This part of the “shepherding” proposal would, as worded, entrench “AI taxation” on the project. I agree that keeping the original AI-assisted contributor engaged can help for task offloading; however, keeping AI in the loop (even when there is a chance that it will succeed) can impose a cost because it may well fail. This cost is exacerbated by the delayed feedback and “broken telephone” effects of distancing the subject matter experts from the AI.
In that sense, the “shepherding” proposal is in danger of being extractive by driving the community to help AI-assisted contributors explore the limits of AI.