Name who looks after it, decide what triggers a retest, and make sure unfinished work reaches a person. Then have someone other than the builder run it once.
Isn’t naming an owner enough?
Only if the next action actually reaches them. A handoff can name the right person and still leave the work sitting untouched.
1. Who does what.
Business owner. Accepts the service, sets priorities for changes and decides whether it still helps.
Technical support. Handles connections, permissions, updates and failures.
Exception owner and backup. Receive unresolved work, know what to check and keep it moving.
Practical check: One person can hold more than one role. Write down who holds each.
2. What to retest when something changes.
Triggers. A source document, the model, an instruction, a connection or an approval rule changes.
Ask which answers or actions could now behave differently.
Respond. Rerun the relevant test cases, and confirm who approves the change.
Practical check: An update isn’t finished just because it’s live.
3. Make handoffs land.
When the agent can’t finish, where does the work go? How does the owner find out, and how does anyone know it was picked up?
Practical check: Follow one unfinished case from start to finish and see where it waits.
4. The backup test.
- A known case
- Can the backup find the instructions and check the result?
- An exception
- Can they see what happened and where the work stopped?
- The fallback
- Can they pause the process and finish the work another way?
Practical check: Record what fails in the handover, not just in the agent.
Read more: Who Maintains an HR AI Tool When Its Builder Leaves? · Why an AI Agent Works in the Demo but Fails on Real Work
Use made-up examples and authorized access for the backup test. This is general guidance, not advice for your situation.