AI System Handover Checklist
This checklist covers the four areas a real handover needs: a tested runbook, a trained operating team, a rehearsed recovery procedure, and a reviewed access list. It exists because 'documented' and 'handed over' aren't the same thing — this pushes toward the second.
Download the CSV, and for each item in the runbook, training, recovery, and access areas, record the owner, current status, and a link to the evidence (the actual document, the training record, the rollback test result). An item without evidence isn't done yet, even if it's marked as documented.
On this page
At a glance
- Four areas, not one: runbook, training, recovery, and access — a handover that's strong on documentation but untested on recovery is still an incomplete handover.
- Every item asks for evidence, not just a status. 'Rollback tested' means it was actually run once, not that a rollback document exists.
- The access item pairs directly with the AI agent permission matrix, so a handover doesn't quietly skip reviewing who can still do what.
Who this is for
This is for whoever is on either side of an AI system changing hands: the engineering team that built it, or the operations team that will run it day to day once it's live.
A handover is complete when the receiving team could recover from a failure without the original builders in the room. Everything on this checklist is aimed at that bar, not at producing documentation for its own sake.
What's inside
One row per handover item, grouped by area:
| Area | Item | Owner | Status |
|---|---|---|---|
| Recovery | EXAMPLE: Rollback procedure tested at least once before go-live | Engineering owner | Not started |
The four example rows cover one item per area. The download adds blank rows for the rest of your own runbook, training, recovery, and access items.
Download the handover checklist (CSV)What if the receiving team wasn't involved in building the system?
That's exactly the situation this checklist is for — every item is written from the receiving team's perspective, not the builder's.
Does 'recovery tested' mean a full disaster-recovery drill?
Not necessarily a full drill, but the rollback or recovery procedure should have actually been executed once, not just documented.
Should this checklist be revisited after go-live?
Yes — treat it as a living document the receiving team can re-run whenever ownership changes again.
Next step
A handover with every item evidenced is what lets a receiving team take real ownership, not just nominal responsibility.
Related: pair this with the AI agent permission matrix for the access review, or the acceptance test plan if recovery testing surfaces a gap.