Executive summary¶
The problem¶
The internal runbook is hand-maintained and drifts out of date. Truth is scattered across Ansible repos, SharePoint, ServiceNow, Power BI and Teams. Senior engineers lose hours re-deriving and searching procedures — and P1s are slowed by stale runbooks across EMEA, Brazil, China and on-prem.
The proposal¶
A self-updating Code wiki: when a source changes, Azure OpenAI drafts the documentation update and opens a pull request; a human owner reviews and merges. A weekly drift report keeps documentation debt visible and shrinking. Phase 2 adds an "ask-the-runbooks" assistant.
-
Pros
- Runbooks stay accurate — changes become reviewed PRs in minutes.
- Faster incident response; on-call finds the right procedure fast.
- Better onboarding; canonical, current map of the estate.
- One search box across wiki + reports + comms (Phase 2).
- Extensible: new source = one adapter.
- Vendor-neutral: pluggable LLM.
-
Cons
- Reviewer load — owners must review auto-draft PRs.
- Prompt/template tuning is a small ongoing effort.
- Needs a named owner per section (some teams resist).
- Content debt becomes visible on day one (feature, looks noisy).
-
Risks (see fixes)
- AI hallucination → PR-gated + cited + confidence-scored.
- Confidential leakage → link-only, never sent to the LLM.
- Runaway cost → quotas + budget alerts + circuit breaker.
Why now, and why this project
It fixes a pain senior+ staff feel weekly, it rides on assets we already own (near-zero cost), and it produces a compounding asset — a clean corpus — that later powers an AI assistant. Ideal first win for the Automation Observatory.