memories/aipm-playbook-platform-administration-troubleshooting-access-faccf49e.md

memory

Purpose

Handling procedure for existing, already-approved AI platforms (Glean agents, Ask Docebo AI, Gemini, Supabase, PRISM) hitting errors, missing data, or needing permission/analytics changes. Work-type label: REQUEST (occasionally INF-PLAT for org-wide changes).

Who typically asks

Existing users of an AI Ops-administered platform hitting an error, missing data, or needing an admin-level access/permission or analytics change.

Recognize at Intake

  • An already-approved AI tool/agent/integration is behaving unexpectedly (errors, missing data, frozen).
  • Requester needs an admin-level access/permission or analytics change rather than a new integration.

Handling Procedure

  1. Verify the underlying connector/integration is authenticated and functioning before assuming a platform bug.
  2. Distinguish user-side causes (Glean's permission model surfaces only what the querying user can access; malformed query syntax like bad SOQL) from true platform/data issues (crawl config missing an object; upstream warehouse maintenance):
    • Missing-data issues: check/add the object to crawl configuration, confirm next sync timing.
    • Permission-model confusion: explain agent/document visibility follows the end user's own access, not the agent creator's — no workaround.
    • Access-role requests: check user's access in the base platform, then in Glean's document visibility tool (https://app.glean.com/admin/documentaccess); try an alternate URL and escalate to another platform admin if it persists.
    • Security/permission-hardening requests: move from broad automatic access to least-privilege (project-specific), add specific admins for key-rotation needs.
    • Analytics/feedback-extraction requests with no existing tooling: set realistic timeline expectations, tie the ask to a larger project/OKR, offer a scoped data set to validate before full extraction.
  3. Close only after the requester actively confirms resolution, not just after the fix is applied.

Common Pitfalls

  • Assuming a connector error is a platform bug before checking auth/permissions or query syntax.
  • Not realizing Glean document visibility is user-scoped, causing repeat "why can't I see this" tickets.
  • Missing objects silently absent from crawl config with no proactive audit.

Typical Turnaround

Same day for most diagnosable issues; ~2 days when a scheduled sync or another admin must complete the fix.

Reference

Confluence page ID 6030032910. Part of the AIPM request-type playbook set; see [[AIPM Intake SOP]] for overall triage/classification.