieval-index verification. Measure accepted/corrected/rejected proposals rather than equating drafts with value.

The Generative & Agentic AI course connects grounded drafting to versioned approval workflows.

Exercise

Create two local Markdown versions and a source memo. Generate a patch, change the base before approval and prove the publisher blocks the stale digest.

Continue learning

This article is part of the Business automation with AI sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Build a meeting-action extractor with source references.
• Continue with Route low-confidence AI outputs to an exception queue.

Reference: NIST AI RMF Generative AI Profile.